Reliability Assessment of

Electric Power Systems

Using Monte Carlo Methods
Reliability Assessment of
Electric Power Systems
Using Monte Carlo Methods

Roy Billinton
University 0/ Saskatchewan
Saskatoon, Saskatchewan, Canada

Wenyuan Li
BC Hydro
Vancouver, British Columbia, Canada

Springer Science+Business Media, LLC

 L i b r a r y o f Congress C a t a l o g i n g - i n - P u b l i c a t i o n Data

B i l l i n t o n , Roy.
R e l i a b i l i t y assessment of e l e c t r i c power systems u s i n g Monte
C a r l o methods / Roy B I l H n t o n , Wenyuan L i .
p. cm.
I n c l u d e s b i b l i o g r a p h i c a l r e f e r e n c e s and I n d e x .
1 . E l e c t r i c power s y s t e m s — R e l i a b i l i t y — M a t h e m a t i c s . 2 . Monte
C a r l o method. I . L i , Wenyuan. I I . Title.
TK1010.B55 1994
6 2 1 . 3 1 9 ' 1 — d C20 94-39034
CIP

ISBN 978-1-4899-1348-7 ISBN 978-1-4899-1346-3 (eBook)

DOI 10.1007/978-1-4899-1346-3

© Springer Science+Business Media New York 1994

Originally published by Plenum Press, New York in 1994
Softcover reprint of the hardcover 1st edition 1994

A l l rights reserved

No part of this book may be reproduced, stored i n a retrieval system, or transmitted i n any form
or by any means, electronic, mechanical, photocopying, microfilming, recording, or otherwise,
without written permission from the Publisher
 To Joyce and Jun
with thanks for constant
encouragement and support
Preface

The application of quantitative reliability evaluation in electric power sys-

tems has now evolved to the point at which most utilities use these techniques
in one or more areas of their planning, design, and operation. Most of the
techniques in use are based on analytical models and resulting analytical
evaluation procedures. Improvements in and availability of high-speed digi-
tal computers have created the opportunity to analyze many of these prob-
lems using stochastic simulation methods and over the last decade there
has been increased interest in and use made of Monte Carlo simulation in
quantitative power system reliability assessment. Monte Carlo simulation is
not a new concept and recorded applications have existed for at least 50 yr.
However, localized high-speed computers with large-capacity storage have
made Monte Carlo simulation an available and sometimes preferable option
for many power system reliability applications.
Monte Carlo simulation is also an integral part of a modern undergrad-
uate or graduate course on reliability evaluation of general engineering
systems or specialized areas such as electric power systems. It is hoped that
this textbook will help formalize the many existing applications of Monte
Carlo simulation and assist in their integration in teaching programs. This
book presents the basic concepts associated with Monte Carlo simulation.
Its primary focus, however, is on the practical application of these techniques
in reliability evaluation of electric power generation, transmission, and distri-
bution systems. The book has arisen from the reliability teaching and
research program conducted at the University of Saskatchewan and applica-
tions at BC Hydro. The authors would like to thank the many graduate
students who have participated in this program and, in particular, Raymond
Ghajar, Ligong Gan, Lalit Goel, Easin Khan, and Guangbin Lian, whose
work is specifically referenced. Particular thanks are due to Raymond Ghajar
for coordinating and completing the work of many graduate students in
preparing the figures in the book. The cheerful assistance of Brenda Bergen
in a wide range of tasks is also gratefully acknowledged. Wenyuan Li would
vii
viii PREFACE

also like to thank Yakout Mansour and Fred Turner of BC Hydro, and
Jiaqi Zhou and Yihong Qin of Chongqing University, for their constant
support and encouragement.
Roy Billinton
Wenyuan Li
Contents

Chapter 1
Introduction
1.1. Probabilistic Analysis of Power Systems. . . . . . . . . . . . . . . . . . 1
1.2. Reliability Evaluation Techniques ...................... 3
1.3. Outline of the Book .................................. 4
1.4. References .......................................... 6

Chapter 2
Basic Concepts of Power System Reliability
Evaluation
2.1. Adequacy and Security . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.2. Functional Zones and Hierarchical Levels. . . . . . . . . . . . . . . . 10
2.3. Requirements for Power System Adequacy Assessment .... 12
2.3.1. Generation System (HLl) Studies ................ 12
2.3.2. Composite System (HL2) Studies ................. 14
2.3.3. Distribution System Studies ...................... 15
2.4. Reliability Cost/Worth. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.5. Reliability Data ...................................... 17
2.5.1. General Concepts .............................. 17
2.5.2. Equipment Reliability Information System. . . . . . . . . . 18
2.5.3. System Performance Data Collection Systems ...... 20
2.6. Adequacy Indices .................................... 22
2.6.1. Adequacy Indices in HLl Studies. . . . . . . . . . . . . . . . .. 23
2.6.2. Adequacy Indices in HL2 Studies. . . . . . . . . . . . . . . . .. 24
2.6.3. Adequacy Indices in Distribution System Evaluation 27
2.6.4. Reliability Worth Indices ........................ 29
2.7. Conclusions.......................................... 29
2.8. References 30
ix
x CONTENTS

Chapter 3
Elements Of Monte Carlo Methods
3.1. Introduction ........................................ 33
3.2. General Concepts .................................... 34
3.2.1. Two Simple Examples .......................... 34
3.2.2. Features of Monte Carlo Methods in Reliability
Evaluation .................................... 35
3.2.3. Efficiency of Monte Carlo Methods. . . . . . . . . . . . . . . . 37
3.2.4. Convergence Characteristics of Monte Carlo
Methods ...................................... 37
3.3. Random Number Generation .......................... 39
3.3.1. Introduction.................................... 39
3.3.2. Multiplicative Congruential Generator ............ 39
3.3.3. Mixed Congruential Generator. . . . ... . . .. ... .... .. 41
3.4. Random Variate Generation .......................... 42
3.4.1. Introduction.................................... 42
3.4.2. Inverse Transform Method ...................... 43
3.4.3. Tabulating Technique for Generating Random
Variates ...................................... " 45
3.4.4. Generating Exponentially Distributed Random
Variates ...................................... " 47
3.4.5. Generating Normally Distributed Random
Variates ...................................... 48
3.4.6. Generating Other Distribution Random Variates .. " 50
3.5. Variance Reduction Techniques ........................ 52
3.5.1. Introduction.................................... 52
3.5.2. Control Variates.. . ... . ... . . . .. . . . . ... ... . ... . .. 53
3.5.3. Importance Sampling . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
3.5.4. Stratified Sampling ........ ,. . . .. . .. . ... ... . ... .. 56
3.5.5. Antithetic Variates .............................. 58
3.5.6. Dagger Sampling. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
3.6. Three Simulation Approaches in Reliability Evaluation .... 60
3.6.1. State Sampling Approach ........................ 60
3.6.2. State Duration Sampling Approach. . . . . . . . . . . . . . . . 62
3.6.3. System State Transition Sampling Approach. . . . . . .. 64
3.7. Evaluating System Reliability by Monte Carlo
Simulation .......................................... 66
3.7.1. Example 1 .................................... 66
3.7.2. Example 2 .................................... 69
3.8. Conclusions.......................................... 71
3.9. References .......................................... 72
CONTENTS xi

Chapter 4
Generating System Adequacy Assessment
4.1. Introduction ........................................ 75
4.2. Single-Area Generating System Adequacy
Assessment-State Duration Sampling Method .......... 76
4.2.1. General Steps .................................. 76
4.2.2. Generating Unit Modeling ...................... 78
4.2.3. Stopping Rules ................................ 83
4.2.4. IEEE R TS Studies .............................. 84
4.3. Single-Area Generating System Adequacy
Assessment-State Sampling Method.................... 91
4.3.1. Single Load Level Case. . . . . . . . . . . . . . . . . . . . . . . . .. 91
4.3.2. Modeling an Annual Load Curve ................ 93
4.3.3. Cluster Technique for a Multistep Load Model. . . . .. 95
4.3.4. Stopping Rules ................................ 96
4.3.5. IEEE RTS Studies .............................. 96
4.4. Multi-Area Generating System Adequacy
Assessment-Maximum Flow Algorithm ................ 100
4.4.1. Basic Procedure ................................ 100
4.4.2. Modifying the Area Available Margin Model
Using the Maximum Flow Algorithm .............. 101
4.4.3. Case Studies. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 103
4.5. Multi-Area Generating System Adequacy
Assessment-Linear Programming Model ................ 106
4.5.1. Basic Procedure ................................ 106
4.5.2. Load Model of a Multi-Area System .............. 109
4.5.3. Linear Programming Model for Multi-Area Reliability
Evaluation .................................... 112
4.5.4. Case Studies. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 115
4.6. Different Supporting Policies in Multi-Area Generating
System Adequacy Assessment .......................... 120
4.6.1. Incorporation of Different Supporting Policies ...... 120
4.6.2. Case Studies. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 124
4.7. Conclusions .......................................... 128
4.8. References .......................................... 129

Chapter 5
Composite System Adequacy Assessment
5.1. Introduction ........................................ 131
5.2. System State Sampling Method and System Analysis
Techniques .......................................... 132
xii CONTENTS

5.2.1. Basic Methodology .............................. 133

5.2.2. Contingency Analysis for Composite Systems ...... 134
5.2.3. Linear Programming Optimization Model .......... 137
5.2.4. Basic Case Studies .............................. 139
5.3. Incorporation of the Annual Load Curve ................ 142
5.3.1. Multistep Model of the Annual Load Curve ........ 142
5.3.2. Ratios of Generation-Transmission Adequacy
Indices ........................................ 143
5.3.3. Effect of the Number of Steps in a Load Model .... 145
5.3.4. Effect of a Nonuniform Load Increment
Step Model .................................... 149
5.4. Generating Unit Derated States ........................ 150
5.4.1. Sampling Multiderated States .................... 150
5.4.2. Case Studies. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 151
5.5. Regional Weather Effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 153
5.5.1. General Concepts .............................. 153
5.5.2. Basic Transmission Line Model Recognizing
Weather Conditions ............................ 155
5.5.3. Sampling Regional Weather States ................ 156
5.5.4. Determination of Transmission Line Forced
Unavailability and Repair Time with Regional
Weather Effects ................................ 157
5.5.5. Test System and Data .......................... 158
5.5.6. Case Studies. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 160
5.6. Transmission Line Common Cause Outages .............. 164
5.6.1. General Concepts .............................. 164
5.6.2. Case Studies. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 166
5.7. Bus Load Uncertainty and Correlation .................. 169
5.7.1. Tabulating Technique for Bus Load Normal
Distribution Sampling .......................... 169
5.7.2. Bus Load Correlation Sampling Technique ........ 171
5.7.3. Case Studies .................................... 172
5.8. Frequency Index and System State Transition Sampling
Technique .......................................... 179
5.8.1. Discussion on the Frequency Index ................ 179
5.8.2. System State Transition Sampling Technique. . . . . . .. 182
5.8.3. Case Studies. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 185
5.9. Security Considerations in Composite System
Adequacy Evaluation .................................. 187
5.9.1. System Model Including Security Considerations .... 187
5.9.2. System Analysis Techniques ...................... 189
5.9.3. Hybrid Methodology ............................ 191
CONTENTS xiii

5.9.4. Case Studies .................................. 193

5.9.5. Application of Variance Reduction Techniques .... 196
5.10. Noncoherence in Composite System Adequacy
Assessment ........................................ 199
5.10.1. Definitions and Measures of Noncoherence ...... 199
5.10.2. A Simple Example ............................ 202
5.11. Conclusions ........................................ 205
5.12. References .......................................... 206

Chapter 6
Distribution System and Station Adequacy
Assessment
6.1. Introduction ........................................ 209
6.2. Basic Concepts and Analytical Techniques for Distribution
System Adequacy Assessment .......................... 210
6.2.1. General Concepts .............................. 210
6.2.2. Basic Distribution Systems ...................... 212
6.2.3. Analytical Techniques .......................... 213
6.3. Monte Carlo Simulation Approach to Distribution
System Adequacy Assessment .......................... 218
6.3.1. Probability Distribution Considerations ............ 218
6.3.2. Component State Duration Sampling Method ...... 219
6.3.3. Load Point Index Distributions .................. 221
6.3.4. System Performance Index Distributions .......... 230
6.4. Station Reliability Assessment . . . . . . . . . . . . . . . . . . . . . . . . .. 232
6.4.1. General Concepts .............................. 232
6.4.2. Station Component Modeling .................... 233
6.4.3. Component State Duration Sampling Procedure .... 236
6.4.4. Numerical Example 1 . . . . . . . . . . . . . . . . . . . . . . . . . . .. 240
6.4.5. Numerical Example 2 ............................ 244
6.5. Conclusions .......................................... 251
6.6. References .......................................... 253

Chapter 7
Reliability Cost/Worth Assessment
7.1. Introduction ........................................ 255
7.2. Customer Surveys and Customer Damage Functions ...... 256
7.2.1. Basic Customer Survey Methods .................. 256
7.2.2. Questionnaire Content and Data Treatment ........ 258
7.2.3. Customer Damage Functions .................... 258
xiv CONTENTS

7.3. Generating System Reliability Worth Assessment .......... 262

7.3.1. Assessment Techniques .......................... 262
7.3.2. Numerical Example ............................ 264
7.3.3. Application of Reliability Worth Assessment in
GenerationPlanning ............................ 265
7.4. Composite System Reliability Worth Assessment .......... 268
7.4.1. Assessment Method ............................ 268
7.4.2. Modified IEEE Reliability Test System and
Additional Data ................................ 272
7.4.3. Numerical Results .............................. 273
7.4.4. Application of Reliability Worth Assessment in
Composite System Planning ...................... 277
7.5. Distribution System Reliability Worth Assessment ........ 279
7.5.1. Assessment Techniques .......................... 279
7.5.2. Numerical Example ............................ 281
7.6. Minimum Cost Assessment in Composite System
Expansion Planning .................................. 282
7.6.1. Basic Concepts ................................ 282
7.6.2. Minimum Cost Assessment Method .............. 284
7.6.3. Case Studies .................................... 288
7.7. Conclusions.......................................... 294
7.8. References .......................................... 296

Appendix A
Reliability Test Systems
A.1. IEEE Reliability Test System (IEEE RTS) .............. 299
A.1.1. Load Model .................................. 299
A.1.2. Generating System ............................ 300
A.1.3. Transmission System .......................... 302
A.1.4. Additional Data .............................. 305
A.2. Roy Billinton Test System (RBTS) .................... 306
A.2.1. Brief Description of the RBTS .................. 309
A.2.2. Load Model .................................. 309
A.2.3. Generating System ............................ 309
A.2.4. Transmission System .......................... 312
A.2.5. Station Data .................................. 313
A.2.6. Reliability Worth Assessment Data .............. 315
A.3. References .......................................... 316
CONTENTS xv

Appendix B
Elements of Probability and Statistics
B.1. Probability Concept and Calculation Rules .............. 317
B.1.1. Probability Concept ............................ 317
B.1.2. Probability Calculation Rules .................... 317
B.2. Probability Distributions of Random Variables .......... 318
B.2.1. Probability Distribution Function and Density
Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 318
B.2.2. Important Distributions in Reliability
Evaluation .................................... 320
B.3. Numerical Characteristics of Random Variables .......... 321
B.3.1. Expectation and Variance ........................ 322
B.3.2. Covariance and Correlation Function ............ 322
B.4. Limit Theorems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 323
B.4.1. Law of Large Numbers. . . . . . . . . . . . . . . . . . . . . . . . .. 323
B.4.2. Central Limit Theorem. . . . . . . . . . . . . . . . . . . . . . . . .. 324
B.5. Parameter Estimation ................................ 324
B.5.1. Basic Definitions .............................. 324
B.5.2. Sample Mean and Sample Variance .............. 325
B.6. References .......................................... 326

Appendix C
Power System Analysis Techniques
c.l. AC Load Flow Models ...... . . . . . . . . . . . . . . . . . . . . . . .. 327
C.l.l. Load Flow Equations .......................... 327
C.1.2. Newton-Raphson Model ........................ 328
C.1.3. Fast Decoupled Model .......................... 328
C.2. DC Load Flow Models .............................. 329
C.2.1. Basic Equations. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 329
C.2.2. Relationship between Power Injections and Line
Flows ........................................ 331
C.3. Optimal Power Flow .................................. 331
C.4. Contingency Analysis ................................ 332
C.5. References .......................................... 336

Appendix 0
Optimization Techniques
D.1. Linear Programming ................................ 337
D.1.1. Basic Concepts ................................ 337
D.1.2. Generalized Simplex Method .................... 338
xvi CONTENTS

D.1.3. Duality Principle .............................. 340

D.1.4. Dual Simplex Method ................. '" ... '" 341
D.1.5. Linear Programming Relaxation Technique. . . . . . .. 342
D.2. Maximum Flow Method .............................. 344
D.2.1. Basic Concepts ................................ 345
D.2.2. Maximum Flow Problem ...................... 346
D.3. References .......................................... 347

Index ...................................................... 349

 1
Introduction

1.1. PROBABILISTIC ANALYSIS OF POWER

SYSTEMS
The primary function of an electric power system is to provide electrical
energy to its customers as economically as possible and with an acceptable
degree of continuity and quality. Modem society has come to expect that
the supply of electrical energy will be continuously available on demand.
This is not possible due to random failures of equipment and the system,
which are generally outside the control of power system personnel. Electricity
supply generally involves a very complex and highly integrated system. Fail-
ures in any part of it can cause interruptions which range from inconvenienc-
ing a small number of local residents, to major and widespread catastrophic
disruptions of supply. The economic impact of these outages is not restricted
to loss of revenue by the utility or loss of energy utilization by the customer
but include indirect costs imposed on society and the environment due to
the outage. In the case of the 1977 New York blackout, the total costs of
the blackouts were suggested to be as high as $350 million, which 84% was
attributed to indirect costs. (1)
In order to reduce the probability, frequency and duration of these
events and to ameliorate their effect, it is necessary to increase investment
in the planning, design and operating phases. A whole series of questions
arise from this concept, including:
• How much should be spent?
• Is it worth spending any money?
• Should the reliability be increased, maintained at existing levels, or
allowed to degrade?
• Who should decide-the utility, a regulator, the customer?
• On what basis should the decision be made?
It is evident that the economic and reliability constraints can conflict and
lead to difficult managerial decisions.
1
2 CHAPTER 1

These problems have been long recognized by power system managers,

designers, planners, and operators. Design, planning, and operating criteria
and techniques have been developed over many years in an attempt to resolve
and satisfy the dilemma created by the economic, reliability, and operational
constraints. The criteria and techniques first used in practical applications
were all deterministically based and many of these criteria and techniques
are still in use today. These include percentage reserves in generation capacity
planning or N - I contingency criteria in transmission planning.
The basic weakness of deterministic criteria is that they do not respond
to, nor do they reflect, the probabilistic or stochastic nature of system
behavior, of customer demands, or of component failures. Deterministic
analysis can consider the outcome and ranking of hazards which may lead
to a dangerous state or a system failure. However, a hazard, even if extremely
undesirable, is of little consequence if it cannot occur or is so unlikely that
it can be ignored. Planning alternatives based on such hazard analyses will
lead to overinvestment. Conversely, if the hazards selected in the determin-
istic analysis are not very severe but have relatively large probabilities of
occurrence, alternatives based on a deterministic analysis of such hazards
will lead to insufficient system reliability. Probabilistic evaluation of a power
system can recognize not only the severity of a state or an event and its
impact on system behavior and operation, but also the likelihood or prob-
ability of its occurrence. Appropriate combination of both severity and likeli-
hood creates indices that truly represent system risk.
The need for probabilistic evaluation of system behavior has been recog-
nized for over forty years(2.3) and it may be questioned why such methods
have not been widely used in the past. The main reasons have been lack of
data, limitations of computational resources, lack of realistic techniques,
aversion to the use of probabilistic techniques, and a misunderstanding of
the significance and meaning of probabilistic criteria and indices. None of
these reasons is valid today as most utilities have relevant reliability data-
bases, computing facilities are greatly enhanced, evaluation techniques are
highly developed, and most engineers have a working understanding of pro-
babilistic techniques. There is, therefore, now no need to artificially constrain
the inherent probabilistic or stochastic nature of a power system into a
deterministic framework.
A wide range of probabilistic techniques has been developed. These
include techniques for reliability and reliability worth evaluation, (4-14) proba-
bilistic load flOW,(IS-17) probabilistic fault analysis,(18-20) probabilistic transi-
ent stability,(21-23) and probabilistic transmission line design.(24-27) The
fundamental and common concept behind each of these developments is
the need to recognize that power systems and system components behave
stochastically and all input and output state and event parameters are proba-
bilistic variables.
INTRODUCTION 3

1.2. RELIABILITY EVALUATION

TECHNIQUES
Power system reliability evaluation has been extensively developed util-
izing probabilistic methods and a wide range of appropriate indices can be
determined. A single all-purpose formula or technique does not exist. The
approach used and the resulting formulas depend on the problem and the
assumptions utilized. Many assumptions must be made in all practical appli-
cations of probability and statistical theory. The validity of the analysis is
directly related to the validity of the models used to represent the system.
Actual failure distributions seldom completely fit the analytical descriptions
used in the analysis, and care must be taken to ensure that significant errors
are not introduced through oversimplification. On the other hand, a poten-
tial user of reliability evaluation techniques should clearly keep in mind
the objective behind the evaluation process. In generation system capacity
assessment, a reference adequacy level can often be obtained in terms of
historical experience of the individual utility. This generally requires a model
as accurate as possible so that the resulting indices can be compared with
the given target level. In evaluating composite generation and transmission
systems, distribution systems, and substation configurations, however, it
is generally necessary to assess the relative benefits between the various
alternatives available, including the option of not doing any reinforcement
at all. The level of analysis need not be any more complex than that which
enables the relative merits to be assessed. The apparent ability to include a
high degree of precision in the calculations should never override the inherent
uncertainties in the forecast data, including load, failure rates, restoration
times, etc. Absolute reliability, although an ideal objective, is virtually impos-
sible to evaluate. This does not weaken the necessity to objectively assess
the relative merits of alternate schemes.
The most important point to note is that it is absolutely necessary to
have a complete understanding of the system. No amount of probability
theory can circumvent this important engineering function. Probability
theory is simply a tool that enables the analyst to transform knowledge of
the system into a prediction of its likely future behavior. Only after this
understanding has been achieved, can a model be derived and an approp-
riate evaluation technique chosen. Both of these must reflect and respond
to the way the system operates and fails. The basic steps involved are to:
• Understand the way in which the components and system operate.
• Identify the way in which they can fail.
• Deduce the consequences of the failures.
• Derive models to represent these characteristics.
• Then select the evaluation technique.
4 CHAPTER 1

There are two main categories of evaluation techniques: analytical and

simulation. Analytical techniques represent the system by analytical models
and evaluate the indices from these models using mathematical solutions.
Monte Carlo simulation methods, however, estimate the indices by simulat-
ing the actual process and random behavior of the system. The method,
therefore, treats the problem as a series of experiments. There are merits
and demerits in both methods. In general, if complex operating conditions
are not considered and/or the failure probabilities of components are quite
small (i.e., the system is very reliable), the analytical techniques are more
efficient. When complex operating conditions are involved and/or the num-
ber of severe events is relatively large, Monte Carlo methods are often prefer-
able. The main advantages of Monte Carlo methods are as follows:
• In theory, they can include system effects or system processes which
may have to be approximated in analytical methods.
• The required number of samples for a given accuracy level is indepen-
dent of the size of the system and therefore it is suitable for large-
scale system evaluation.
• They can simulate probability distributions associated with compo-
nent failure and restoration activities. This generally cannot be per-
formed using analytical methods.
• They can calculate not only reliability indices in the form of expected
values of random variables, but also the distributions of these indices,
which analytical techniques generally cannot.
• Nonelectrical system factors such as reservoir operating conditions,
weather effects, etc. can also be simulated.
It should be appreciated that there can be considerable differences
between various analytical techniques or Monte Carlo methods. The evalua-
ted indices are only as good as the model derived for the system, the appro-
priateness of the evaluation technique, and the quality of the input data
used in the models and techniques. This is an important point to appreciate.

1.3. OUTLINE OF THE BOOK

Analytical techniques for power system reliability evaluation have been
highly developed and there are many technical papers and several books
which summarize the developed techniques.(4-7) Developments in Monte
Carlo simulation can generally be said to have occurred later than compar-
able analytical techniques. Considerable effort has, however, been expanded
in developing Monte Carlo simulation methods, and in a number of areas
INTRODUCTION 5

these techniques provide viable alternatives to the analytical approaches.

This book provides a basic reference on this area and is based on research
activities performed at the University of Saskatchewan and BC Hydro in
Canada. The book does not include every known and available technique but
provides a systematic description of the topic, including basic mathematical
principles, and presents applications in each power system functional zone.
Chapter 2 presents basic concepts of power system reliability evaluation.
These concepts are used in both analytical techniques and Monte Carlo
methods. Chapter 3 describes elements of the Monte Carlo method. This
description, based on the mathematical fundamentals of Monte Carlo simu-
lation, places emphasis on the application aspects of Monte Carlo methods in
system reliability evaluation. Chapters 2 and 3 are the basis for the following
chapters.
Chapter 4 illustrates applications of the component state duration sam-
pling and the system state sampling methods in generation system adequacy
assessment. Both single area and multi-area generation systems are dis-
cussed. A maximum network flow approach and linear programming models
are utilized in the multi-area system evaluation.
Chapter 5 illustrates applications of the system state sampling and the
system state transition sampling methods in composite generation and trans-
mission system adequacy assessment. Contingency analysis and linear pro-
gramming optimal power flow techniques are utilized in this aspect. Various
system problems are considered. These include considerations such as annual
load duration curves, generating unit derated states, regional weather effects,
transmission line common cause failures, bus load forecast uncertainty and
correlation, etc. Two special aspects are discussed. The first one is considera-
tion of steady state security in composite system adequacy assessment and
the second one is possible noncoherence of composite systems.
Chapter 6 describes applications of the component state duration sam-
pling method in distribution system and station adequacy assessment. A
basic emphasis is placed on the simulation of distributions of component
restoration times and the calculation of probability distributions of reliability
indices.
Chapter 7 illustrates reliability cost/worth assessment of generation,
composite, and distribution systems using Monte Carlo methods. Applica-
tions of reliability worth assessment in power system planning are also dis-
cussed and a minimum cost assessment method for composite system
expansion planning is described.
There are four appendices provided. Appendix A presents the data for
the two reliability test systems which are used in the case studies throughout
the book. Appendix B provides essential elements of probability and statis-
tics, while Appendices C and D, respectively, present basic power system
6 CHAPTER 1

analysis techniques and the optimization methods used in the various

applications.
It is important to note that although the book describes applications
of Monte Carlo methods in power system reliability assessment, the basic
system analysis techniques and mathematical models such as contingency
analysis, optimal power flow, maximum network flow approach, linear pro-
gramming models for various evaluation purposes, etc. can also be utilized in
analytical applications of power system reliability assessment. The essential
difference between analytical and Monte Carlo methods is basically how to
select system state scenarios to be evaluated.
This book does not pretend to include all the known and available
materials on this topic. It will, however, place the reader in a position where-
by he/she can accommodate a wide range of problems and gain a wider and
deeper appreciation of the materials that have been published and which
will, no doubt, be enhanced in the future.

1.4. REFERENCES
1. R. Sugarman, "New York City's Blackout: A $3S0 Million Drain," IEEE Spectrum Com-
pendium: Power Failure Analysis and Prevention, 1979, pp. 48-S0.
2. G. Calabrese, "Generating Reserve Capacity Determined by the Probability Method,"
AlEE Trans., No. 66, 1947, pp. 1439-1450.
3. C. W. Watchorn, "The Determination and Allocation of the Capacity Benefits Resulting
from Interconnecting Two or More Generating Systems," AlEE Trans., No. 69, 19S0,
Pt. II, pp. 1180-1186.
4. R. Billinton and R. N. Allan, Reliability Evaluation of Power Systems, Plenum, New York,
1984.
S. R. Billinton and R. N. Allan, Reliability Assessment of Large Electric Power Systems,
Kluwer, Boston, 1988.
6. R. Billinton, R. N. Allan, and L. Salvaderi, Applied Reliability Assessment in Electric Power
Systems, IEEE Press, New York, 1991.
7. J. Endrenyi, Reliability Modeling in Electric Power Systems, Wiley, Chichester, 1978.
8. R. Billinton, "Bibliography on the Application of Probability Methods in Power System
Reliability Evaluation," IEEE Trans. on PAS, Vol. 91, 1972, pp. 649-660.
9. IEEE Committee Report, "Bibliography on the Application of Probability Methods in
Power System Reliability Evaluation, 1971-1977," IEEE Trans. on PAS, Vol. 97, 1978,
pp. 223S-2242.
10. R. N. Allan, R. Billinton, and S. H. Lee, "Bibliography on the Application of Probability
Methods in Power System Reliability Evaluation, 1977-1982," IEEE Trans. on PAS,
Vol. 103, 1984, pp.27S-282.
11. R. N. Allan, R. Billinton, S. M. Shahidehpour, and C. Singh, "Bibliography on the Applica-
tion of Probability Methods in Power System Reliability Evaluation, 1982-1987," IEEE
Trans. on PS, Vol. 3, No.4, 1988, pp. ISSS-lS64.
12. R. Billinton, G. Wacker, and E. Wojczynski, "Comprehensive Bibliography of Electrical
Service Interruption Costs," IEEE Trans. on PAS, Vol. 102, 1983, pp. 1831-1837.
INTRODUCTION 7

13. G. Tollefson, R. Billinton, and G. Wacker, "Comprehensive Bibliography on Reliability

Worth and Electrical Service Consumer Interruption Costs: 1986-1990," IEEE Trans. on
PS, Vol. 6, No.4, 1991, pp. 1508-1514.
14. CIGRE Working Group 38.03, "Power System Reliability Analysis-Application Guide,"
CIGRE Publications, Paris, 1988.
15. M. Th. Schilling, A. M. Leite da Siva, R. Billinton, and M. A. EI-Kady, "Bibliography
on Power System Probabilistic Analysis (1962-1988)," IEEE Trans. on PS, Vol. 5, No.1,
1990, pp. 1-11.
16. B. Borkowska, "Probabilistic Load Flow," IEEE Trans. on PAS, Vol. 93, 1974, pp. 752-
759.
17. A. M. Leite da Silva, S. M. P. Ribeiro, V. L. Arienti, R. N. Allan, and M. B. Do Coutto
Filho, "Probabilistic Load Flow Techniques Applied to Power System Expansion Plan-
ning," IEEE Trans. on PS, Vol. 5, No.4, 1990, pp. 1047-1053.
18. G. L. Ford and K. D. Srivastava, "Probabilistic Short Circuit Design of Substation Bus
Systems," IEEE Winter Meeting, 1978, Paper A78 211-5.
19. G. L. Ford and S. S. Sengupta, "Analytical Methods for Probabilistic Short Circuit
Studies," Electric Power System Research, Vol. 5, 1982, pp. 13-20.
20. M. D. Germani, G. L. Ford, E. G. Neudorf, M. Vainberg, M. A. EI-Kady, and R. W. D.
Ganton, "Probabilistic Short Circuit Uprating of Station Strain Bus System-Overview
and Application," IEEE Trans. on PWRD, Vol. 1, 1986, pp. 111-117.
21. R. Billinton and P. R. S. Kurunganty, "A Probabilistic Index for Transient Stability,"
IEEE Trans. on PAS, Vol. 99, 1980, pp. 195-206.
22. R. Billinton and P. R. S. Kurunganty, "Probabilistic Evaluation of Transient Stability in
Multimachine Power Systems," lEE Proc. C, Vol. 126, 1980, pp. 321-326.
23. P. M. Anderson, A. Bose, K. J. Timko, and F. E. Villaseca, "A Probabilistic Approach
to Stability Analysis," EPRI Report EL-2797, 1983.
24. EPRI Report, "Reliability Based Design of Transmission Line Structures: Final Report,"
EL-4793, 1987.
25. E. Ghannoum and G. Orawski, "Reliability Based Design of Transmission Lines According
to Recent Advances by IEC and CIGRE," in Probabilistic Methods Applied to Electric
Power Systems, Pergamon, Oxford, 1987, pp. 125-142.
26. International Electrotechnical Commission (lEC), "Draft oflEC/TCll Recommendations
for Overhead Lines," 1985.
27. S. G. Krishnasamy and N. Ramani, "Wind Direction and Extreme Wind Loads for Over-
head Power Line Design," in Probabilistic Methods Applied to Electric Power Systems,
Pergamon, Oxford, 1987, pp. 399-411.
 2
Basic Concepts of
Power System
Rei iabi I ity
Evaluation

2.1. ADEQUACY AND SECURITY

The term reliability has a very wide range of meaning and cannot be associ-
ated with a single specific definition such as that often used in the mission-
oriented sense. (\) It is therefore necessary to recognize this fact and to use
the term to indicate, in a general rather than a specific sense, the overall
ability of the system to perform its function. Power system reliability assess-
ment can therefore be divided into the two basic aspects of system adequacy
and system security as shown in Figure 2.1. The terms adequacy and security
can be described as follows:
Adequacy relates to the existence of sufficient facilities within the system
to satisfy the consumer load demand or system operational constraints.
These includes the facilities necessary to generate sufficient energy and the
associated transmission and distribution facilities required to transport the
energy to the actual consumer load points. Adequacy is therefore associated
with static conditions which do not include system dynamic and transient
disturbances.
Security relates to the ability of the system to respond to dynamic or
transient disturbances arising within the system. Security is therefore associ-
ated with the response of the system to whatever perturbations it is subject
to. These include the conditions associated with both local and widespread
disturbances and the abrupt loss of major generation or/and transmission
9
10 CHAPTER 2

system reliability

Figure 2.1. Subdivision of system reliability.

facilities which can lead to dynamic, transient, or voltage instability of the

system.
It is important to appreciate that most of the probabilistic techniques
presently available for reliability evaluation are in the domain of adequacy
assessment. The ability to assess security is therefore very limited. Probabilis-
tic transient stability evaluation(2) lies in this domain together with tech-
niques for quantifying unit commitment and response risk. (3) This limitation
is due to the complexities associated with modeling the system in the security
domain. Most of the indices used at the present time are adequacy indices
and not overall reliability indices. Indices which are obtained by assessing
past system performance encompass the effect of all system faults and failures
irrespective of cause, and therefore include the effect of insecurity as well as
those due to inadequacy. This fundamental difference is an important point
which should be clearly recognized.

2.2. FUNCTIONAL ZONES AND

HIERARCHICAL LEVELS
The basic techniques for adequacy assessment can be categorized in
terms of their application to segments of a complete power system. These
segments are shown in Figure 2.2 and can be defined as the functional
zones of generation, transmission, and distribution.(4) This division is an
appropriate one as most utilities are either divided into these zones for the
purposes of organization, planning, operation, and/or analysis or are solely
responsible for one of these functions. Adequacy studies can be, and are,
conducted in each of these three functional zones.
The functional zones shown in Figure 2.2 can be combined to give the
hierarchical levels shown in Figure 2.3. These hierarchical levels can also be
BASIC CONCEPTS OF POWER SYSTEM RELIABILITY EVALUATION 11

generation
facilities

transmission
facilities

distribution
facilities

Figure 2.2. Basic functional zones.

used in adequacy assessment. Hierarchical Levell (HLI) is concerned with

only the generation facilities. Hierarchical Level 2 (HL2) includes both
generation and transmission facilities while HL3 includes all three functional
zones in an assessment of consumer load point adequacy. HL3 studies are
not usually conducted directly due to the enormity of the problem in a
practical system. Analysis is usually performed in the distribution functional
zone in which the input points mayor may not be considered to be fully
reliable.
Functional zone studies are often done which do not include the hier·
archicallevels above them. These studies are usually performed on a subset
---------------------i
r-------------------, I
.-----------------, I I
I II I
I .1.
I generation I I I hierarchical level I
:

I facilities L o.
iI:

I II
..:. - - HLI
II..
-------- -------- '
• "I I
II
II
II
II
I I
I I

!oI:';t--- hierarchical
I I
transmission level II
facilities I I HLII
II

--------
I I
L. -------- .J I
I
I
I
I
I

distribution :.....- hierarchical level III

facilities : HUll
I
I
L _____________________ ~I

Figure 2.3. Hierarchical levels.

12 CHAPTER 2

of the system in order to examine a particular configuration or topological

change. These analyses are frequently undertaken in the subtransmission
and distribution system functional zones as these areas are less affected by
the actual location of the generating facilities. In transmission planning
studies, especially for those relatively low-voltage transmission subsystems
which are far away from generation sources, generation facilities are usually
assumed to be 100% reliable in order to highlight effects of different transmis-
sion components and configurations on system reliability.
In composite system (HL2) adequacy studies of relatively large scale
transmission systems, it is reasonable to limit a study area and, in doing so,
provide more realistic results than by evaluating the whole system. This is
due to the fact that the addition of a transmission facility may considerably
affect a local area but have basically little impact on remote parts of the
system. The contribution to overall reliability of a large system due to a
local transmission line addition may be so small that it is masked by compu-
tational errors and consequently cannot be reflected in the reliability change
of the whole system. This contribution, however, can be a relatively large
proportion of the reliability change in the local area. When the planning
task is to investigate reinforcement alternatives of the subsystem represented
by the local area, evaluating reliability of the whole system may lead to a
"no addition" conclusion. In this case, it is desirable to limit the size of
the studied system or/and utilize system equivalent techniques. Reliability
equivalence is a useful tool/concept(5) in this application area.

2.3. REQUIREMENTS FOR POWER SYSTEM

ADEQUACY ASSESSMENT

2.3.1. Generation System (HL1) Studies

In a generation system study, the total system generation is examined
to determine its adequacy to meet the total system load requirement. This
activity is usually termed "generating capacity reliability evaluation." The

8
system model at this level is shown in Figure 2.4.

total total
system G f--+-----~~ .. system
generation load

Figure 2.4. Hierarchicallevel1 (HLl) model.

BASIC CONCEPTS OF POWER SYSTEM RELIABILITY EVALUATION 13

generation load
model model

risk
model

Figure 2.5. Conceptual tasks for HLl evaluation.

The transmission system and its ability to move the generated energy
to the consumer load points is ignored in generating system adequacy assess-
ment. The basic concern is to estimate the generating capacity required to
satisfy the system demand and to have sufficient capacity to perform correc-
tive and preventive maintenance on the generating facilities. The basic tech-
nique, used in the past, to determine the capacity requirement was the
percentage reserve method. In this approach, the required reserve is a fixed
percentage of either the installed capacity or the predicted load. This and
other criteria, such as a reserve equal to one or more of the largest units,
have now been largely replaced by probabilistic methods which respond to
and reflect the actual factors that influence the reliability of the system. (3)
The basic modeling approach(3) for an HLl study is shown in Figure
2.5. Analytical methods and Monte Carlo simulation utilize different tech-
niques to assess generation and load models. The essential concept shown
in this figure, however, is basically the same for both techniques.
Limited transmission considerations can be, and are, included in HLl
studies. These include the modeling of remote generation facilities and inter-
connected systems. In the latter case, only the interconnections between
adjacent systems are modeled, not the internal system connections or intra-
connections. The latter is often termed multi-area generating system
adequacy assessment. In the case of remote generation, the capacity model
of the remote source is modified by the reliability of the transmission link
before being added to the system capacity model. In the case of intercon-
nected systems, the available assistance model should be considered. The
most widely used modeling methods for a multi-area generation system are
network flow (maximum flow-minimum cut) techniques.
14 CHAPTER 2

2
1
1
1
1
1
1
2 17
1 1
1
1
3 4

Figure 2.6. A sample composite system for HL2 studies.

2.3.2. Composite System (HL2) Studies

In HL2 studies, the simple generation-load model shown in Figure 2.4
is extended to include bulk transmission. Adequacy analysis at this level is
usually termed composite system or bulk transmission system evaluation. A
sample composite system(3) is shown in Figure 2.6.
HL2 studies can be used to assess the adequacy of an existing or pro-
posed system including the impact of various reinforcement alternatives at
both the generation and transmission levels. These effects can be assessed
by evaluating two sets of indices: individual bus (load points) indices and
overall system indices. These indices are complementary and not alternatives.
The system indices provide an assessment of overall adequacy. The bus
indices show the effect at individual busbars and provide input values to
subsequent distribution system adequacy evaluation (HL3). Both system and
bus indices can be categorized as annualized and annual indices. Annualized
indices are calculated using a single load level (normally the system peak
load level) and expressed on a one-year basis. Annual indices are calculated
considering detailed load variations throughout one year. In general, annual-
ized indices provide satisfactory indications when comparing adequacies of
different reinforcement alternatives. Annual indices show a more realistic
and comprehensive picture of system adequacy. Annual indices should be
utilized when attempting to calculate actual damage cost of system unreli-
ability or evaluate reliability worth. Although these indices add realism by
including transmission, they are still adequacy indicators. They do not
BASIC CONCEPTS OF POWER SYSTEM RELIABILITY EVALUATION 15

include the system dynamics or the ability of the system to respond to

transient disturbances. They simply measure the ability of the system to
adequately meet its requirements in a specified set of probabilistic states.
There are many complications in HL2 studies associated with load flow
calculations, contingency analysis, overload alleviation, generation reschedu-
ling, load curtailment philosophy, etc. In the area of system state selection,
considerations of independent, dependent, common-cause, and station-
related outages, regional weather effects, and bus load uncertainty and corre-
lation, etc., introduce additional complexities. These aspects are discussed
further in Chapter 5. It should be noted, however, that the primary indices
are expected values and are highly dependent on the modeling assumptions.
Adequacy assessment at HL2 is still very much in the development phase.
There are many power utilities and related organizations doing interesting
and innovative work in this area.

2.3.3. Distribution System Studies

The overall problem of HL3 evaluation can become very complex in
most systems as this level involves all three functional zones, starting at the
generating stations and terminating at the individual consumer load points.
For this reason, the distribution functional zone is usually analyzed as a
separate entity. HL3 indices can be evaluated, however, by using the HL2
load-point indices as the input values of the distribution functional zone
being analyzed.
Distribution system adequacy evaluation involves assessment of suitable
adequacy indices at the actual consumer load points. There are two basic
types of distribution system: meshed and radial configurations. Evaluation
methods for meshed distribution systems are conceptually the same as those
used for composite systems. The techniques for a radial distribution system
are based on failure-mode analysis including considerations of all realistic
failure and restoration processes. These techniques also apply to substation
adequacy assessment as a substation has similar configurations and failure
modes.
Individual customer adequacy due to the overall system is reflected by
HL3 indices. In most systems, inadequacy of the individual customer points
is caused mainly by the distribution system. HL2 adequacy indices generally
have a relatively small effect on individual customer load-point indices.
The HLl and HL2 indices are very important, however, because failures in
these segments affect large sections of the system and therefore can have
widespread and perhaps catastrophic consequences for both society and the
16 CHAPTER 2

environment. Failures in the distribution system, although more frequent,

have much more localized effects.

2.4. RELIABILITY COST/WORTH

Adequacy studies of a system are only part of the required overall
assessment. The economics of alternate facilities play a major role in the
decision-making process. The simplest approach which can be used to relate
economics with reliability is to consider the investment cost only. In this
approach, the increase in reliability due to the various alternatives is evalua-
ted together with the investment cost associated with each scheme. Dividing
this cost by the increase in reliability gives the incremental cost of reliability,
i.e., how much it will cost for a per-unit increase in reliability. This approach
is useful when comparing alternatives given that the reliability of a section
of the power system is inadequate. In this case, the lowest incremental cost
of reliability is the most cost effective. This is still a significant step forward
from comparing alternatives and making major capital investment decisions
using deterministic techniques.
The weakness of this approach is that it is not related to either the
likely return on investment or the real benefit accruing to the consumer,
utility, and society. In order to make a consistent appraisal of economics
and reliability, albeit only the adequacy, it is necessary to compare the
adequacy cost (the investment cost needed to achieve a certain level of
adequacy) with adequacy worth (the benefit derived by the utility, consumer,
and society). A step in this direction can be achieved by setting a level of
incremental cost which is believed to be acceptable to consumers. Schemes
costing less than this level would be considered viable but schemes costing
more would be rejected. A complete solution, however, requires a detailed
knowledge of adequacy worth.
This type of economic appraisal is a fundamental and important area
of engineering application, and it is possible to perform this kind of evalua-
tion at the three hierarchical levels. A goal for the future should be to extend
this adequacy comparison within the same hierarchical structure to includl
security, and therefore to arrive at reliability-cost and reliability-worth evalu·
ation. The extension of quantitative reliability analysis to the evaluation 01
service worth is deceptively simple but is fraught with potential misapplica-
tion. The basic concept is relatively simple and can be illustrated using the
cost/reliability curves of Figure 2.7.
Figure 2.7 shows that utility costs will generally increase as consumers
are provided with higher reliability. On the other hand, consumer costs
associated with supply interruptions will decrease as the reliability increases.
BASIC CONCEPTS OF POWER SYSTEM RELIABILITY EVALUATION 17

system reliability
Figure 2.7. Consumer, utility, and total cost as a function of system reliability.

The total costs to society are the sum of these two individual costs. This
total cost exhibits a minimum point at which an "optimum" or target level
of reliability is achieved.
There have been many studies on the subject of interruption and outage
costs. (6) These analyses show that, although trends are similar in virtually
all cases, the costs vary over a wide range and depend on the country of
origin and the type of consumer. It is apparent, therefore, that considerable
research is still needed on power system interruption costs. This research
should consider both the direct and indirect costs associated with the loss
of supply.

2.5. RELIABILITY DATA

2.5.1. General Concepts

Any discussion of quantitative reliability evaluation invariably leads to
a discussion of the data available and the data required to support such
studies. Valid and useful data are expensive to collect, but it should be
recognized that in the long run it will be even more expensive not to collect
them. It is sometimes argued as to which comes first: reliability data or
reliability methodology. Some utilities do not collect data because they have
18 CHAPTER 2

not fully determined a suitable reliability methodology. Conversely, they do

not conduct reliability studies because they do not have any data. It should
be remembered that data collection and reliability evaluation must evolve
together and that the process is therefore iterative.
In conceptual terms, data can be collected for two fundamental reasons:
assessment of past performance and/or prediction of future system perform-
ance. Past assessment looks back at the past behavior of the system while
predictive procedures look forward at future system behavior. In order to
predict, however, it is essential to transform past experience into suitable
models for future prediction. Collection of suitable data is therefore essential
as it forms the input to relevant reliability models, techniques, and equations.
It should also be appreciated that the data requirements should reflect
the needs of the predictive methodology. The data must be sufficiently com-
prehensive to ensure that the methods can be applied, but restrictive enough
to ensure that unnecessary data are not collected nor irrelevant statistics
evaluated. The data should therefore reflect and respond to the factors that
affect system reliability and enable it to be modeled and analyzed. It should
relate to the two main processes involved in component behavior, namely,
the failure process and the restoration process. It cannot be stressed too
strongly that, in deciding data to be collected, a utility must consider the
factors that have an impact on its own planning and design considerations.
The quality of the data and the indices evaluated depend on two impor-
tant factors: confidence and relevance. The quality of the data, and thus the
confidence that can be placed in it, is clearly dependent on the accuracy and
completeness of the information compiled by operating and maintenance
personnel. It is therefore essential that they should be made fully aware of
the future use of data and the importance they will play in the later develop-
ment of the system. The quality of the statistical indices is also dependent
on how the data are processed, how much pooling is done, and the age of
the data currently stored.

2.5.2. Equipment Reliability Information

System
In order to establish a common reporting procedure and realize a practi-
cal database, all the major electric power utilities in Canada participate in
a single data collection and analysis system called the Equipment Reliability
Information System (ERIS) of the Canadian Electrical Association
(CEA).(7) CEA started collecting data on generation and transmission out-
ages in 1977 and since then has published a number of annual or periodic
BASIC CONCEPTS OF POWER SYSTEM RELIABILITY EVALUATION 19

ERIS

Generation Equipment Transmission Equipment Distribution Equipment

Status Reporting System Outage Reporting System Outage Reporting System

Figure 2.8. Basic components of the ERIS.

reports. A distribution equipment outage reporting system was initiated on

January 1, 1992 and participating utilities commenced collecting data at that
time. Annual reports will be issued commencing in 1993. The basic structure
of ERIS is shown in Figure 2.8.
In order to provide a general indication of the basic equipment reli-
ability parameters, selected CEA data are presented. Tables 2.1 to 2.4 show
average failure rates and mean outage times of major generation and trans-
mission components obtained from the 1990 annual ERIS reports.(8,9) These

Table 2.1. Generating Unit Failure Data

Unit capacity Failure rate Mean outage time
(MW) (occ./yr) (hr/occ.)
Hydraulic units
24-99 3.44 93.29
100-199 3.76 71.35
200-299 6.14 74.94
300-399 6.03 41.46
400-499 2.82 64.50
500 & over 2.42 112.58
Fossil units
69-99 11.50 44.72
100-199 14.53 37.72
200-299 13.79 25.72
300-399 16.54 46.55
400-599 8.79 45.30
Nuclear units
400-599 3.40 369.35
600-799 4.90 27.17
800 & over 4.49 111.64
20 CHAPTER 2

Table 2.2. Transmission Line Failure Data

Line-related Terminal-related
Voltage
level failure rate outage time failure rate outage time
(kV) (occ./yr/l00 km) (hr/occ.) (occ./yr) (hr/occ.)
110-149 1.0858 6.6 0.1295 5.5
150-199 0.5513 1.6 0.0959 4.1
200-299 0.5429 12.4 0.1351 4.7
300-399 0.3010 0.8 0.0836 4.0
500-599 0.5468 22.5 0.2752 22.9
600-799 0.1689 7.3 0.1184 26.2

Table 2.3. Transformer Failure Data

Subcomponent-related Terminal-related
Voltage
level failure rate outage time failure rate outage time
(kV) (occ./yr) (hr/occ.) (occ./yr) (hr/occ.)
110-149 0.0088 1293.2 0.0570 18.3
150-199 0.0484 23.5 0.0200 8.0
200-299 O.Ol2l 495.0 0.0582 20.6
300-399 0.0245 2386.2 0.0235 32.6
500-599 0.0142 228.8 0.0477 14.4
600-799 0.0593 4545.1 0.0385 811.6

Table 2.4. Circuit Breaker Failure Data

Subcomponent-related Terminal-related
Voltage
level failure rate outage time failure rate outage time
(kV) (occ./yr) (hr/occ.) (occ./yr) (hr/occ.)
110-149 0.0101 422.0 0.0268 21.2
150-199 0.0097 64.8 0.0277 274.4
200-299 0.0186 267.9 0.0369 32.4
300-399 0.0229 204.2 0.013l 304.2
500-599 0.0348 151.4 0.0638 104.1
600-799 0.0810 541.0 0.0295 424.2

data are based on statistics for the period January 1, 1986, to December 31,
1990.

2.5.3. System Performance Data Collection

Systems
CEA has also initiated an Electric Power System Reliability Assessment
(EPSRA) procedure(7) designed to provide data on the past performance of
BASIC CONCEPTS OF POWER SYSTEM RELIABILITY EVALUATION 21

Electrical Power System

Reliabili~ Assessment
PSRA

/
Bulk Electricity Systems
\
Distribution Systems

~a~Significant Power Interruptions (a) Service Continuity Statistics

b Delivery Point Interruptions

Figure 2.9. Basic structure of EPSRA.

an overall power system. This procedure is in the process of evolution and,

at the present time, contains systems for compiling information on bulk
system disturbances, bulk system delivery point performance, and customer
service continuity statistics. In the latter area, these data have been collected
and published for many years by the Distribution Section of the Engineering
and Operating Division of CEA. The basic structure of EPSRA is shown in
Figure 2.9.
The bulk electricity system (BES) parameters provide valuable HL2
data. The service continuity statistics collected at the distribution system
level provide overall HL3 indices. The following is a brief description of the
salient features of the EPSRA components.

(a) BE5-Significant Power Interruptions. This is a relatively

simple reporting procedure in which each participating utility provides
annual data on the frequency and severity of significant power interruptions
on its system. Typically, these events will involve widespread customer inter-
ruptions or localized power interruptions of an extended duration. It is
expected that these events will occur infrequently. Disturbance severity is
defined as the unsupplied energy in an event and is measured in MW-
minutes. This is transformed into "System Minutes" by dividing the unsup-
plied energy by the annual peak system load.

(b) BES-Delivery Point Interruptions. This reporting pro-

cedure was initiated on January 1, 1988, and is intended to provide a central-
ized source of BES delivery point performance data at the national level
which will allow utilities to compare their performance with that of other
Canadian utilities. The measurement system will focus on the collection of
all delivery point interruptions due to problems within the BES and therefore
22 CHAPTER 2

Table 2.5. Canadian Service Continuity Statistics

System average interruption frequency index = 4.03 int/syst-cust
System average interruption duration index=4.55hr/syst-cust
Customer average interruption duration index = 1.13 hr/cust
Index of reliability = 0.999480

will include both adequacy and security considerations. The interruptions

will be divided into momentary and sustained events. It is expected that
this reporting procedure will provide some very important information for
Canadian utilities.

(c) Service Continuity Statistics. Service continuity statistics

have been collected by many Canadian utilities for over thirty years. These
data were previously compiled by the Distribution Section of the CEA
Engineering and Operating Division. The responsibility for the procedure
was assigned to CEA headquarters in 1986. The report contains individual
utility statistics and overall Canadian data and, in 1990, covered 6.5 million
electric system customers. Table 2.5 shows the basic service continuity statis-
tics on a national basis for the 1990 period.
These indices are HL3 parameters and provide a valuable indication of
customer service. The report also provides a breakdown of the primary
causes of interruption. Approximately 23% of the recorded customer inter-
ruptions were due to loss of supply, which is a measure of the BES contribu-
tion to customer service levels. Approximately 17% of the customer hours
of interruption were attributed to loss of supply.

2.6. ADEQUACY INDICES

There are many possible indices which can be used to measure the
adequacy of a power system at one of the three hierarchical levels and
different countries/utilities use different indices. The following section
presents some widely used power system adequacy indices. These indices are
used in the following chapters. Most adequacy indices are basically expected
values of a random variable, although the probability distribution can be
calculated in some cases. There is considerable confusion both inside and
outside the power industry on the specific meaning of expectation indices and
the use that can be made of them. An expected value is not a deterministic
parameter. It is the long-run average of the phenomenon under study. Expec-
tation indices provides valid adequacy indicators which reflect various fac-
BASIC CONCEPTS OF POWER SYSTEM RELIABILITY EVALUATION 23

tors such as system component availability and capacity, load characteristics

and uncertainty, system configurations and operational conditions, etc.

2.6.1. Adequacy Indices in H L1 Studies

The basic indices in generating system adequacy assessment are loss of
load expectation (LOLE), loss of energy expectation (LOEE), loss of load
frequency (LOLF), and loss ofload duration (LOLD) and can be calculated
using quite different approaches. Chapter 4 presents the equations for calcul-
ating these indices using Monte Carlo methods. Conceptually, these indices
can be described by the following mathematical expressions:

(1) LOLE (days/yr or hr/yr)

LOLE= LPiT (2.1)
iE8

where Pi is the probability of system state i and S is the set of all system
states associated with loss of load. The LOLE is the average number of days
or hours in a given period (usually one year) in which the daily peak load
or hourly load is expected to exceed the available generating capacity.
It should be noted that the LOLE index in days/yr or in hr/yr has
quite different meanings. When it is in days/yr, Pi depends on a comparison
between the daily peak load and the available generating capacity. When it
is in hr/yr, Pi depends on a comparison between the hourly load and the
available generating capacity. The LOLE index does not indicate the severity
of the deficiency nor the frequency nor the duration of loss of load. Despite
these shortcomings, it is at present the most widely used probabilistic
criterion in generating capacity planning studies.

(2) LOEE (MWh/yr)

LOEE = L 8760CiPi (2.2)
iE8

where Pi and S are as defined above; C is the loss of load for system state
i. The LOEE index is the expected energy not supplied by the generat-
ing system due to the load demand exceeding the available generating
capacity.
The LOEE incorporates the severity of deficiencies in addition to the
number of occasions and their duration, and therefore the impact of energy
shortfalls as well as their likelihood is evaluated. It is hence believed that
this index will be used more widely in the future, particularly for situations
24 CHAPTER 2

in which alternate energy replacement sources are being considered. The

complementary value of energy not supplied, i.e., energy actually supplied,
can be divided by the total energy demanded to give a normalized index
known as the energy index of reliability (EIR). This index can be used to
compare the adequacy of systems that differ considerably in size.

(3) LOLF (occ.jyr)

LOLF= L (Fi-f;) (2.3)
ieS

where Fi is the frequency of departing system state i and f; is the portion of

Fi which corresponds to not going through the boundary wall between the
loss-of-Ioad state set and the no-Ioss-of-Ioad state set.

(4) LOLD (hrjdisturbance)

LOLD= LOLE (2.4)

LOLF
Frequency and duration are a basic extension of the LOLE index in
that they identify the expected frequency of encountering a deficiency and
the expected duration of the deficiencies. They contain additional physical
characteristics which makes them sensitive to additional generation system
parameters and provide more information to power system planners. The
criterion has not been used very widely in generation system reliability ana-
lysis, although the concept of frequency and duration of failures is used
extensively in transmission or composite system studies.

2.6.2. Adequacy Indices in H L2 Studies

The adequacy index concepts used in HLl studies can be extended to
composite system adequacy assessment. More indices, however, are required
to reflect composite system features. References 3 and 10 present a range of
adequacy indices for HL2 studies. The following indices are used in this
book:

(1) PLC-Probability of Load Curtailments

PLC= L Pi (2.5)
ieS

where Pi is the probability of system state i and S is the set of all system
states associated with load curtailment.
BASIC CONCEPTS OF POWER SYSTEM RELIABILITY EVALUATION 25

(2) EFLC-Expected Frequency of Load Curtailments (occ./yr).

This index has the same definition as the LOLF expressed by equation
(2.3) at HLl studies. In composite system adequacy assessment, it is a diffi-
cult task to calculate the frequency index. This is due to the fact that for
each load curtailment state i, it is necessary to identify all the no-Ioad-
curtailment states which can be reached from state i in one transition.
Expected Number of Load Curtailments (ENLC) is often used to replace
the EFLC index:
ENLC=I Fj (2.6)
jeS

The ENLC is the sum of occurrences of load curtailment states and

therefore an upper bound of the actual frequency index. It is possible to
have a large difference between the EFLC and the ENLC. The ENLC may
lead to an overestimation of the actual frequency. In many HL2 studies,
however, absolute adequacies of the systems are not required. The basic
planning task is to compare different alternatives, which only requires a
relative comparison between the adequacy indices of the alternatives. In
many cases, the ENLC index provides a sufficiently reasonable indication.
The system state frequency Fj can be calculated by the following relationship
between the frequency and the system state probability Pi:
(2.7)

where Ak is the departure rate of the component corresponding to system

state i and N is the set of all possible departure rates corresponding to state i.

(3) EDLC-Expected Duration of Load Curtailments (hr/yr)

EDLC = PLC x 8760 (2.8)

(4) ADLC-Average Duration of Load Curtailments (hr/disturbance)

ADLC = EDLC/EFLC (2.9)

This index is similar to the LOLD index in HLl studies.

(5) ELC-Expected Load Curtailments (MW/yr)
ELC= I C;Fj (2.10)
ieS

where Cj is the load curtailment in system state i.

26 CHAPTER 2

(6) EDNS-Expected Demand Not Supplied (MW)

EDNS=L CiPi (2.11)
ieS
It should be noted that the ELC and the EDNS are different indices.

(7) EENS-Expected Energy Not Supplied (MWh/yr)

EENS= L CiFiDi= L 8760CiPi (2.12)
ieS ieS
where Di is the duration of system state i. This index is similar to the LOEE
index in HLl studies. It is an important index in composite system adequacy
assessment.

(8) BPII-Bulk Power Interruption Index (MW/MW-yr)

BPII= L C;F;jL (2.13)
ieS
where L is the annual system peak load in MW. This index can be interpreted
as the equivalent per-unit interruption of the annual peak load. One complete
system outage during peak load conditions ·contributes 1.0 to this index.

(9) BPECI-Bulk Power/Energy Curtailment Index (MWh/MW-yr)

BPECI= EENS/L (2.14)

(10) BPACI-Bulk Power-Supply Average MW Curtailment Index

(MW/disturbance)
BPACI = ELC/EFLC (2.15)

(11) MBPCI-Modified Bulk Power Curtailment Index (MW/MW)

MBPCI = EDNS/L (2.16)

(12) SI-Severity Index (system min/yr)

SI = BPECI x 60 (2.17)
This index can be interpreted as the equivalent duration (in min) of the loss
of all load during peak load conditions. One complete system outage during
peak conditions contributes by its duration (in min) to this index.
It can be seen that items (1) to (7) are basic indices and items (8) to
(12) are calculated from these basic indices. The advantage of indices (8) to
(12) is that they can be used to compare adequacies of systems having
BASIC CONCEPTS OF POWER SYSTEM RELIABILITY EVALUATION 27

different sizes. Indices (1) to (7) can be applied to either a whole system or
to one single bus. Indices (8) to (12) only apply to an overall system. These
indices can be calculated at the peak load and expressed on a one-year basis
(annualized index) or by considering the annual load duration curve (annual
index).

2.6.3. Adequacy Indices in Distribution

System Evaluation
There are three basic load-point indices in distribution system adequacy
assessment. They are load-point failure rate A, load-point outage duration
r, and load-point annual unavailability U. The analytical equations and
Monte Carlo simulation formulas for these three basic indices are given in
Chapter 6. Overall distribution system performance indices can be calculated
using the three basic load-point indices. The definitions and formulas for
the system performance indices are as follows. These indices can be used to
indicate the past or future performance of the system.

(1) SAIFI-System Average Interruption Frequency Index (inter-

ruptions/system customer/yr)

SAIFI =
L· .E
RA;Ni
(2.18)
LiERNi
where Ai and Ni are the failure rate and the number of customers at load
point i respectively; R is the set of load points in the system.

(2) SAIDI-System Average Interruption Duration Index (hr/system

customer/yr)
~. R U;Ni
SAIDI = _L....:..::..=..::....-_
E
(2.19)
LiER Ni
where Ui is the annual unavailability or outage time (in hr/yr) at load
point i.

(3) CAIFI-Customer Average Interruption Frequency Index (inter-

ruptions/customer affected/yr)

(2.20)
28 CHAPTER 2

where Mi is the number of customers affected at load point i. The customers

affected should be counted only once, regardless of the number of interrup-
tions they may have experienced in the year. This index is particularly useful
when a given calendar year is compared with other calendar years since, in
any given calendar year, not all customers will be affected and many will
experience complete continuity of supply.

(4) CAIDI-Customer Average Interruption Duration Index (hr/cus-

tomer interruption)

LieR UNi SAIDI

CAIDI (2.21)
LieR ANi SAIFI

(5) ASAI-Average Service Availability Index

~ "eR 8760Ni- ~" R UNi

ASAI = ,-" '-',e (2.22)
LieR8760Ni

(6) ASUI-Average Service Unavailability Index

~" R UNi
ASUI = _'-'.:.='e=----__ (2.23)
LieR 8760Ni

(7) ENS-Energy Not Supplied (kWh/yr)

ENS= L PaiUi (2.24)
ieR
where Pai is the average load (in kW) connected to load point i and Ui is
the annual outage time (in hr/yr) at the load point.

(8) AENS-Average Energy Not Supplied (kWh/customer/yr)

ENS
AENS (2.25)

(9) ACCI-Average Customer Curtailment Index (kWh/customer

affected/yr )

ACCI= ENS (2.26)

LieRMi
BASIC CONCEPTS OF POWER SYSTEM RELIABILITY EVALUATION 29

2.6.4. Reliability Worth Indices

In general, it is not possible to measure reliability worth directly, and
therefore system reliability worth is usually assessed by considering unreli-
ability cost. Interruption costs are used as a surrogate of reliability worth.
The three hierarchical level studies utilize conceptually the same reliability
worth indices. They are:

(1) EIC-Expected Interruption Cost (k$jyr)

EIC= I CFiW(Di) (2.27)
ieV

where Ci is the load curtailment of load loss event i; Fi and Di are the
frequency and duration of load loss event i; W(Di) is the unit interruption
cost which is a function of the interrupted duration; V is the set of all load
loss events.

(2) lEAR-Interrupted Energy Assessment Rate ($jkWh)

lEAR = EICjEENS (2.28)
where EENS is the expected energy not supplied. In HLl studies, this is the
LOEE index and, in distribution system studies, it is the ENS index.
It should be noted that in generation system assessment the EIC and
lEAR are system indices, while in composite system and distribution system
assessments they can either be system or load point (bus) indices depending
on the particular focus of the study.

2.7. CONCLUSIONS
This chapter describes several philosophical aspects of power system
reliability. These basic concepts are fundamental to an appreciation of the
applications proposed in Chapters 4-7. Reliability assessment of a power
system can be divided into the two aspects of system adequacy and system
security. Most presently available reliability evaluation techniques are in the
domain of adequacy assessment. There should be some conformity between
the reliability of various segments of the system and a balance is required
between the generation, transmission, and distribution functional zones. This
does not imply that the reliability of each should be equal. Different levels
of reliability are justified as different functional zones are associated with
different impact ranges. Generation and transmission failures can cause
30 CHAPTER 2

widespread outages while distribution failures tend to be highly localized.

The most useful concept for planning purposes is to utilize a reliability cost/
worth approach. The main difficulty with such a concept is the perceived
uncertainty in the consumers' valuation.
Reliability data collection is an essential and important aspect of quanti-
tative reliability evaluation. Data collection and reliability methodology
should evolve together. Updating reliability data is a long-term and continu-
ous task. Adequacy assessment at different hierarchical levels utilizes differ-
ent indices, although they are fundamentally similar in a probabilistic sense.
These indices are expected values of probability distributions which are not
deterministic parameters. This is a very important point which should be
clearly understood.
Power system reliability evaluation has been highly developed over the
last twenty years and there are many publications available on this
subject.(1l-17) Billinton et al.(17) contains a collection of important publica-
tions which highlight important developments and fundamental techniques.

2.B. REFERENCES
1. R. Billinton and R. N. Allan, Reliability Evaluation of Engineering Systems: Concepts and
Techniques, Plenum, New York, 1983.
2. R. Billinton and P. R. S. Kuruganty, "A Probabilistic Index for Transient Stability," IEEE
Trans. on PAS, Vol. 99, 1980, pp. 195-207.
3. R. Billinton and R. N. Allan, Reliability Evaluation of Power Systems, Plenum, New York,
1984.
4. R. Billinton and R. N. Allan, Reliability Assessment of Large Electric Power Systems,
Kluwer, Boston, 1988.
5. R. Billinton, H. J. Koglin, and E. Roos, "Reliability Equivalents in Composite System
Reliability Evaluation," lEE Proc. C, Vol. 134, 1987, pp. 224-233.
6. R. Billinton, G. Wacker, and E. Wojczynski, "Comprehensive Bibliography of Electrical
Service Interruption Costs," IEEE Trans. on PAS, Vol. 102, 1983, pp. 1831-1837.
7. R. Billinton, K. Debnath, M. Oprisan, and 1. M. Clark, "The CEA Equipment Reliability
and System Performance Reporting Procedure," Transactions of the Canadian Electrical
Association, 1988.
8. CEA, ERIS 1990 Annual Report-Generation Equipment Status, 1991.
9. CEA, ERIS 1990 Annual Report-Forced Outage Performance of Transmission Equip-
ment, 1991.
10. Working Group on Performance Records for Optimizing System Design, Power System
Engineering Committee, "Reliability Indices for Use in Bulk Power Supply Adequacy
Evaluation," IEEE Trans. on PAS, Vol. 97, No.4, 1978, pp. 1097-1103.
II. R. Billinton, "Bibliography on the Application of Probability Methods in Power System
Reliability Evaluation," IEEE Trans. on PAS, Vol. 91, 1972, pp. 649-660.
12. IEEE Committee Report, "Bibliography on the Application of Probability Methods in
Power System Reliability Evaluation, 1971-1977," IEEE Trans. on PAS, Vol. 97, 1978,
pp. 2235-2242.
BASIC CONCEPTS OF POWER SYSTEM RELIABILITY EVALUATION 31

13. R. N. Allan, R. Billinton, and S. H. Lee, "Bibliography on the Application of Probability

Methods in Power System Reliability Evaluation, 1977-1982," IEEE Trans. on PAS,
Vol. 103, 1984, pp. 275-282.
14. R. N. Allan, R. Billinton, S. M. Shahidehpour, and C. Singh, "Bibliography on the Applica-
tion of Probability Methods in Power System Reliability Evaluation, 1982-1987," IEEE
Trans. on PS, Vol. 3, No.4, 1988, pp. 1555-1564.
15. G. Tollefson, R. Billinton, and G. Wacker, "Comprehensive Bibliography on Reliability
Worth and Electrical Service Consumer Interruption Costs: 1986-1990," IEEE Trans. on
PS, Vol. 6, No.4, 1991, pp. 1508-1514.
16. CIGRE Working Group 38.03, "Power System Reliability Analysis-Application Guide,"
CIGRE Publications, Paris, 1988.
17. R. Billinton, R. N. Allan, and L. Salvaderi, Applied Reliability Assessment in Electric Power
Systems, IEEE Press, New York, 1991.
 3
Elements of Monte
Carlo Methods

3.1. INTRODUCTION
The Monte Carlo method is the general designation for stochastic simulation
using random numbers. Monte Carlo is the name of the suburb in Monaco
made famous by its gambling casino. The name was also used as the secret
code for atomic bomb work performed during World War II involving
random simulation of the neutron diffusion process. Monte Carlo methods
have been used in many areas since that time.
The basic concept of the Monte Carlo method dates back to the 18th
century when the French scientist Buffon presented the famous needle throw
test method(l) to calculate tr in 1777. The method is as follows. A needle of
length d is thrown randomly onto a plane on which some parallel lines with
equal width a have been drawn, where d < a. It can be shown that the
probability of the needle hitting a line is P=2d/tra. Since the probability
can be estimated as the ratio of the number of throws hitting a line to the
total number of throws, the value of tr can be obtained by tr = 2d/Pa. This
is the earliest and most interesting example of Monte Carlo method
application.
This example indicates that the Monte Carlo method can be used to
solve not only stochastic but also deterministic problems. Applications of
Monte Carlo techniques can be found in many fields such as complex mathe-
matical calculations, stochastic process simulation, medical statistics, engin-
eering system analysis, and reliability evaluation.
This book applies Monte Carlo methods to power system reliability
evaluation. In this chapter, the basic concepts of the Monte Carlo
method are presented and discussed from a reliability evaluation point of
view.
33
34 CHAPTER 3

3.2. GENERAL CONCEPTS

3.2.1. Two Simple Examples

Two simple examples are given in order to illustrate the basic concepts
of the Monte Carlo method.

Example 1: A fair die is thrown. What is the probability of a one

occurring on the upper face? This is obviously 1/6 as each of the six faces
has equal probabilities of occurring. This probability can be estimated by
sampling simulation. Throw the die N times and record the times number
one occurs. Let this be I times. The estimation of the probability is liN. As
N increases sufficiently, liN approaches 1/6.

Example 2: Calculate the following integral by sampling simulation:

1= f g(x)dx

It is well known that the integral equals the shaded area in Figure 3.1. A
point is thrown randomly N times in the range [O~x~ 1, O~y~ 1] and the

x
o 1
Figure 3.1. One-dimension integral by Monte Carlo simulation.
ELEMENTS OF MONTE CARLO METHODS 35

number of points hitting the shaded area is M. The integral is therefore

equal to the probability of the point hitting the shaded area, namely,
M
I=p-:::::,-
N
The sampling simulation can be conducted on a computer. Two
uniformly distributed random numbers Xi and Yi between [0, 1] can be gener-
ated and checked to see if the inequality Yi~g(Xi) is satisfied or not to obtain
M. The most basic aspect of Monte Carlo simulation is the generation of
random numbers. This is discussed in detail later in this chapter.

3.2.2. Features of Monte Carlo Methods in

Reliability Evaluation
A fundamental parameter in reliability evaluation is the mathematical
expectation of a given reliability index. Salient features of the Monte Carlo
method for reliability evaluation therefore can be discussed from an expecta-
tion point of view.
Let Q denote the unavailability (failure probability) of a system and Xi
be a zero-one indicator variable which states that
Xi =0 if the system is in the up state
Xi = 1 if the system is in the down state
The estimate of the system unavailability is given by
_ 1 N
Q=- L Xi (3.1)
N i =1
where N is the number of system state samples.
The unbiased sample variance is
1 N
V(x)=- L (Xi-Q)2 (3.2)
N-l i =1
When the sample size is large enough, equation (3.2) can be approximated
by
1 N
V(x)=- L (Xi- Q)2 (3.3)
N i =1
Because Xi is a zero-one variable, it follows that
N N
L X;= LXi (3.4)
i=1 i=1
36 CHAPTER 3

Substituting equations (3.1) and (3.4) into equation (3.3) yields

IN IN _IN_
V(X)=- L
Ni=l
xt--Ni=lL 2XiQ+-Ni=lL 12
=Q-2i2+i2
=Q-i2 (3.5)
It is important to note that equation (3.1) gives only an estimate of the
system unavailability. The uncertainty around the estimate can be measured
by the variance of the expectation estimate:
- 1
V(Q)=- V(x)
N
1 - 712
=-(Q-~ ) (3.6)
N
The accuracy level of Monte Carlo simulation can be expressed by the
coefficient of variation, which is defined as

a = JV(Q)/Q (3.7)
Substitution of equation (3.6) into equation (3.7) gives

a=Jl;QQ (3.8)

Equation (3.8) can be rewritten as

N=I-Q (3.9)
a 2Q

This equation indicates two important points:

1. For a desired accuracy level a, the required number of samples N
depends on the system unavailability but is independent of the size of the
system. Monte Carlo methods are therefore suited to large-scale system
reliability evaluation. This is an important advantage of Monte Carlo
methods compared to analytical enumeration techniques for reliability
evaluation.
2. The unavailability (failure probability) in practical system reliability
evaluation is usually much smaller than 1.0. Therefore,
1
N~~ (3.10)
aQ
ELEMENTS OF MONTE CARLO METHODS 37

This means that the number of samples N is approximately inversely propor-

tional to the unavailability of the system. In other words, in the case of a
very reliable system, a large number of samples is required to satisfy the
given accuracy level.

3.2.3. Efficiency of Monte Carlo Methods

Different Monte Carlo techniques can be used to solve the same prob-
lem. These include different random number generation methods, different
sampling approaches, and different variance reduction techniques, etc. It is
therefore sometimes neccessary to compare the efficiency of different Monte
Carlo methods.
Suppose two Monte Carlo methods are used to evaluate the same system
and the expectation estimates of the reliability index obtained using these
two methods are statistically the same. Let II and 12 denote computing times
and O'r and O'~ be the variances of the reliability index for the two methods,
respectively. If the ratio
I1O'r I (3.11)
"'=-2<
120'2

then the first method can be considered to be more efficient than the second
method. The efficiency of the Monte Carlo method depends on the comput-
ing time multiplied by the variance of the estimate, but not simply on the
number of required samples.
In conducting reliability evaluation of power systems using Monte Carlo
methods, the computing time and the variance are directly affected by the
selected sampling techniques and system analysis requirements. Sampling
techniques include random number (or variate) generation methods, vari-
ance reduction techniques, and different sampling approaches and are dis-
cussed in the following sections of this chapter. The purpose of system
analysis is to judge if a system state is good or bad. The system analysis
requirements are different in generation (HLI), composite (HLII), and distri-
bution system reliability evaluation and are discussed in Chapters 4,5, and
6, respectively.

3.2.4. Convergence Characteristics of

Monte Carlo Methods
(a) Convergence Process. Monte Carlo simulation creates a fluc-
tuating convergence process as shown in Figure 3.2 and there is no guarantee
38 CHAPTER 3

estimate value

accurate value 1--""""'l~+f--->C""""'lH~---\-----:~~Pr-

number of samples
Figure 3.2. Convergence process in Monte Carlo simulation.

that a few more samples will definitely lead to a smaller error. It is true,
however, that the error bound or the confidence range decreases as the
number of samples increases.

(b) Convergence Accuracy. The variance of the expectation esti-

mate is given by equation (3.6). The standard deviation of the estimate can
be obtained as follows:
- JV(x)
a=JV(Q)=-- (3.12)
.jN
This indicates that two measures can be utilized to reduce the standard
deviation in a Monte Carlo simulation: increasing the number of samples
and decreasing the sample variance. Variance reduction techniques can be
used to improve the effectiveness of Monte Carlo simulation. The variance
cannot be reduced to be zero and therefore it is always necessary to utilize
a reasonable and sufficiently large number of samples.

(c) Convergence Criteria. The coefficient of variation shown in

equation (3.7) is often used as the convergence criterion in Monte Carlo
simulation. In power system reliability evaluation, different reliability indices
have different convergence speeds. It has been found that the coefficient of
variation of the EENS index has the lowest rate of convergence. This
coefficient of variation should therefore be used as the convergence criterion
in order to guarantee reasonable accuracy in a multi-index study.
ElEMENTS OF MONTE CARLO METHODS 39

3.3. RANDOM NUMBER GENERATION

3.3.1. Introduction
A random number can be generated by either a physical or a mathemati-
cal method. The mathematical method is most common as it can guarantee
reproducibility and can be easily performed on a digital computer. A random
number generated by a mathematical method is not really random and
therefore is referred to a pseudorandom number. In principle, a pseudo-
random number sequence should be tested statistically to assure its
randomness.
The basic requirements for a random number generator are:
1. Uniformity: The random numbers should be uniformly distributed
between [0, 1].
2. Independence: There should be minimal correlation between random
numbers.
3. Long period: The repeat period should be sufficiently long.
There is a wide range of random number generation methods. Two
commonly used congruential generators are given in the following sections.

3.3.2. Multiplicative Congruential Generator

The multiplicative congruential generator was presented in 1949 by D.
H. Lehmer(2) and is based on the following recursive relationship:

Xi+! =axi(mod m) (3.13)

where a is a multiplier and m is the modulus; a and m have to be nonnegative
integers. The module notation (mod m) means that
Xi+! =axi-mki (3.14)
where k i = [ax;/m] denotes the largest positive integer in ax;/m. For example,
assume that aXi = 32 and m = 30,
1
30 T32
30
2
therefore 32(mod 30) = 2.
40 CHAPTER 3

Given an initial value xo, equation (3.13) generates a random number

sequence which lies between [0, m]. A random number sequence in the inter-
val [0, 1] can be obtained by

Xj
Uj =- (3.15)
m
Obviously, such a sequence will repeat itself in at most m steps and therefore
it will be periodic. If the period of the sequence equals m, the random number
generator is considered to have a full period.
Different choices of the parameters a and m produce large impacts on
the random number statistical features. If these parameters are properly
chosen, the initial value Xo has little or no effect on the statistical features
of the generated random numbers. The conventional principles in choosing
the parameters are(3)

I. m = 2k and k equals the integer word-length of the computer

2. a = 8d ± 3 and d is any positive integer
3. Initial value Xo is any odd number
These principles are, however, not completely satisfactory. When these
three principles are satisfied, the maximum period length of a generated
random number sequence is 2k - 2• This means that in the interval [0, m-I],
only m/4 integers can probably appear and the other 3m/4 integers cannot
be made use of. In 1965, D. W. Hutchinson (see Zaigen(4» proved that, if
the modulus m is a prime number instead of 2k, the maximum possible
period length can be m-l. When the integer word-length of the computer
is 31, for example, the largest prime on the computer is 231 _1. Therefore
m = 231 - 1 is a good choice. It is more difficult to choose a good multiplier
a. A number of statistical test results indicate that the following two multi-
plier values produce satisfactory statistical features in the random numbers:
a= 16807 or 630360016.
An overflow problem may occur on a computer, however, when m =
231 _1 and a= 16807 or 630360016. This is because places ofaxj in the binary
system can be as high as 46 for a= 16807 and 53 for a=630360016. Several
alternatives have been proposed to overcome this difficulty. The following
is' one of these alternatives and its proof can be found in Fang Zaigen. (4)
Choose two integers b and c which satisfy
O<b<m
O::;;c<a
ab+c=m
ELEMENTS OF MONTE CARLO METHODS 41

The new multiplicative congruential generator is given by

X:+l =a(x;(mod b»-k,-c (3.16)

where k;= [x;/b] denotes the largest positive integer in x;/b, and

ifx:+l>O
(3.17)
ifx:+l <0

When using equation (3.16), x/(mod b) is calculated instead of

ax;(mod m) so that the very large integer ax; is avoided.

3.3.3. Mixed Congruential Generator

In 1961, M. Greenberger generalized the multiplicative congruential
generator to the mixed congruential generator, which is based on the follow-
ing congruence relationship(5):

Xi+l = (ax; + c) (mod m) (3.18)

It can be seen that, in addition to parameters a and m, a new parameter c

is added into the multiplicative congruential generator. The quantity c is
called the increment and it also has to be a nonnegative integer.
The mixed congruential generator can have a full period if and only if
the following conditions are satisfied(3):

1. c is relative prime to m
2. a(mod g) = 1 for every prime factor g of m
3. a(mod 4) = 1 if m is a multiple of 4
These are sufficient and necessary conditions for the mixed congruential
generator to have a full period. These conditions, however, do not guarantee
good statistical features of generated random numbers. As in the case of the
multiplicative congruential generator, the choices of parameters a, c, and m
have quite large impacts on the statistical features of the random numbers.
Choosing "good" parameters is always a difficult task. D. E. Knuth sug-
gested using the following principles in selecting the parameters(6):

1. m = 2k and k equals the integer word-length of a computer

2. a(mod 8) = 5 and a satisfies:

m
-<a<m-Jiij
100
42 CHAPTER 3

3. c is an odd number and satisfies:

c I I
-~---,j3~0.21132
m 2 6

A number of statistical test results indicates that the following two sets
of parameters give quite satisfactory statistical features in the generated
random numbers:

a = 314159269, c = 453806245
c=1
It is noteworthy that there always exists a very weak correlation between
random numbers generated by the congruential generators. Greenberger
showed that the correlation coefficient between Xi and Xi+ 1 has the following
upper and lower bounds(7):

(3.19)

In the case of the multiplicative congruential generator (i.e., c = 0),

when a = fm, the correlation coefficient p achieves its minimum upper limit,
which equals 2/ fm. This indicates that if m is large enough, correlation
between generated random numbers can be very weak.

3.4. RANDOM VARIATE GENERATION

3.4.1. Introduction
A random variate refers to a random variable following a given distribu-
tion. The random number generation methods given in the previous section
are essentially ones which generate a random variate following a uniform
distribution between [0, I]. Generators of random variates which follow
other distributions are based on uniformly distributed random numbers
between [0, I].
The procedures for generating nonuniformly distributed random vari-
ates can be generally catagorized into three techniques: (I) inverse transform
method; (2) composition method; and (3) acceptance-rejection method.
There are also particular methods for specific distributions. (8) This section
emphasizes the inverse transform method, which is most frequently used.
The exponential and normal distributions are the most important ones in
ELEMENTS OF MONTE CARLO METHODS 43

reliability evaluation. Some techniques for generating random variates for

these two distributions are therefore discussed in detail. Methods for other
distributed random variates are given in Section 3.4.6.

3.4.2. Inverse Transform Method

The inverse transform method is based on the following proposition:

If a random variate U follows a uniform distribution in the interval

between [0,1], the random variate X=F-1(U) has a continuous cumulative
probability distribution function F( x).

The proof of the proposition is straightforward. Because X =F- 1( U),

P(X ~x) = P(F- 1( U) ~x) (3.20)

If F(x) is a cumulative distribution function, F(x) is a monotonic increasing

function. This leads to

P(F- 1( U) ~x) = P( U ~F(x» (3.21)

As U is a uniformly distributed random variate,

P[U~F(x)]=F(x) (3.22)

Therefore

P(X ~x) = F(x) (3.23)

This means that F(x) is the cumulative distribution function of X obtained

by setting X=F-1(U).
This proposition can be generalized to the case of a discrete distribution
and, in this case, the inverse function of F(x) is defined as

X=F-1(U)=min{x: F(x)~ U} (O~ U~ 1) (3.24)

The proposition is clarified by Figure 3.3.

The procedure for generating random variates using the inverse trans-
form method is as follows:
Step 1: Generate a uniformly distributed random number sequence U
between [0, 1].
Step 2: Calculate the random variate which has the cumulative prob-
ability distribution function F(x) by X=F-1(U).
 44 CHAPTER 3

U = F(X)

1~------------------------------~=-=--

x
Figure 3.3. Explanation of the inverse transform method.

Example: Given the following probability density function:

O~x~1
otherwise

the cumulative probability distribution function is

x<O
O~x~1

x>1

The random variate which follows this cumulative probability distribution

function is obtained using

(O~ U~ 1)

where U is a uniformly distributed random number sequence between [0, 1].

ELEMENTS OF MONTE CARLO METHODS 45

3.4.3. Tabulating Technique for Generating

Random Variates
This is an approximate technique for generating random variates and
is based on the inverse transform method. It has reasonable accuracy, but
requires much less computing time than the direct inverse transform method.
The tabulating technique involves two procedures: tabulating and sampling.

(a) Tabulating Procedure. The values of a cumulative probability

distribution function range from zero to one. The interval between [0, 1] is
divided into k subintervals which have the same length 11k. All values in
each subinterval are represented by their midpoint value. The number k
should be sufficiently large, i.e., k = 500.
Using the inverse transform method, the k values of a random variate
following the cumulative probability distribution function F(x) can be calcu-
lated using

I
(i-O.5)
_F- 1 - -
x·- k (i= 1, ... , k) (3.25)

This is shown in Figure 3.4. The subinterval numbers, values of the cumul-
ative probability distribution function (CPDF), and values of the random
variates (RV) are shown in Table 3.1.

(b) Sampling Procedure. This is a "table-look-up" procedure

which can be described as follows:
Step 1: Generate a uniformly distributed random number between
[0, 1],
Step 2: Judge in which subinterval the generated random number falls,
Step 3: Pick up the value of the random variate from the formed table
according to the subinterval number.
The advantage of the tabulating technique is that it is not necessary to
calculate the inverse function value for each sample as the k values of the
inverse function of F(x) are calculated in advance. The tabulating technique
is therefore much faster than the direct inverse transform method.
The tabulating technique is a discretized approach in which a continu-
ous cumulative probability distribution function is changed into a discrete
one. The purpose of the discretization, however, is to perform sampling but
not to conduct enumerating. The tabulating technique involves two errors:
a discretization error and a truncation error. When the number of subinterv-
als is sufficiently large, the discretization error is smaller than the standard
46 CHAPTER 3

F(X)

1~--------------------------~~==-

(k - 1)1k

11k

O~~L-~~-------------------------'X

Figure 3.4. Calculating k values of a random variate.

deviation of sampling and therefore negligible. The truncation error is caused

by the last subinterval. In order to avoid this error, the direct inverse trans-
form method can be used only for the last subinterval. If the generated
random number falls in the last interval, the inverse function value of F(x)
is calculated instead of "looking up the table." The probability of a random
number falling in the last subinterval is 11k and is very small when k is
sufficiently large, which leads to little increase in computing time.
The tabulating technique has sufficient accuracy for most practical
applications. For example, when the number of subintervals is 500, the
square root error of a normal distribution random variate obtained using
the tabulating technique is 0.005.

Table 3.1. Tabulating Technique Table

Subinterval Values of Values of
No. CPDF RV
I O.5/k XI
2 1.5/k X2
3 2.5/k X3

k k-O.5/k Xk
ELEMENTS OF MONTE CARLO METHODS 47

3.4.4. Generating Exponentially Distributed

Random Variates
(a) Method 1: The Inverse Transform Method. An exponen-
tially distributed random variate has the probability density function

(3.26)

Its cumulative probability distribution function is

F(x)=I-e- Ax (3.27)

By the inverse transform method

U=F(x) = l-e- Ax

so that

_ 1
X=F I(U)=--ln(l-U) (3.28)
A

Since (1 - U) distributes uniformly in the same way as U in the interval

[0, 1],

1
X=--ln U (3.29)
A

where U is a uniformly distributed random number sequence and X follows

an exponential distribution.

(b) Method 2: Alternate Method. In reliability evaluation, it is

sometimes necessary to simultaneously generate several exponentially distri-
buted random variates. In this case, the inverse transform method requires
calculation of the natural logarithm for each variate. An alternate method
can be used which requires only one computation of the natural logarithm
for generating n exponential distribution random variates. This alternate
method is based on the following proposition(1):

UI , ... , Un, Un+ I, . . . , U2n - 1 are independent and uniformly distributed

random numbers. If Ui, ... , U~_I represent the order statistics corresponding
to the random numbers Un+ l , ••• , U2n - 1 and define Ub=O and U~= 1, then
48 CHAPTER 3

the random variates

n U/
n

Yk=(U k- l - Uk) In (k= 1, ... , n) (3.30)

1=1

are independent and follow standard exponential distributions.

According to the proposition, the algorithm generating n standard expo-

nential distribution random variates can be stated as follows:
Step 1: Generate 2n - 1 uniformly distributed random numbers
U1 , ••• , Un, Un+ I , ••• , U2n - I ,
Step 2: Arrange the random numbers Un + I, •.. , U2n - 1 in the order of
increasing magnitudes to obtain Uj, ... , U~-t.
Step 3: Calculate Yk (k= 1, ... ,n) in terms of equation (3.30).

3.4.5. Generating Normally Distributed

Random Variates
(a) Method 1: The Approximate Inverse Transform
Method. A normally distributed random variate cannot be generated by
the accurate inverse transform method as there exists no analytical inverse
function of the normal cumulative probability distribution function F(x).
The inverse function of F(x), however, has the following approximate
expression(9) :

(3.31)

where
t=v'-2ln Q (3.32)

co=2.515517 dl = 1.432788
CI = 0.802853 d2 =0.189269
C2 =0.010328 d3 = 0.001308
The implications of z and Q are shown in Figure 3.5, where f(z) is the
standard normal probability density function:

f(z) =_1 exp(- ~) (3.33)

.j2i 2
ELEMENTS OF MONTE CARLO METHODS 49

f(z)

o z

Figure 3.5. Area under normal density function, Q(z).

The algorithm for generating a normally distributed random variate

can be stated as follows :
Step 1: Generate a uniformly distributed random number sequence U
between [0, 1],
Step 2: Calculate the normally distributed random variate X by

if 0.5 < U::;; 1.0

X={ -z~ if U=0.5
if 0::;; U<0.5

where z is obtained using equations (3.31) and (3.32) and Q in equation

(3.32) is given by

I-U if 0.5 < U::;; 1.0

Q= {
U if 0::;; U::;;0 .5

The maximum absolute error in equation (3.31) IS smaller than

0.45 x 10-4 and is generally negligible.
50 CHAPTER 3

(b) Method 2: The Box-Muller Method. Box and Muller

presented the following proposition(IO):

If VI and V 2 are two independent, uniformly distributed random numbers

between [0, 1], then
and (3.34)

are two independent, normally distributed random variates.

The algorithm for generating two normally distributed random variates

simultaneously is as follows:
Step 1: Generate two uniformly distributed random number sequences
between [0, 1],
Step 2: Calculate XI and X 2 using equation (3.34).

3.4.6. Generating Other Distribution

Random Variates
In this section, algorithms for generating log-normal, gamma, and
Weibull distributed random variates, which are useful in power system reli-
ability evaluation, are presented.

(a) Log-Normal Distributed Random Variate. In reliability

evaluation, the time to repair a component is sometimes considered as a
random variate from a log-normal distribution with the following probabil-
ity density function:

{
I [(In
exp -
y- .u )2J (O<y< 00)
/(y) = ::nay 2a 2 (3.35)
(y~O)

If X follows a normal distribution, then Y = t? follows the log-normal distri-

bution. This is because equation (3.35) can be obtained from the general
relationship between probability density functions of Yand X which is given
by equation (B.II) in Appendix B when Y = t? and X has a normal distribu-
tion density function. Note that .u and a in equation (3.35) are not the
mean and the standard deviation of the log-normal distribution. They are the
mean and the standard deviation of the normal distribution corresponding to
the log-normal distribution. The mean and the variance of the log-normal
ELEMENTS OF MONTE CARLO METHODS 51

distribution should be calculated using equations (B.22) and (B.23) in

Appendix B.
The algorithm for generating a log-normally distributed random variate
is as follows:
Step 1: Generate a random variate Z following a standard normal
distribution.
Step 2: X = Jl + (JZ, where Jl and (J are the parameters given in the
equation (3.35). If the mean and variance of the log-normal
distribution are specified, the parameters Jl and (J can be calcu-
lated using equations (B.24) and (B.25) in Appendix B.
Step 3: Y = C where Y is a log-normally distributed random variate.

(b) Wei bull Distributed Random Variate. The Weibull distribu-

tion has the following probability density function:

(3.36)

°
where ~ x <
function is
00, a> 0, and P> O. The cumulative probability distribution

F(x) = 1-exp [ -(~rJ (3.37)

By the inverse transform method,

U=F(x) = 1-exp [ -(~rJ

and
X= a[-ln(l- U)fIP (3.38)
where U is a uniformly distributed random variate between [0, 1]. Since
1- U is also a uniformly distributed random variate between [0, 1], equation
(3.38) becomes
X = a( -In U)I/P (3.39)

The algorithm for generating Weibull distributed random variates is as

follows:
Step 1: Generate a uniformly distributed random number sequence U
between [0, 1];
Step 2: Calculate X by equation (3.39).
52 CHAPTER 3

(C) Gamma Distributed Random Variate. The gamma distribu-

tion has the following probability density function:

f(x) xP-
p
1
-~
exp( )
(3.40)
a r(P) a
°
where =s;;x < 00, a >0, and P>O, and it is denoted by G(P, a). When P=
1, the gamma distribution becomes the exponential distribution with param-
eter 1/a. When P equals an integer, the gamma distribution is known as
the Erlangian distribution. It has been shown(l) that if Xi (i= 1, ... ,n) is a
sequence of independent random variates following G(Pi' a), then X=
I;=I Xi is a random variate following G(P, a) where P=I;=I Pi. The
Erlangian distribution with parameter P= n therefore can be obtained by
summing n independent exponential random variates with parameter I/a,
that is,

n Ui
n n
X=a I (-In Ui)=-a In (3.41)
i=1 i=1

The algorithm for generating gamma distributed random variates is as

follows:
Step 1: Generate n independent, uniformly distributed random number
sequences Ui (i= 1, ... ,n) between [0,1] where n is the integer
p,
Step 2: Calculate X by equation (3.41).

3.5. VARIANCE REDUCTION TECHNIQUES

3.5.1. Introduction
Equation (3.12) shows that reducing the sample variance can decrease
the standard deviation of the Monte Carlo estimator and that decreasing
the sample variance has a similar effect on the accuracy of a Monte Carlo
simulation as increasing the number of samples. The application of variance
reduction techniques is an important concept in Monte Carlo simulation.
The following points, however, should be clearly understood:
1. Variance reduction is simply a means to use known information
about the problem. Before variance reduction is achieved, a price
has to be paid to obtain and deal with required information.
2. Not all theoretical variance reduction techniques can be applied
effectively in power system reliability evaluation.
ELEMENTS OF MONTE CARLO METHODS 53

Five variance reduction techniques which are useful in power system

reliability evaluation are discussed in this section. They are: control variates,
importance sampling, stratified sampling, antithetic variates, and dagger
sampling.

3.5.2. Control Variates

The basic feature of the control variate technique is the use of informa-
tion obtained from an analytical model to reduce the variance. If the problem
contains a part which can be solved by means of an analytical model, then
Monte Carlo simulation is only used to calculate the difference between the
solution of the problem and this part. (II)
If Z is a random variate which is strongly correlated with another
random variate F, define a new random variate Y:
Y=F-Z+E(Z) (3.42)
where the mean value E(Z) can be obtained by an analytical model.
It can be seen that Y and F have the same expected value because
E( Y) = E(F) - E(Z) + E(Z) = E(F) (3.43)
The variance of Y is given by
V( Y) = V(F) + V(Z) - 2 cov(F, Z) (3.44)
where cov(F, Z) is the covariance between F and Z. Since F and Z are
strongly correlated, the absolute value of the negative term in equation (3.44)
may be larger than the positive term V(Z). In this case, therefore, the
variance of Y is smaller than the variance of F. Variate Z is called the control
variate.
This variance reduction technique can be used to evaluate system indices
of a composite generation and transmission system. Let the load curtailment
due to the generating capacity shortage be the control variate Z and that
due to the composite generation and transmission system be the variate F.
Quantities Z and F are strongly correlated and E(Z) can be easily obtained
by an analytical model such as the generating capacity outage table. The
technique cannot be used to evaluate bus indices of composite systems unless
a proper control variate for this purpose can be found.

3.5.3. Importance Sampling

Importance sampling is a procedure for changing the probability density
function of sampling in such a way that the events which make greater
54 CHAPTER 3

,,
: ,.,.,. .. --~ ....

,,.'1" '"',,
/ ~
../ \
\

/
.. I \
\
\
I \
I \
I \
I \ ,

,
,, \\ 'II
I " \

"~
1/

"
,I:, ,~\
/ : :\
/ " I, , ,
I ,

~" : : '
o 1 x
Figure 3.6. Explanation of importance sampling.

contributions to the simulation results have greater occurrence

probabilities. (12)
An integral can represent an expected value of a parameter and there-
fore the problem of estimating an integral by the Monte Carlo method is
equivalent to the problem of estimating an adequacy index in reliability
evaluation. The importance sampling technique can be illustrated using the
problem of estimating an integral.

r
Consider the following integral between [0, 1]:

1= g(x) dx (3.45)
o

Using the expected value estimation method, the integral can be estimated
by
1 N
I=E(g(V))~- L g(Xi) (3.46)
Ni=1

where V is a uniformly distributed random number sequence between [0, 1].

As shown in Figure 3.6, three random numbers VI, V2 , and V3 of x
are obtained by uniform sampling between [0,1]. These three random
ELEMENTS OF MONTE CARLO METHODS 55

numbers make different contributions to the integral value; U I has a greater

contribution than U2 and U3 , which indicates that uniform sampling is
unreasonable. If the probability density function for sampling is changed
from the uniform distribution to f(x), which has the same shape as g(x),
then the random numbers which can make a greater contribution to the
integral value have larger occurrence probabilities.
Multiplying and dividing g(x) by f(x), the integral given in equation
(3.45) can be rewritten as

1= I I g(x)
-f(x)dx
o f(x)

The new probability density functionf(x) is called the importance sampling

density function.
Let ()=g(x)/f(x). According to the definition of the expected value,
the integral equals the expected value of ():

I=E[g(x)/f(x)] (3.47)

This means that if a random variate is sampled, which has the importance
sampling density function f(x) instead of a uniformly distributed random
variate, the integral can be obtained by calculating the expected value of ().
The variance of () is

V«())=fll(X) f(x) dx-I 2 (3.48)

o f2(x)

Iff(x)=g(x)/I, the variance of () would be zero:

V«()) =1 f o
g(x) dx- 12 =0

It is impossible to makef(x) = g(x)/I because lis unknown. The vari-

ance, however, can be considerably reduced if f(x) has a shape similar to
the shape of g(x).
In power system reliability evaluation, the importance sampling tech-
nique can be used to sample load or hydrological states. It is a relatively
difficult problem in practical application to find a proper importance sam-
pling density functionf(x). A trial Monte Carlo sampling can be conducted
to develop an importance sampling density function f(x). This requires
additional computing time.
56 CHAPTER 3

Figure 3.7. Explanation of stratified sampling.

3.5.4. Stratified Sampling

The idea of stratified sampling is similar to that of importance sampling.
In order to reduce the variance, more samples are drawn in the subintervals
which make greater contributions to the simulation results. (13)
Consider the integral between [0, 1] given in equation (3.45). The inter-
val [0, 1] is divided into m disjoint subintervals fit U= 1, ... ,m), as shown
in Figure 3.7. The total integral can be expressed as

1= I0
I m

g(x) dx= j~1 ~ (3.49)

where ~ is the integral in the jth subinterval. By the expected value estimation
method, ~ can be estimated by

(3.50)
ELEMENTS OF MONTE CARLO METHODS 57

where ~ is a uniformly distributed random number sequence in the jth

subinterval and Nj is the number of random numbers in the jth subinterval.
The total integral therefore can be estimated by

m
= L ~=I (3.51)
j=1

This indicates that () is the unbiased estimator of I. The variance of () is

m d~ NJ
= L ~ L varfg(~)]
j=1 N j ;=1
m d~
= L -Lvarfg(~)] (3.52)
j=l~
where the estimation of var[g( [f;)] can be obtained by
I N, I~
Vfg(~)]=-- L l(x;)-~ (3.53)
~-I ;=1 dj

It has been shown that if the sample size Nj in each subinterval is

proportional to ~, then the variance of the stratified sampling given in
equation (3.52) is smaller than or at most equal to the variance obtained
when applying the expected value estimation method to the whole interval
[0,1]. When

N.=N ~Uj
] L7=1 ~Cj

where N=L~=I ~ and Uj= {varfg(~)]}1/2, V«() achieves its minimum

value. In oth'er words, the minimum V( () occurs when the ~ are propor-
tional to ~(jj.
58 CHAPTER 3

In power system reliability evaluation, high load level points in the

annual load curve make greater contributions to unreliability indices than
low load level points. The idea of stratified sampling therefore can be used
when considering the annual load curve.

3.5.5. Antithetic Variates

This technique is based on the following concept. (14) If there are two
unbiased estimators (}1 and (}2, then
() = !( (}1 + (}2) (3.54)
is still an unbiased estimator. The variance of () is
!
V( ()) = ! V( (}1) + ! V( (}2) + cov( (}1, (}2) (3.55)
If cov( (}1, (}2) is strongly negative, the estimator (3.54) can lead to reduction
of the variance.

f
For example, the integral given in equation (3.45) can be calculated by

I=! [g(x)+g(l-x)]dx (3.56)

Therefore, it can be estimated by

() = !E[g( U) + g(l- U)]
1 N
=- L [g(Xi)+g(l-Xi)] (3.57)
2N i =1
where U is a uniformly distributed random number sequence between [0, 1].
Since U and 1- U are strongly negatively correlated, the estimator (3.57)
has smaller variance than the estimator (3.46). Note that the computing
time required by equation (3.57) is twice that required by equation (3.46).
Therefore, the estimator (3.57) is more effective than the estimator (3.46)
only if the variance corresponding to the former is smaller than half of that
corresponding to the latter. This can be guaranteed if g(x) is a monotonic
function.

3.5.6. Dagger Sampling

This technique is suited to two-state variables and small probability
eventsYS) In power system reliability evaluation, system component states
ELEMENTS OF MONTE CARLO METHODS 59

remainder part

oI
0.151
I
0.30
I
0.45
I
0.60
I
0.75
I
0.90\1.0
I I

trial 1 2 3 4 5 6
Figure 3.8. Explanation of dagger sampling.

can be simulated by a two-state (up and down) variable and system compo-
nent failure events are usually small probability events. This technique is
discussed from a reliability evaluation point of view.
Suppose that a system component has failure probability p. Generate
a uniformly distributed random number Ubetween [0, I]. If U~p, the com-
ponent is in the failure state, and if U> p, it is in the normal state. Each
random number therefore corresponds to one trial of the component state.
This form of sampling is called direct Monte Carlo sampling.
Let S be the largest integer not larger than lip. The interval between
[0, 1] is divided into S equal subintervals and the length of each subinterval
is p. For example, the failure probability of a component p=0.15, IIp=
6.66, and S = 6. In this case, as shown in Figure 3.8, there are six subintervals,
the length of each subinterval is 0.15, and there is also a remainder part. In
dagger sampling, each subinterval corresponds to a trial of the component.
If the generated random number falls in subinterval i, the component failure
is assumed to occur in trial i and not to occur in other trials. In this example,
there are six trials associated with one random number. If the random num-
ber falls in the second subinterval, the component failure occurs in trial 2
but not in the other five trials. If the random number falls in the remainder
part beyond all the subintervals, then the component failure does not take
place in any of the six trials. In this sampling procedure, only one uniform
random number determines S trials of the component state. It is as if one
uniform random number pierces in S subintervals and therefore this method
is called "dagger sampling."
Combination of all component states provides a system state vector.
Let Z, , Z2, ... , ZN be N system state vectors obtained by dagger sampling.
The system unavailability can be estimated by
1 N
Q=- L X(Zj) (3.58)
N j =,
where X is an indicator variable:
X(Zj) =0 if Zj is a nonfailure state vector
X(Zj) = 1 if Zj is a failure state vector
60 CHAPTER 3

Generally, the variance of Q can be calculated by

V(Q) = ~2 {;tl V[X(Z;)] + ;~j cov[X(Z;), X(Zj)]} (3.59)

In the case of direct Monte Carlo sampling, there is no correlation

between different random vectors. The variance of the direct Monte Carlo
sampling is the first term of equation (3.59), that is,
1 N
V(QD)=2 I V[X(Z;)] (3.60)
N ;=1

Here, the subscript D denotes direct Monte Carlo sampling.

In dagger sampling, different random vectors are correlated. If a compo-
nent failure occurs in a trial, then it definitely does not occur in all other
trials corresponding to the common random number. If Zi/c and Zjk are two
elements associated with component k in the system state vectors Z; and Zj>
respectively, and the trials i and j are generated by a common random
number for component k, then they have a negative covariance:
cov(Zi/c, Zjk) =E(Zi/cZjk) - E(Zi/c)E(Zjk) = -PZ<O (3.61)

This indicates that the correlation between two system state random vectors
is negative as long as these two vectors have some elements corresponding
to a common random number. Because variable X(Z;) is an indicator vari-
able, the negative correlation between Z; and Zj also applies to X(Z;) and
X(Zj). Therefore, the dagger sampling has smaller variance than direct
Monte Carlo sampling.

3.6. THREE SIMULATION APPROACHES IN

RELIABILITY EVALUATION
The basic principles of three simulation approaches(I6-20) are described
in this section. Further discussion and their application to power system
reliability evaluation are developed in the following chapters.

3.6.1. State Sampling Approach

A system state depends on the combination of all component states and
each component state can be determined by sampling the probability that
the component appears in that state. (21,22)
ELEMENTS OF MONTE CARLO METHODS 61

The behavior of each component can be described by a uniform distri-

bution between [0,1]. Assume that each component has two states offailure
and success and that component failures are independent events. Let Si
denote the state of the ith component and PFi denotes its failure probability.
Draw a random number Ui distributed uniformly between [0, 1] for the ith
component,

s.={01
I
(success state)
(failure state)
if Ui~PFi
ifO::; Ui<PF i
(3.62)

The state of the system containing m components is expressed by the

vector S,
(3.63)

Assuming that each system state has the probability P(S) and the
reliability index function F(S), the mathematical expectation of the index
function of all system states is given by
E(F) = I F(S)P(S) (3.64)
SeG

where G is the set of system states.

Substituting the sampling frequency of the state S for its probability
P(S) gives

E(F) = I F(S) n(S) (3.65)

SeG N
where N is the number of samples and n(S) is the number of occurrences
of state S; F(S) can be obtained by appropriate system analysis. The advan-
tages of the state sampling approach are:
I. Sampling is relatively simple. It is only necessary to generate
uniformly distributed random numbers between [0, I]. It is not neces-
sary to sample a distribution function.
2. Required basic reliability data are relatively few. Only the compo-
nent-state probabilities are required.
3. The idea of state sampling not only applies to component failure
events but also can be easily generalized to sample states of other
parameters in power system reliability evaluation, such as load,
hydrological, and weather states, etc.

The disadvantage of this approach is that it cannot be used by itself to

calculate the actual frequency index.
62 CHAPTER 3

3.6.2. State Duration Sampling Approach

The state duration sampling approach is based on sampling the prob-
ability distribution of the component state duration. In this approach, chron-
ological component state transition processes for all components are first
simulated by sampling. The chronological system state transition process is
then created by combination of the chronological component state transition
processes. (23,24)
This approach uses the component state duration distribution functions.
In a two-state component representation, these are the operating and repair
state duration distribution functions and are usually assumed to be exponen-
tial. Other distributions, however, can be easily used.
The state duration sampling approach can be summed up in the follow-
ing steps:
Step 1: Specify the initial state of each component. Generally, it is
assumed that all components are initially in the success or up
state.
Step 2: Sample the duration of each component residing in its present
state. For example, given an exponential distribution, the sam-
pling value of the state duration is

1
T./=--ln Ur
Ai I

where Ui is a uniformly distributed random number between

[0, 1] corresponding to the ith component; if the present state
is the up state, Ai is the failure rate of the ith component; if
the present state is the down state, Ai is the repair rate of the
ith component.
Step 3: Repeat Step 2 in the given time span (yr) and record sampling
values of each state duration for all components. Chronological
component state transition processes in the given time span for
each component can be obtained and have the forms shown in
Figure 3.9.
Step 4: The chronological system state transition process can be
obtained by combining the chronological component state tran-
sition processes of all components. The chronological system
state transition process for the two components is shown in
Figure 3.10.
Step 5: Conduct system analysis for each different system state to
obtain the reliability index F(S) and use the concept given in
equation (3.64) to calculate E(F).
ELEMENTS OF MONTE CARLO METHODS 63

up
r:-
I
- component 1
I
I
I
I
down
time

up f-- ····· ····· component 2

down
T time
Figure 3.9. Chronological component state transition process.

1 up 2 up - -

1 down 2 up -
1 up 2 down

1 down 2 down L -_ _ _ _ _ _ _ _ _ _--L~_ _ _ _ __

time
Figure 3.10. Chronological system state transition process.

The advantages of the state duration sampling approach are:

1. It can be easily used to calculate the actual frequency index.
2. Any state duration distribution can be easily considered.
3. The statistical probability distributions of the reliability indices can
be calculated in addition to their expected values.
The disadvantages of this approach are:
1. Compared to the state sampling approach, it requires more comput-
ing time and storage because it is necessary to generate a random
64 CHAPTER 3

variate following a given distribution for each component and store

information on chronological component state transition processes
of all components in a long time span.
2. This approach requires parameters associated with all component
state duration distributions. Even under a simple exponential
assumption, these are all transition rates between states of each com-
ponent. In some case, especially for a multistate component represen-
tation, it might be quite difficult to provide all these data in an actual
system application.

3.6.3. System State Transition Sampling

Approach
This approach focuses on state transition of the whole system instead
of component states or component state transition processes. (21.25)
Assume that a system contains m components and that the state dura-
tion of each component follows an exponential distribution. The system can
experience a system state transition sequence {S(1), ... , sen)} = G where G
is the system state space. Let us suppose that the present system state is S(k)
and the transition rate of each component relating to S(k) is Ai (i= 1, ... ,m).
The state duration T/ of the ith component corresponding to system state
S(k) therefore has the probability density function: j;( I) = Ai exp( - Ail). Tran-
sition of the system state depends randomly on the state duration of the
component which departs earliest from its present state, i.e., the duration T
of the system state S(k) is a random variable which can be expressed by
T=min{Ti } (3.66)
i

It can be proved that since the state duration of each component Ti follows
an exponential distribution with parameter A;, the random variable T also
follows an exponential distribution with the parameter A=L~=I Ai, i.e., T
has the probability density function

f( I) = i~1 At exp( - i~1 Att)

Assume that system state S(k) starts at instant 0 and the transition of
the system state from S (k) to S (k+ I) takes place at instant 10. The probability
that this transition is caused by departure of the jth component from its
present state is the following conditional probability: Pj =P(1j= lo/T= 10)'
According to the definition of conditional probability and equation (3.66),
ELEMENTS OF MONTE CARLO METHODS 65

it follows that
Pj = P(1j= to/T= to)
=P(1j= to ('\ T= to)/P(T= to)
= P[1j= to ('\ (Ti~ to, i= 1, ... ,m)]/P(T= to)
m
= P(1j= to) I1 P(TI~ to)/ P(T= to) (3.67)
i=l,i;ej

Since Ti (i= 1, ... , m) and T follow exponential distributions,

P(Ti~tO)= foo Aie-.t,t dt=e-.t,to (3.68)

to
P(1j=to)= lim (Aje-A}tO)~t (3.69)
dt-+O

(3.70)

Substituting equations (3.68), (3.69), and (3.70) into equation (3.67) yields

Pj=P(1j=to/T=to)=A;!r Ai (3.71)
i=1

State transition of any component in the system may lead to system

state transition. Consequently, starting from state S(k), the system containing
m components has m possible reached states. The probability that the system
reaches one of these possible states is expressed by equation (3.71) and
obviously

(3.72)

Therefore, the next system state can be determined by the following

simple sampling. The probabilities of m possible reached states are success-
ively placed in the interval [0,1] as shown in Figure 3.11. Generate a

u
t Pm 1.0
o ~

Figure 3.11. Explanation of system state transition sampling.

66 CHAPTER 3

uniformly distributed random number U between [0, 1]. If U falls into the
segment corresponding to Pj , this means that transition of the jth component
leads to the next system state. A long system state transition sequence can
be obtained by a number of samples and the reliability of each system state
can be evaluated.
The advantages of the system state transition sampling approach are:
1. It can be used to calculate the exact frequency index without the
need to sample the distribution function and storing chronological
information as in the state duration sampling approach.
2. In the state sampling approach, m random numbers are required to
obtain a system state for an m-component system. This approach
requires only a random number to produce a system state.
The disadvantage of this approach is that it only applies to exponentially
distributed component state durations. It should be noted, however, that the
exponential distribution is the most commonly used distribution in reliability
evaluation.

3.7. EVALUATING SYSTEM RELIABILITY BY

MONTE CARLO SIMULATION
In order to illustrate the utilization of Monte Carlo techniques in
reliability evaluation, two simple examples are given in this section. (26,27)

3.7.1. Example 1
Two independent repairable components, A and B, operate in parallel.
System operation requires at least one component in service. The state dura-
tions of both A and B follow exponential distributions. Their failure rates
are denoted by AA and AB and repair rates by JJA and JJB respectively.

(a) Calculate the System Reliability Rs for a Mission Time

of T. The reliability of the system without considering repair can be
obtained using the following analytical expression:
(3.73)
If repairs are taken into consideration, the problem is relatively complex to
solve by an analytical method. It can be handled easily, however, by Monte
Carlo simulation.
ELEMENTS OF MONTE CARLO METHODS 67

The reliability of the system can be estimated by observing the results

of random experiments. An experiment is designed as a system behavior
before time T. The first step is to simulate the transition processes of the
two components as stated in Section 3.6.2. If there is no overlapping repair
in the time interval [0, T], the experiment is said to be a "success" (see
Figure 3.9); otherwise, it is a "failure." By defining the following indicator
variables:

x.={O
, 1
if the ith experiment is a "success"
if the ith experiment is a "failure"
the reliability of the system for a mission time of T can be obtained by
IN\X;
R =1--'=-- (3.74)
s N

where N is the number of experiments.

Consider the following data: AA =0.001 failures/hr, AB=0.0024
failures/hr, J.l.A = 0.003 repairs/hr, J.l.B = 0.005 repairs/hr and T= 1000 hr. In
the case of no repair being considered, the analytical solution using equation
(3.73) is Rs =0.425224 and the estimated value by Monte Carlo simulation
(106 samples) is Rs = 0.425057. In the case where repairs of the two compo-
nents are considered, the estimated reliability of the system is Rs = 0.61167.
Figure 3.12 shows the Monte Carlo convergence process as the number of
experiments, N, increases.

(b) Calculate the System Availability As. The availability of the

system can be calculated analytically using

(3.75)
As=l-CA:J.l.JCB:J.l.J

The availability of the system using Monte Carlo simulation can be estimated
by observing the chronological system state transition process over a
sufficiently long time as shown in Figure 3.10. The value of As can be calcula-
ted using
A = Total system up time
(3.76)
s Total simulation time
Using the data of AA, AB, J.l.A, and J.l.B given earlier, the analytical value for
the system availability by equation (3.75) is As =0.918919 and the estimated
value by Monte Carlo simulation with the total simulation time of 106 hr is
As=O.915562. Figure 3.13 shows the Monte Carlo convergence process.
68 CHAPTER 3

0.620
0.615
0.610
0.605

~ 0.600
:is 0.595 Component A:
.~ failure rate AA = 0.001 flhr
a; repair rate Po A .. 0.003 rlhr
a: 0.590
Component B:
0.585 failure rate AS = 0.0024 flhr
repair rate 11 B - O.OOSO rlhr
0.580
Mission time T = 1000 hours
0.575
0.570
0 1 2 3 4 5 6 7 8 9 10
Number of experiments N (x1 04 )
Figure 3.12. Reliability of the two repairable component parallel system vs. the number of
experiments.

1.00
Component A:
0.98 failure rate AA = 0.001 flhr
repair rate 11 A =0.003 rlhr
0.96 Component B:
failure rate AB .. 0.0024 flhr
repair rate 11 B = O.OOSO rlhr
~ 0.94
:g
~
·iii
0.92
~
0.90

0.88

0.86
0 1 2 3 4 5 6 7 8 9 10
Simulation time (x1 05 hours)
Figure 3.13. Availability of the two repairable component parallel system vs. simulation
time.
ELEMENTS OF MONTE CARLO METHODS 69

Figure 3.14. A two repairable component standby system.

3.7.2. Example 2
A standby system which contains two independent repairable compo-
nents, A and B, is shown in Figure 3.14. When the normally operating
component A fails, component B is switched in and component A is placed
in the standby mode after being repaired. The switch is assumed to be fully
reliable and switching time is negligible. It is assumed that a component
cannot fail in a standby mode. The state durations of the two components
follow an exponential distribution.

(a) Calculate the System Reliability Rs for a Mission Time

of T. The analytical method to solve this problem involves the solution of
Markov differential equations using a numerical algorithm. The Monte
Carlo technique can be easily applied by considering a series of experiments
which simulate the system behavior before time T. As shown in Figure 3.15,
the sequential time to failure (TIF) chain is generated first starting with

,, ,, TTF time chain

I, ,I
,
,,, ,, ,,,
I
',,
! TTRA1 ! , ,
!'TTRBf-+'

TTR time chain

, ,, ,i,
,,
,, ,,, ,,, ,, ,,
,, ,, ,,, ,,, ,,
,, ,,
,,, ,, ,, ,, ,,
,, ,, ,, ,, ,
TIME T
Figure 3.15. Simulation of the TIF and the TIR time chains.
70 CHAPTER 3

0.76
Component A:
0.75 failure rate AA • 0.001 flhr
repair rate 11 A .. 0.003 rlhr
Component B:
0.74 failure rate AB = 0.0024 flhr
l: repair rate 11 B .. 0.0050 rlhr
:g Mission time T = 1000 hours
.~ 0.73
a;
a:
0.72

0.71

0.70
0 1 2 3 4 5 6 7 8 9 10
Number of experiments N (X104)
Figure 3.16. Reliability of the two repairable component standby system vs. the number of
experiments.

component A. Another time chain associated with repair is also created

starting with component A and originates from the end of TTFAI. If the
time to repair (TIR) of a failed component is greater than the TIF of the
operating component, then that failed component cannot be repaired before
the other fails. Thus, the experiment is a "failure." Otherwise, it can be
considered a "success." Similarly, by defining indicator variables Xi as in
the first example, system reliability can be estimated using equation (3.74).
Using the same data for AA, AB, J.l.A, J.l.B, and T as given in the first
example, the estimated value of the system reliability with repairs
(105 samples) is Rs = 0.72399. The reliability of the system without repairs
is 0.56585. Figure 3.16 shows the Monte Carlo convergence process in this
case with repairs.

(b) Calculate the System Availability As. The Monte Carlo

procedure for estimating the availability of the system is the Same as that
discussed in the first example using equation (3.76). The calculation of
system down time, however, is different. As shown in Figure 3.17, the
operating time is first simulated. The component repair time is simulated
according to the operating time sequence. Whenever the repair time TIR of
the failed component is longer than the operating time TIF of the operating
ELEMENTS OF MONTE CARLO METHODS 71

system down time

..... TTFA1 -+1......- - TTFB1--.... ~ ~ TTF~ •

I
I
I
I

--+!-t•.rl
I
I
I
I I

l0III.1---- TTRA1
I •
I TTRB1

n===9
I
I
I
I
I
I
I
I
I
I
I
I
I I I
I I I
I I I

TIME
Figure 3.17. Calculation of the system down time.

component, a random repair time of the operating component is drawn.

The system down time is then equal to the overlapping repair time.
With the same data given earlier, the system availability As for a total
simulation time of 106 hr is As = 0.9408385. Figure 3.18 shows the Monte
Carlo convergence process in this case. The analytical solution for the system
availability is As=O.943109.
It can be seen from the above two examples that Monte Carlo simula-
tion provides considerable flexibility in evaluating system reliability. Success-
ful application relies on a complete understanding of the problem.

3.S. CONCLUSIONS
This chapter describes the basic elements of Monte Carlo simulation.
In general, random number/variate generation and variance reduction
techniques are the most basic and important aspects of the Monte Carlo
simulation method. These elements are discussed in Sections 3.3, 3.4, and
3.5. This material is not intended to be an exhausive description of these
concepts but to provide a sufficient introduction. Readers can find more
material in specialized texts and papers on Monte Carlo simulation. Monte
Carlo methods are also discussed from a reliability application point of
view in Sections 3.2 and 3.6. These considerations include convergence
72 CHAPTER 3

1.00

Analytical answer = 0.943109

0.95

:g~ 0.90

1i Component k.
~ 0.85 failure rate 1A • 0.001 flhr
repair rate Jl A .. 0.003 rlhr
Component B:
0.80 failure rate 18 • 0.0024 flhr
repair rata "8 • 0.0050 rlhr

0.7S 0
.f................1............2
.........-ro-.-
3...............4,..................
S.............,S............7"T"""""'...........8.............9T""""""......,10

Simulation time (x10 5 hours)

Figure 3.18. Availability of the two repairable component standby system vs. simulation
time.

characteristics and criteria and three basic sampling approaches for reli-
ability evaluation. The material in this chapter provides the theoretical fun-
damentals required to apply Monte Carlo methods to reliability assessment.
There is no unified mode or procedure for any given problem and two simple
illustrative examples for general application are given in Section 3.7.

3.9. REFERENCES
1. R. Y. Rubinstein, Simulation and the Monte Carlo Method, Wiley, New York, 1981.
2. D. H. Lehmer, "Mathematical Methods in Large-Scale Computing Units," Ann. Comp.
Lab., Harvard Univ., 1951, No. 26, pp. 141-146.
3. B. J. T. Morgan, Computer Simulation and Monte Carlo Method, Chapman and Hail,
London, 1984.
4. Fang Zaigen, Computer Simulation and Monte Carlo Method, Publishing House of Beijing
IndustriaI Institute, 1988.
5. M. Greenberger, "Notes in a New Pseudo-Random Number Generator," J. Assoc. Compo
Math., No.8, 1961, pp. 163-167.
6. D. E. Knuth, The Art of Computer Programming: Seminumerical Algorithms, Vol. 2,
Addison-Wesley, Massachusetts, 1969.
7. M. Greenberger, "An a Priori Determination ofSeriaI Correlation in Computer Generated
Random Numbers," Math. Comp., No. 15, 1961, pp. 383-389.
ElEMENTS OF MONTE CARLO METHODS 73

8. A. M. Law and W. D. Kelton, Simulation Modelling and Analysis, McGraw-Hili, New

York,1982.
9. M. Abramovitz and 1. A. Stegun, Handbook of Mathematical Functions, National Bureau
of Standards, 1968.
10. G. E. P. Box and M. E. Miiller, "A Note on the Generation of Random Normal Deviates,"
Ann. Math. Stat., No. 29, 1958, pp. 610-611.
11. J. M. Hammersley and D. C. Handscomb, Monte Carlo Methods, Wiley, New York, 1964.
12. A. W. Marshall, "The Use of Multi-Stage Sampling Schemes in Monte Carlo Computa-
tions," in Symposium on Monte Carlo Methods, Wiley, New York, 1956, pp. 123-14.
13. W. S. Cochran, Sampling Techniques, Wiley, New York, 1966.
14. J. M. Hammersley and K. W. Morton, "A New Monte Carlo Technique: Antithetic Vari-
ates," Proc. Cambridge Phi/os. Soc., No. 52, 1956, pp.449-474.
15. H. Kumamoto, K. Tanaka, K. Inoue, and E. J. Henley, "Dagger-Sampling Monte Carlo
for System Unavailability Evaluation," IEEE Trans. on Reliability, R-29, No.2, 1980.
16. G. Fishman, Principles of Discrete Event Simulation, J. Wiley, New York, 1978.
17. L. T. Blank, Statistical Procedures for Engineering, Management and Science, McGraw-
Hill, New York, 1980.
18. A. M. Mood, F. A. Graybill, and D. C. Boes, Introduction to the Theory of Statistics,
McGraw-Hili, New York, 1974.
19. S. J. Yakowitz, Computational Probability and Simulation, Addison-Wesley, Massachusetts,
1977.
20. J. Banks and J. S. Carson, Discrete Event System Simulation, Prentice-Hall, Englewood
Cliffs, NJ, 1984.
21. E. J. Henley and H. Kumamoto, Probabilistic Risk Assessment: Reliability Engineering,
Design, and Analysis, IEEE Press, New York, 1992.
22. R. Billinton and W. Li, "Hybrid Approach for Reliability Evaluation of Composite Genera-
tion and Transmission Systems Using Monte Carlo Simulation and Enumeration
Technique," lEE Proc. C, Vol. 138, No.3, 1991, pp. 233-241.
23. J. Endrenyi, Reliability Modeling in Electric Power Systems, Wiley, New York, 1978.
24. J. Kleijnen, Statistical Techniques in Simulation: Part I and II, Marcel Dekker, New York,
1975.
25. R. Billinton and W. Li, "A System Transition Sampling Method for Composite System
Reliability Evaluation," IEEE Trans. on PS, Vol. 8, No.3, 1993, pp. 761-770.
26. L. Gan, "Multi-Area Generation System Adequacy Assessment by Monte Carlo Simula-
tion," M.Sc. Thesis, University of Saskatchewan, 1991.
27. R. Billinton and R. N. Allan, Reliability Evaluation of Engineering Systems: Concepts and
Techniques (2nd edition), Plenum, New York, 1992.
 4
Generating System
Adequacy
Assessment

4.1. INTRODUCTION
Generating system adequacy assessment (HLl) is used to evaluate the
ability of the system generating capacity to satisfy the total system load.
This assessment can be conducted using either an analytical technique or the
Monte Carlo method. A considerable number of papers involving analytical
techniques are listed in ReferencesY-7) This chapter does not attempt to
reiterate these analytical techniques but focuses an application of the Monte
Carlo method in generating system adequacy assessment. Monte Carlo
simulation can be considered to be more preferable than an analytical
approach in situations which involve, for example, the following
considerations:

1. Time-dependent or chronological issues are considered.

2. The duty cycle of peaking units are modeled.
3. Nonexponential component state duration distributions are
considered.
4. Distributions of reliability indices are required.
5. A large unacceptable set of states (unfeasible range) in multi-area
generation system studies is involved.

Generating system adequacy assessment is normally divided into two

aspects: single-area and multi-area generating systems. Modeling a single-
area generating system provides the basis for modeling multi-area systems.
Two fundamental simulation methods-state duration sampling and system
75
76 CHAPTER 4

state sampling-are described by application to single-area generating sys-

tems in Sections 4.2 and 4.3.
Most electric power companies operate as members of an intercon-
nected power system because of the mutual benefits associated with intercon-
nected operation and planning. Multi-area generating system adequacy
assessment is therefore as important but also more complex than a single
area analysis. It involves not only generating capacity models and load
models of each area, but also tie line models and supporting policies between
areas. A maximum flow algorithm and a linear programming model for
multi-area generating system adequacy assessment are presented in Sections
4.4 and 4.5, respectively. A wide variety of supporting policies and their
effects on multi-area system reliability indices are discussed in Section 4.6.

4.2. SINGLE-AREA GENERATING SYSTEM

ADEQUACY ASSESSMENT-STATE
DURATION SAMPLING METHOD
As described in Section 3.6, there are three basic simulation methods
in reliability evaluation. State duration sampling and system state transition
sampling are often called sequential methods because they advance time or
system states sequentially. Correspondingly, system state sampling can be
called a nonsequential method because it considers each time point or system
state independent of another. In this section, the state duration sampling
method is applied to single-area generating system adequacy assessment.
The main advantages of this method are as follows:
1. Frequency indices can be easily calculated.
2. Nonexponential distributions of generating unit state duration can
be considered.
3. Peaking unit operating cycle can be easily modeled.

4.2.1. General Steps

The state duration sampling method for single-area generating system
adequacy assessment can be illustrated as follows. (8)
The first step is to generate operating histories for each generating unit
by drawing sample values of TIF (Time-to-Failure) and TTR (Time-to-
Repair) of the unit. The operating history of each unit is in the form of
chronological up-down-up or up-derated-down-up operating cycles. The sys-
GENERATING SYSTEM ADEQUACY ASSESSMENT 77

Unit #1

II II
Unit #2

II II
§ System Capacity

TIME Study Period

Figure 4.1. Available capacity models of each unit and the system.

tem available capacity can then be obtained by combining the operating

cycles of all units. Figure 4.1 shows the combination.
The second step is to superimpose the system available capacity curve
on the chronological hourly load curve to obtain the system available margin
model. A positive margin denotes that the system generation is sufficient to
meet the system load, while a negative margin implies that the system load
has to be curtailed. Figure 4.2 shows the superimposition.
The third step is to calculate appropriate reliability indices. In each
sampled year, for example in year i, the loss of load duration (LLDi) in hr,
the loss of load occurrence (LLOi), and the energy not supplied (ENSi) in
MWh can be obtained by observing the available margin model. The reli-
ability indices in N sampling years therefore can be estimated using the
following equations:

(l) Loss of Load Expectation (LOLE), hr/yr

I~ 1 LLDi
LOLE - ' = - - (4.1)
N
(2) Loss of Energy Expectation (LOEE), MWh/yr
~~lENSi
LOEE =_L.,,::..-.=-"---_ (4.2)
N
78 CHAPTER 4

system available capacity

hourly system load

TIME (hours)
Figure 4.2. Superimposition of the system available capacity model on the load model.

(3) Loss of Load Frequency (LOLF) occ./yr

IN I LLO;
LOLF=-'=-- (4.3)
N
The mean value obtained by equations (4.1) to (4.3) are expectation
estimates of the index over a period of N sampling years. The variance of
the estimate of a reliability index can be obtained by

(1'2= 1
N(N-l)
I [X;-E(x)f
;=1
(4.4)

where E(X) denotes the estimated expectation of any index and X; is the
sample value of the index in year i.

4.2.2. Generating Unit Modeling

It can be seen from the above general steps that the most important
factor is generating unit modeling. Other work is nothing more than the
combination of the operating cycles of all units, superimposition of the
system available capacity curve on the chronological load curve, records of
observed values, and simple calculations. The generating unit model pro-
vides an artificial operating history of the unit in state duration sampling
simulation.
GENERATING SYSTEM ADEQUACY ASSESSMENT 79

1IMTIF
UNIT UP UNIT DOWN

1/MTIR
Figure 4.3. Two-state model for a base load unit.

Generating units can be divided into two types: base load units and
peaking units. Base load units have long operating cycles while peaking units
have short operating cycles. Both types of unit can be modeled using their
state space diagrams based upon state durations. The development of the
state space diagram of a unit requires an understanding of the physical and
logical operation of the unit.

(a) Two-State Model for Base Load Units. A conventional

two-state model for a base load unit is shown in Figure 4.3 in which both
the operating and repair times are exponentially distributed. In this figure,
MTTF is the mean time to failure and MTTR the mean time to repair.
Sampling values of the TIF (time to failure) and the TTR (time to repair)
can be obtained by drawing random variates following the exponential
distributions with parameters).= I/MTTF and J.l = I/MTTR, respectively,
i.e.,

TTF=-MTIFln U (4.5)

and
TTR = -MTTR In U' (4.6)

where U and U' are two uniformly distributed random number sequences
between [0, I].
An up-and-down cycle of a two-state unit can be generated starting
from an initial state by sampling values of the TIF and TTR, as shown in
Figure 4.4.

(b) Derated State Model for Base Load Units. A generating

unit can operate at a reduced-capacity state designated as a derated state.
The state space diagram of the three-state model of a unit is shown in Figure
4.5, where ).ij denotes the transition rate from state i to state j.
Assuming that the unit resides in the up state at the beginning of the
simulation, there are two possible states to follow: the derated state or the
down state. If the unit transits to the derated state, the sampling value of
80 CHAPTER 4

..---

. -TTF1----- TTR1 TTF2 1+

TIME (hour)
Figure 4.4. Up--down-up cycle of a two-state unit.

UNIT
UP
1

UNIT A.:32 UNIT

DERATED DOWN
2 3
Figure 4.5. Three-state model for a base load unit.

the up-state duration is given by

1
Tupl = --;-In UI (4.7)
1\,12

If the unit transits to the down state, the sampling value of the up-state
duration is given by
1
Tup2 = - -;-In U2 (4.8)
1\,\3

where UI and U2 are two uniformly distributed random numbers. The sam-
pling value of the up-state duration is given by
(4.9)
GENERATING SYSTEM ADEQUACY ASSESSMENT 81

Up

Derated

Down omp!ete Partial

repair repair
Down

TIME(hour)
Figure 4.6. Operating cycle of a three-state unit.

Equation (4.9) also indicates the next state of the unit. If Tup = T upl , the
next state is the derated state. If Tup = T up2 , the next state is the down state.
When the unit resides in the derated or the down state, similar sampling can
be conducted. A unit operating cycle can therefore be obtained by simulating
the state duration, as shown in Figure 4.6. This simulation technique can
be generalized to a multiderated-state case.

(c) Peaking Unit Model. The model of a base load unit is inad-
equate for modeling peaking units. This is due to the fact that when the unit
is forced out of service, it may not be needed by the system, and when it is
in the operating state, periods of service may be interrupted by reserve
shutdown. The IEEE Task Group on Models for Peaking Service Units
proposed the four-state model(9) shown in Figure 4.7.
:rhe parameters in Figure 4.7 are: T, average reserve shutdown time
between periods of need, excluding scheduled outage; S, average in-need
time per occasion of demand; m, average in-service time between occasions
of forced outage, excluding forced outage as a result of failure to start;
r, average repair time per forced outage occurrence; p., probability of a
starting failure reSUlting in inability to serve load during all or part of a
demand period; a repeated attempt to start during one demand period
should not be counted as more than one failure to start.
The simulation procedure for peaking units can be explained as follows:
A system available capacity margin model, as shown in Figure 4.2, without
82 CHAPTER 4

( 1- Ps)fT
RESERVE UNIT IN SERVICE
SHUTDOWN
1/S

1/r PsrT 1/r 11m

1fT
FORCED OUT BUT FORCED OUT IN
NOT NEEDED PERIOD OF NEED
1/S

Figure 4.7. Four-state model for a peaking unit.

considering the peaking units, can be obtained first. The starting instants
and in-need times (corresponding to S) requiring peaking units can be deter-
mined from this model (see the shaded parts). Peaking units can then be
incorporated to modify the system available capacity margin model. When-
ever a peaking unit is needed, a uniformly distributed random number
between [0, I] is generated. If this random number is smaller than p., then
the unit fails to start up. Otherwise, it starts. If the unit starts, a sampling
value of time to failure corresponding to m is drawn. If the unit becomes
unnecessary before the failure event occurs (i.e., the in-need time is less than
the time to failure), this peaking unit is in service only during the in-need
time. If the unit fails to start during the in-need time, a sampling value of
time to repair corresponding to r is drawn. If the unit becomes unnecessary
before the repair event occurs, this peaking unit is out of service during the
whole or part of the in-need time. When the unit is in the forced out but
not needed state, the reserve shutdown time (the duration between two
starting instants requiring peaking units) is compared with the repair com-
pletion time and the most imminent event is applied. This procedure applies
to all starting instants requiring a peaking unit. The start-up duration is
assumed to be so short that it is negligible. If the system has more than one
peaking unit, a "one-by-one" policy is used.

(d) Nonexponential Distributions of State Durations. The

general steps are the same as described earlier when nonexponential distribu-
GENERATING SYSTEM ADEQUACY ASSESSMENT 83

tions for the state duration are considered. The unique difference is that
sampling values of the state durations are obtained by drawing random
variates following the given distribution. For example, the generating unit
repair times can be modeled by log-normal distributions. In this case, not
only the mean value of the repair time of a unit but also its standard devia-
tion should be specified. The sampling values of the repair time can be
calculated using the procedures given in Section 3.4.6(a) instead of equation
(4.6). Similarly, if the operating time or repair time is modeled by other
distributions, the sampling values of the state duration can be obtained using
the procedures corresponding to the specific distribution, which are given
in Sections 3.4.5 and 3.4.6.

4.2.3. Stopping Rules

Monte Carlo simulation is a fluctuating convergence process. As the
simulation proceeds, the estimated indices will approach their "real" values.
The simulation should be terminated when the estimated reliability indices
achieve a specified degree of confidence. The purpose of a stopping rule is
to provide a compromise between the accuracy needed and the computation
cost.
As noted in Section 3.2.4, the coefficient of variation is often used
as the convergence criterion in Monte Carlo simulation. The coefficient of
variation of an index is defined as
a = a/E(X) (4.10)
where E(X) is the estimated expectation of the index and a the standard
deviation ofthe estimated expectation obtained from equation (4.4). A num-
ber of calculations indicate that the coefficient of variation for the LOEE
index has the lowest convergence speed compared to other indices. It is
therefore advisable when calculating multiple indices to use the LOEE
coefficient of variation as the convergence criterion.
Two stopping rules can be used:
1. The simulation stops when the coefficient of variation is less than
the prespecified tolerance value.
2. The simulation pauses at a given number of samples and it is checked
to see if the coefficient of variation is acceptable. If not, the number
of samples can be increased.

These two stopping rules are quite general. The basic idea presented
here is applied in all the other applications in this book.
84 CHAPTER 4

4.2.4. IEEE RTS Studies

The IEEE RTS has 32 generating units ranging from 12 MW to
400 MW. The annual load curve consists of 8736 hourly load points. All
the data for the IEEE R TS are given in Appendix A.l.

(a) Base Case. In this case, the generating unit state durations are
assumed to be exponential and no derated states are considered.
The system available capacity is obtained by combining the operating
cycles of all the units. Figure 4.8 shows the system available capacity model
in a typical sample year. The annual load model for the IEEE RTS is shown
in Figure 4.9. Figure 4.10 shows the system available margin model in the
typical sample year.
The estimated reliability indices are given in Table 4.1. These results are
for 2500 sampling years. Figures 4.11, 4.12, and 4.13 show the convergence
processes of the LOLE, LOEE, and LOLF indices, respectively, as the num-
ber of sample years increases. It can be seen that of these indices, the LOEE
index has the most difficulty in converging and the LOLF index takes the
shortest time to stabilize.
The distribution of an index can be obtained from the simulated history
of the index. Figures 4.14, 4.15, and 4.16 show the distributions of LOLE,
LOEE, and LOLF indices, respectively, for a system peak load of 2850 MW.

3400

3200

-
-
3:
~ 3000
~
'0
as
0..
as 2800
0

2600

2400
0 1092 2184 3276 4368 5460 6552 7644 8736
Hour
Figure 4.8. System available capacity model in a typical sample year.
GENERATING SYSTEM ADEQUACY ASSESSMENT 85

2900
System peak load =2850 MW

2500

~
-
~
"C
as
.2
2100

>-
"I:
:::l
1700
0
J:

1300

900~~~~~~~~~~~~~~~~~~~
o 1092 2184 3276 4368 5460 6552 7644 8736

Hour
Figure 4.9. Annual hourly load model for the IEEE RTS.

2600

2200

- 1800

-
~
~ 1400
c:::
.~
as 1000
~
600

200

-200
0 1092 2184 3276 4368 5460 6552 7644 8736
Hour
Figure 4.10. System available margin model in a typical sample year.
86 CHAPTER 4

Table 4.1. IEEE RTS Reliability Indices

Annual system peak load (MW)
Index 2750 2850 2950 3050
LOLE (hr/yr) 4.8516 9.3716 17.3696 30.7172
LOEE (MWh/yr) 586.4907 1197.4448 2335.7295 4384.6909
LOLF (occ./yr) 1.0348 1.9192 3.4228 5.8652

14

12 Analytical answer:
LOLE = 9.39418 hours/yr

'C'
10
~ 8
-
;:,
0
.c
W
..J
6
9 4 System peak load =2850 MW
2

0
0 500 1000 1500 2000 2500
Sample years
Figure 4.11. LOLE vs. the number of sample years.

Note that these distributions are not probability densities. Instead, they are
in the number of years. The probability density can be obtained by dividing
the number of years by the total number of sample years (2500). The distri-
bution of each index provides interesting and important insight into the
random behavior of the system. For instance, it can be seen from Figure
4.14 that among the 2500 sample years, about 1560 years actually experience
loss of load for less than 2.5 hr (including 0 hr).

(b) Derated State Case. This case is the same as the base case,
except that the 350 MW and 400 MW units are assumed to be able to operate
at 50010 derated capacities. The unit derating data are given in Table 4.2.
These derating data are such that the derating-adjusted two-state model of
a unit is identical to that in the base case. (10) The simulation results for 2500
GENERATING SYSTEM ADEQUACY ASSESSMENT 87

2000

1750
System peak load =2850 MW
1500
'C'
~ 1250

-
3:
:E
w
w
0
1000

750 Analytical answer:

-I =
LOEE 1176 MWh/yr
500

250

0
0 500 1000 1500 2000 2500
Sample years

Figure 4.12. LOEE vs. the number of sample years.

3.0

2.5

'C' 2.0

f-
u. 1.5
-I
9 1.0
System peak load =2850 MW
0.5

0.0
0 500 1000 1500 2000 2500
Sample years
Figure 4.13. LOLF vs. the number of sample years.
88 CHAPTER 4

16

14 2500 sample years

Std. Dev. (0) = 16.2 hr/yr
L:" 12
>-
0
0 10
..-
C
>-
0 8
c:
CD
:+- MEAN = 9.3716 hr/yr
I
::J 6
CT
....
CD I
I
U. I
4 I
I
I
I

0
0 10 20 30 40 50 60 70 80 90 100

LOLE (hr/yr)

Figure 4.14. Distribution of the LOLE index.

18

16
2500 sample years
"§' 14 Std. Dev. (0) = 3100 MWh/yr
>-
0
0
12
..-
C 10
~
c: 8
CD I
::J
CT
6
: . - MEAN = 1197.4 MWh/yr
....
CD
U.
I
I
I
I
4 I
I
I
I
2 I
I

0
0 1600 3200 4800 6400 8000
LOEE (MWh/yr)

Figure 4.15. Distribution of the LOEE index.

GENERATING SYSTEM ADEQUACY ASSESSMENT 89

12

10 2500 sample years

Std. Dev. (0) = 2.65 occ/yr
~
>.
0
0 8
,...
C
>. 6
0
r::
Q)
::J I
I
C"
~ 4 : . - MEAN = 1.9192 occ/yr
u.

o~~~~~~~~~~~~~
o 234 5 6 7 8 91011121314151617181920
LOLF (occ/yr)

Figure 4.16. Distribution of the LOLF index.

sample years are shown in Table 4.3. It can be seen, by comparing the results
in Tables 4.1 and 4.3, that including the unit derated states provides a more
optimistic appraisal of generating system adequacy.

(e) Peaking Unit Case. This case is the same as the base case for
the annual peak load of2850 MW, except that additional 25-MW gas turbine

Table 4.2. Unit Derating Data for the IEEE RTS

Unit Derated
Mean duration (hr)
size capacity
(MW) (MW) Up Derated Down
350 175 1150 60 70
400 200 1100 100 100

Table 4.3. IEEE RTS Reliability Indices for the Derated State Case
Annual system peak load (MW)
Index 2750 2850 2950 3050
LOLE (hr/yr) 2.6860 5.5404 11.0608 21.0664
LOEE (MWh/yr) 290.4749 642.0654 1350.1066 2729.4775
LOLF (occ.jyr) 0.5952 1.2140 2.3748 4.3752
90 CHAPTER 4

Table 4.4. Peaking Unit Data for the IEEE RTS

Mean time to Average repair Starting
Capacity failure time failure
(MW) (hr) (hr) probability
25 550 75 0.01

Table 4.5. IEEE RTS Reliability Indices for the

Peaking Unit Case
Number of peaking units added
Index Zero units One unit Two units
LOLE (hr/yr) 9.3716 7.9108 6.6132
LOEE (MWh/yr) 1197.4448 990.0966 817.2875
LOLF (occ./yr) 1.9192 1.6208 1.3876

units are installed as peaking units. The data for the peaking units are given
in Table 4.4. The simulation results for 2500 sample years are shown in
Table 4.5. The addition of peaking units improves the generating system
adequacy.

(d) Nonexponential Distribution Case. This case is the same as

the base case for the annual peak load of 2850 MW, except that unit operat-
ing durations or repair durations are assumed to be nonexponential. The
following four conditions were considered:

1. Times to failure of all units follow a Weibull distribution with the

mean value = MTTF and the shape parameter p = 0.5.
2. Times to repair of all units follow a Weibull distribution with the
mean value=MTTR and the shape parameter P=4.0.
3. Times to repair of all units follow a normal distribution with the
standard deviation which equals one-third of the mean value
(MTTR).
4. Times to repair of all units follow a log-normal distribution with
the standard deviation which equals one-third of the mean value
(MTTR).

The simulation results for 2500 sample years are shown in Table 4.6.
It is interesting to note that the reliability indices of the generating system
in a mean value sense do not display great differences when the generating
unit state durations are assumed to follow different distributions.
GENERATING SYSTEM ADEQUACY ASSESSMENT 91

Table 4.6. IEEE RTS Reliability Indices Considering Nonexponential

Distributions of Generating Unit State Durations
Index Cond.l Cond.2 Cond.3 Cond.4
LOLE (hr/yr) 9.7388 9.2260 9.2407 9.1866
LOEE (MWh/yr) 1231.6710 1135.4628 11181.0088 1157.0930
LOLF (occ./yr) 1.9388 1.9096 1.8890 1.8883

4.3. SINGLE-AREA GENERATING SYSTEM

ADEQUACY ASSESSMENT-STATE
SAMPLING METHOD
This section describes the application of the state sampling method in
single-area generating system adequacy assessment. Compared to the state
duration sampling method, this method cannot be used to calculate the
frequency index in a strict sense. However, it has the following advantages:

1. It requires less computing time and memory storage than the state
duration sampling method, particularly for a large-scale system.
2. Basic data are the generating unit state probabilities and no transi-
tion rates between generating unit states nor other data are required.
It is easier for an electric power company to provide generating unit
state probabilities than transition rates between possible generating
unit states, particularly in the case of multistate generating unit
representations.

4.3.1. Single Load Level Case

For a single load level, generating system adequacy depends on the
possible generating capacity states of the system. A system state is a random
combination of all generating unit states. The behavior of each generating
unit can be simulated by a uniformly distributed random number sequence
in [0, 1]. In the case of a two-state representation, let Sj denote the state of
the ith generating unit and FUj be its forced unavailability. A uniformly
distributed random number Uj is drawn in [0,1] for the ith generating unit,

(up state) ifUj~FUj

S.={O (4.11)
I 1 (down state) ifO~Uj<FUj
92 CHAPTER 4

In situations in which generating unit derated states are considered, let PDRi
be the probability of a single derated state of the ith generating unit,

(up state) if Ui~PDRi+ FUi

(down state) ifPDRi~ Ui<PDR i + FUi (4.12)
(derated state) ifO~ Ui<PDR i

The above state sampling technique can be easily extended to the case of
multiderated states of generating units without any increase in computation
effort.
The available capacity of each generating unit can be determined
according to its state. The generating capacity state of the system can be
expressed using a generation capacity vector {G ik , i= I, ... ,m}, where Gik
is the available capacity of the ith generating unit in the kth sampling and
m is the number of generating units. For a given load level D, the demand
not supplied due to insufficient generating capacity in the kth sampling is
given by

DNSk = max{0, D - i~l Gik} (4.13)

The estimates of the annualized reliability indices for N samples can be

calculated using the following equations:

(1) Expected Demand Not Supplied (EDNS), MW

EDNS= k~l
IN DNS
k (4.14)
N

(2) Loss of Energy Expectation (LOEE), MWh/yr

I:-l DNSk x 8760

LOEE - - - - - - - (4.15)
N

(3) Loss of Load Expectation (LOLE), hr/yr

~N Ik(DNS k)
LOLE L...k~ 1 x 8760 (4.16)
N

where h is an indicator variable which means that

ifDNSk=O
ifDNSk:;cO
GENERATING SYSTEM ADEQUACY ASSESSMENT 93

The variance of the estimate of a reliability index can be calculated by

1 N
(52= I [Xk-E(X)f (4.17)
N(N-l) k=\

where E(X) denotes the expected estimate of any index and X k is the kth
sample value of the index. It should be noted that equations (4.4) and (4.17)
have the same form, but N in these two equations is different; N in equation
(4.4) is the number of sample years and N in equation (4.17) is the number
of samples.

4.3.2. Modeling an Annual Load Curve

The most basic approach to consider the annual load curve is to scan
all hourly points of the chronological load curve. This is, however, not
practical because of computational requirements. The following three alter-
natives can be used to model the annual load curve:
1. Sample load states in terms of the chronological load curve. Each
hourly load in one year has an equal occurrence probability. A uniformly
distributed random number therefore can be drawn to select a particular
load state. This load state and the drawn states of all generating units form
a sample of the system state.
2. Create a multistep model of the annual load curve. The chronological
load curve can be changed into the load duration curve in a load level
descending order. And the load duration curve can be approximated by a
multistep model as shown in Figure 4.17. The method given in the previous
subsection for a single load level can be used directly for each load level
step. Annual reliability indices can be obtained by weighting the annualized
indices for each load level by the load step probability. Assume that the
load duration curve is divided into NL level steps and the ith step includes
NI; load points. The probability of the ith load step is
NI·
p.=--' (4.18)
, 8760

The annual reliability indices considering the multistep model of the annual
load curve can be obtained using
NL
EDNS = I EDNS;P; (4.19)

NL
LOEE= I LOEEiPi (4.20)
;=\
94 CHAPTER 4

8760

Figure 4.17. A multistep model of the load duration curve.

NL
LOLE= L LOLEiPi (4.21)
;=1

where EDNS i , LOEEi , and LOLE i are the annualized indices for the ith
load level step.
3. Sample the load levels using the multistep model. The probabilities
of all steps Pi are put successively in the interval [0, 1]. Draw a uniformly
distributed random number in [0, 1]. A load level can be determined in each
sample according to the location of the random number as shown in Figure
4.18.
Method 2 is more effective than Method I or 3 from a computational
point of view. This is mainly due to the fact that load points in the flat
segments of a load duration curve provide almost the same contribution to
the total indices and that a number of the low load points, which may have

random number

!
~1--~~--~ff~~'------~------------~--------1~
o P1 Pi-1 Pi 1
Figure 4.18. Explanation of how to sample load levels.
GENERATING SYSTEM ADEQUACY ASSESSMENT 95

relatively high probabilities of occurrence, make little or no contribution to

the total indices. Method 2 requires enumeration of load level steps. Essen-
tially, however, it is still a Monte Carlo simulation method because sampling
of generating unit capacity states is conducted for each load level. It can be
seen as a variance reduction technique as it makes use of analytical load
level step probabilities.
Different load level steps make different contributions to the total
annual indices. It is therefore possible to use different tolerance values for
the coefficient of variation convergence criteria in the different load level
steps. This is similar to a stratified sampling approach. The low load level
steps, which make little or no contribution to the total indices, can often be
excluded from enumeration. The following discussion focuses on Method 2
and the numerical results given in Section 4.3.5 are based on this method.

4.3.3. Cluster Technique for a Multistep

Load Model
The cluster technique(II.12) can be used to create a multistep model of the
annual load duration curve. This technique is called the K-mean algorithm or
nearest centroid sorting. Assume that the load duration curve is divided into
NL load levels. This corresponds to grouping 8760 load points into NL
clusters. Each load level is the mean value of those load points in a cluster.
This analysis can be conducted in the following steps:
Step 1: Select initial cluster means Mi where i denotes cluster i (i =
1, ... ,NL).
Step 2: Calculate the distance Dki from each hourly load point Lk (k =
1, ... , 8760) to each cluster mean Mi by
(4.22)
Step 3: Group load points by assigning them to the nearest cluster and
calculate new cluster means by

M.=LkEICLk (4.23)
I NIi

where NIi is the number of load points in the ith cluster and
IC denotes the set of the load points in the ith cluster.
Step 4: Repeat Steps 2 and 3 until all cluster means remain unchanged
between iterations.
The obtained Mi and NIi are the load level (in MW) and the time length
(in hr) for the ith step in the multistep load model, respectively.
96 CHAPTER 4

It should be noted that, theoretically, the cluster technique is a local

optimal sorting approach and therefore may lead to multisolutions. It is
important to specify the proper number of clusters (steps) and initial cluster
means. This can be determined by testing for a given annual load curve.
Generally, selecting relatively high initial cluster means is beneficial.

4.3.4. Stopping Rules

The two general stopping rules in Monte Carlo simulation are given in
Section 4.2.3. In order to obtain the coefficient of variation expressed in
equation (4.10), the variance of the estimate of a reliability index has to be
first calculated. In the case of a single load level it can be estimated directly
by equation (4.17). In the case of a multistep model of the annual load
curve, the following equation similar to that used in stratified sampling can
be applied:
NL p2 N,
u 2= L j L [Xk -Ej (X)]2 (4.24)
j=1 N j (Nj -l) k=1

where NL is the number of load levels; Nt. Pj, and Ej(X) are the number
of samples, the probability, and the expected estimate of any index for the
ith load level, respectively; and Xk is the kth sample value of the index.
Equation (4.24) can also be rewritten as
NL
u2 = L plu; (4.25)
j=1

where u; is the variance of the estimate of the reliability index associated

with the ith load level.
Generally, the relatively high load level steps make greater contributions
to the total annual indices than to those at relatively low load levels. A
smaller coefficient of variation tolerance can be used as the stopping criterion
for the high load level steps and larger ones for the low load level steps.

4.3.5. IEEE RTS Studies

The 8736 hourly load points given in Appendix A.I were replaced by
a 20-step load level model obtained using the cluster technique. The 20-step
load level model is given in Table 4.7. The load levels are in per unit of the
annual peak load.
GENERATING SYSTEM ADEQUACY ASSESSMENT 97

Table 4.7. The 20-Step Load Level Model for the IEEE RTS
Load Load
Step level Step level
No. (p.u.) Probability No. (p.u.) Probability
1 0.9900 0.0006 11 0.6792 0.0711
2 0.9505 0.0034 12 0.6481 0.0738
3 0.9210 0.0061 13 0.6179 0.0754
4 0.8896 0.0l7l 14 0.5866 0.0630
5 0.8612 0.0236 15 0.5519 0.0695
6 0.8348 0.0371 16 0.5184 0.0805
7 0.8068 0.0482 17 0.4864 0.0949
8 0.7782 0.0499 18 0.4512 0.0769
9 0.7467 0.0517 19 0.4149 0.0690
10 0.7126 0.0590 20 0.3733 0.0292

(a) Base Case. In this case, no generating unit derated states are
considered. The stopping rule includes the following two aspects:
1. Two criteria are specified for each load level step. The maximum
number of samples is 80,000 and the LOEE index coefficient of
variation tolerance is 0.05. When either ofthem is reached, the simu-
lation in a particular load level step ends.
2. When the contribution of a load level step to the total LOEE
index is smaller than 3%, all load steps lower than this level are
ignored.
Tables 4.8 and 4.9 show the results for annual peak loads of 2850 MW
and 3050 MW, respectively. It can be seen that the LOLE and LOEE indices
obtained using the state sampling method are consistent with those obtained
using the state duration sampling method. The coefficients of variation for
the system LOEE are quite small (0.0227 and 0.0189, respectively), which
indicates that the calculated results have sufficient accuracy. The stopping
rule given earlier is reasonable. The relatively high load level steps, which
make major contributions to the total reliability indices, are associated with
the relatively small coefficients of variation. On the other hand, although
the relatively low load level steps have relatively large coefficients of varia-
tion, this does not materially affect the accuracy of the total system indices
as their contributions are quite small. For example, in the case of the
2850 MW annual peak load, the coefficient of variation for the LOEE in
Step 9 is quite high (0.2089). It has, however, no great impact on the accuracy
of the total system index because its contribution is only 2.27%. The
contributions due to Steps 10-20 are very small and therefore have been
neglected.
98 CHAPTER 4

Table 4.8. The Results for an Annual Peak

Load of 2850 MW
Coefficient of
LOLE LOEE variation for
(hr/yr) (MWh/yr) LOEE index
Index 9.2185 1147.3279 0.0227
Contributions of Each Load Step (p.u.)
Step 1 0.0545 0.0406 0.0438
Step 2 0.1586 0.1444 0.0487
Step 3 0.1556 0.1499 0.0482
Step 4 0.2183 0.2128 0.0490
Step 5 0.1565 0.1641 0.0578
Step 6 0.1223 0.1165 0.0776
Step 7 0.0720 0.0976 0.1038
Step 8 0.0466 0.0514 0.1388
Step 9 0.0156 0.0227 0.2089
Step 10-20 0.0000 0.0000

Table 4.9. The Results for an Annual Peak

Load of 3050 MW
Coefficient of
LOLE LOEE variation for
(hrjyr) (MWhjyr) LOEE index
Index 30.4516 4379.6152 0.0189
Contributions of Each Load Step (p.u.)
Step 1 0.0441 0.0437 0.0442
Step 2 0.1372 0.1103 0.0413
Step 3 0.1347 0.1208 0.0463
Step 4 0.2389 0.2155 0.0493
Step 5 0.1552 0.1701 0.0490
Step 6 0.1445 0.1525 0.0480
Step 7 0.0780 0.1084 0.0577
Step 8 0.0482 0.0494 0.0758
Step 9 0.0192 0.0293 0.1100
Step 10-20 0.0000 0.0000

(b) Derated State Case. In this case, the derated states of the 350-
MW and two 400-MW units are considered. The unit derating data are
given in Table 4.2. The state probabilities for these generating units were
obtained using their state space model and are shown in Table 4.10. The
stopping rule is the same as that in the base case. The calculated results are
given in Tables 4.11 and 4.12.
GENERATING SYSTEM ADEQUACY ASSESSMENT 99

Table 4.10. The State Probabilities of the

Two Three-State Generating Units
Unit State probability
size
(MW) up 50"10 derated down
350 0.84616 0.07692 0.07692
400 0.89843 0.04688 0.05469

Table 4.11. The Results Considering Derated

States for the Annual Peak Load of 2850 MW
Coefficient of
LOLE LOEE variation for
(hr/yr) (MWh/yr) LOEE index
Index 5.4591 613.7520 0.0263
Contributions of Each Load Step (p.u.)
Step 1 0.0704 0.0588 0.0440
Step 2 0.1840 0.1643 0.0467
Step 3 0.1728 0.1651 0.0490
Step 4 0.2304 0.2223 0.0532
Step 5 0.1489 0.1638 0.0764
Step 6 0.1073 0.1166 0.1052
Step 7 0.0582 0.0800 0.1502
Step 8 0.0280 0.0300 0.2626
Step 9-20 0.0000 0.0000

Table 4.12. The Results Considering Derated

States for the Annual Peak Load of 3050 MW
Coefficient of
LOLE LOEE variation for
(hr/yr) (MWh/yr) LOEE index
Index 21.2280 2741.4656 0.0196
Contributions of Each Load Step (p.u.)
Step 1 0.0573 0.0528 0.0472
Step 2 0.1655 0.1424 0.0443
Step 3 0.1543 0.1473 0.0448
Step 4 0.2432 0.2237 0.0490
Step 5 0.1461 0.1544 0.0498
Step 6 0.1227 0.1298 0.0495
Step 7 0.0658 0.0945 0.0751
Step 8 0.0311 0.0352 0.1140
Step 9 0.0140 0.0199 0.1695
Step 10-20 0.0000 0.0000
100 CHAPTER 4

4.4. MULTI-AREA GENERATING SYSTEM

ADEQUACY ASSESSMENT-
MAXIMUM FLOW ALGORITHM
The generating capacity adequacy of a power system can be improved
by interconnecting the system to other power utilities. When the total avail-
able capacity in an area is insufficient to meet its load, assistance can be
received from neighboring areas. The amount of assistance depends on the
following four factors:
1. The load level of the supported area
2. The available generating capacities of the supporting areas
3. The tie line constraints
4. The import/export agreement between the areas
Two different fundamental approaches to deal with assistance between
areas and tie line constraints are presented in this section and in the next
section, respectively. The problem of import/export agreements is discussed
in Section 4.6.

4.4.1. Basic Procedure

The multi-area generating system adequacy assessment approach
described in this section is based on a combination of the state duration
sampling method and the maximum flow algorithm. (13,14) It consists of the
following general steps:
1. Construct a generating capacity model and an appropriate load
model for each area.
2. Combine the generating capacity model and the load model to obtain
an available capacity margin model for each area.
3. Incorporate the tie line network and the supporting policy, and calcu-
late the reliability indices for each area and the total system.
Steps I and 2 have been clearly explained in Section 4.2. A tie line can
also be simulated using state duration sampling to obtain the up-down-up
operating cycles. Given an initial state, a chain of time to failure and time
to repair can be generated by random samplings, which are similar to the
simulation of a generating unit. The operating cycles of the tie lines and the
available capacities of the tie lines are then used to modify the available
capacity margin models for each area. This can be done using the maximum
flow algorithm described in the next subsection.
GENERATING SYSTEM ADEQUACY ASSESSMENT 101

The reference period, as in the case of a single system study, is 1 yr.

Each sample year is divided into 8760 hr and therefore the time unit is 1 hr.
The maximum flow algorithm is applied for all hourly points where at least
one area has a negative capacity margin. A loss of load situation is consid-
ered to arise in an area when the available assistance through interconnec-
tions cannot offset the capacity deficiency of the area. Consequently, the
available capacity margin model for each area can be modified. The reli-
ability indices in each sample year, that is, the loss of load duration (LLDi)
in hr, the loss ofload occurrence (LLOi), and the energy not supplied (ENS i )
in MWh for an area considering interconnections, can be obtained by observ-
ing the modified available capacity margin model of the area. The estimates
of the area reliability indices in N sample years can be calculated using
equations (4.1)-(4.3). The LOLE and LOEE indices reflecting the overall
system reliability can be calculated using
'\'~1 LLDsi
LOLE =_L...._-_ _ (4.26)
S N
and
,\,m '\'~ ENS.
LOEE 4=1 L....=1 k.
(4.27)
S N
where N is the number of sample years, m is the number of areas, LLDsi is
the total system state duration when one or more areas are in the loss of
load situation in sample year i, and ENSki is the energy not supplied of area
k in sample year i.

4.4.2. Modifying the Area Available Margin

Model Using the Maximum Flow
Algorithm
A multi-area generating system can be modeled by a transportation
network with stochastic branch capacities. (8,13,14) A tie line is represented by
a branch, which may be either unidirectional or bidirectional depending on
the allowable orientation of power transfer. Each area is represented by a
node. Two additional fictitious nodes are introduced: a generation source
node s and a load sink node t. Node s is assumed to have an infinite installed
capacity. A fictitious unidirectional branch from node s to each area node
is introduced to represent the maximum available area assistance. A fictitious
unidirectional branch from each area node to node t is created to represent
the required supported demand. For simplicity, all the branches can be
102 CHAPTER 4

(a) A three-area generating system

(b) The transportation network

Figure 4.19. A three-area generating system and its transportation network.

assumed to be bi-directional and therefore a pair of capacities (Cij, Cji) are

associated with each branch (i,j). For the unidirectional branches, Cij=O
or Cji = o. Capacity Cij is not necessarily equal to capacity Cji for bidirectional
branches. As an example, Figure 4.19 shows a three-area generating system
and its transportation network.
The capacity pair of each branch (i,j) (i,j=l=s, t) at a time point is either
(Cij, Cji) or (0,0), depending on the state of the corresponding tie line at
that time point. If tie line (i,j) is in an operating state, the capacity of
branch (i,j) is (Cij, Cji); otherwise, it is (0, 0). The capacity of each fictitious
branch (s, i) or (i, t) at a time point is determined by the available capacity
margin of area i (i =l=s, t) at the time point. Let di denote the available
capacity margin of area i, Csi and Cit are obtained by the following rules:

i ifdi>O
C.={d
0 (4.28)
SI
if di 5.0
and

i ifdi<O
Cit = {-d0 (4.29)
ifdi~O
GENERATING SYSTEM ADEQUACY ASSESSMENT 103

(a) (b)

(c) (d) (e)

Figure 4.20. Five multi-area systems.

After the transportation network is constructed, the maximum flow

algorithm, which is described in Appendix D.2, can be used to calculate
supporting power. For a supported area i, a failure occurs when
Xit<-d;

where Xit is the calculated supporting power for area i. The available capa-
city margin model of each area without considering interconnection can
be modified in terms of the results obtained by the maximum flow
algorithm.
It can be seen that the transportation network topology is fixed all the
time. The only changeable item is the branch capacity Cij, which is deter-
mined by the state of a tie line or the available capacity margin of an area.
At a time point in a sampling year, where the available capacities of one or
more area cannot meet the loads, the maximum flow algorithm is utilized
to find the maximum available assistance for the supported areas.

4.4.3. Case Stud ies

In order to demonstrate the application of the maximum flow algorithm
to multi-area generating system adequacy assessment, five multi-area con-
figurations created by connecting identical IEEE R TSs are considered. These
configurations are shown in Figure 4.20. Each area in Figure 4.20 is an
IEEE RTS with 32 generating units. The annual peak load is 2850 MW.
104 CHAPTER 4

Table 4.13. Reliability Indices for the

Two-Area System
LOLE LOEE LOLF
(hr/yr) (MWh/yr) (occ'/yr)
Area 1 4.534 542.696 0.9617
Area 2 4.542 535.751 0.9731
System 8.568 1078.457

Each tie line is represented by a two-state model with MTTF = 4380 hr and
MTTR = 10 hr. The capacity of each tie line is Cij= Cji = 100 MW.

(a) Two-Area System. The estimated reliability indices for the two-
area system are given in Table 4.13. The number of sample years is 10,000.
Figures 4.21, 4.22, and 4.23 show the area LOLE, LOEE, and LOLF indices
as functions of sample years, respectively. It can be observed from these
figures that the rough location at which the estimated mean value begins to
stabilize is about 2000 sample years. This can be compared to about 800
sample years in a single-area IEEE RTS study (see Figures 4.11, 4.12, and
4.13). The required simulation time for multi-area system analysis is nor-
mally longer than that required in single-area system studies. This is due to
the fact that a multi-area system is more reliable than a single-area system.

6.5
6.0 - - - AREA 1 1
1
.---AREA2.
5.5
"C
5.0
~
4.5
- 4.0
::J
0
.r.
W
...J
0
...J 3.5
r:I 100 MW r:l
3.0 2850 MW ~ ~ 2850 MW
MTTF =4380 hr
MTTR = 10 hr
2.5
2.0
0 1 2 3 4 5 6 7 8
Sample years (x1000)
Figure 4.21. Area LOLE vs. the number of sample years.
GENERATING SYSTEM ADEQUACY ASSESSMENT 105

700

600

-;:-
500
~
3:
-
~
w
400 1--AREA1
AREA 2
I
w
0
-J 300
100MW
200 2850MWGJ 02850MW
MTTF = 4380 hr
MTTR= 10hr

100
0 1 2 3 4 5 6 7 8
Sample years (x1000)

Figure 4.22. Area LOEE vs. the number of sample years.

1-- 1
1.3

1.2 AREA 1
--AREA2
1.1
-;:-
1.0
~
.2- 0.9
u.
-J
0 0.8
-J

0.7 100MW r:l

2850 MW GJ
1 ~ 2850 MW
MTTF = 4380 hr
0.6 MTTR = 10 hr

0.5
0 1 2 3 4 5 6 7 8
Sample years (x1000)

Figure 4.23. Area LOLF vs. the number of sample years.

106 CHAPTER 4

7
10000 sample years
6

-
0
>. 5
0
0,...

-
4
x
>-
0
c::: 3
Q)
:J
C'"
....
Q) 2
u. • AREA 1

1
o AREA2

0
0 2 4 6 8 10 12 14 16 18 20
Area LOLE (hr/yr)
Figure 4.24. Distribution of the area LOLE index.

Figures 4.24, 4.25, and 4.26 show the distributions of the area LOLE, LOEE,
and LOLF indices for a sampling period of 10,000 years.

(b) Other Multi-Area Systems. The multi-area systems shown in

Figure 4.20b, c~ d, and e were simulated using the maximum flow algorithm
under the same conditions as those used for the two-area system but for the
different configurations. The obtained reliability indices for 10,000 sample
years for these four multi-area systems are shown in Tables 4.14, 4.15, 4.16,
and 4.17, respectively.

4.5. MULTI-AREA GENERATING SYSTEM

ADEQUACY ASSESSMENT-LINEAR
PROGRAMMING MODEL

4.5.1. Basic Procedure

This section describes a method for multi-area generating system
adequacy assessment based on a combination of the state sampling technique
and a linear programming model.(I5) The basic steps of the method can be
 GENERATING SYSTEM ADEQUACY ASSESSMENT 107

7
10000 sample years

-...
6
( I)
>. 5
0
0
0
..... 4
~
>.
(J
c: 3
Q)
::J

...LLC"
Q) 2
• AREA 1
1 o AREA2

0
0 240 480 720 960 1200
Area lOEE (MWh/yr)
Figure 4.25. Distribution of the area LOEE index.

6 10000 sample years

-!!?
>. 5
0
0
0
.....
-
4
x
>.
(J
c: 3
Q)
::J

...LLC"
Q) 2
• AREA 1
1 o AREA2

0
o 2 4 6 8 10 12 14 16 18 20
Area lOlF (occ/yr)
Figure 4.26. Distribution of the area LOLF index.
108 CHAPTER 4

Table 4.14. Reliability Indices for the

Radially Connected Three-Area System
LOLE LOEE LOLF
(hr/yr) (MWh/yr) (occ./yr)
Area 1 2.203 228.234 0.4700
Area 2 4.478 526.761 0.9626
Area 3 4.349 498.581 0.9546
System 10.629 1253.576

Table 4.15. Reliability Indices for the

Delta-Connected Three-Area System
LOLE LOEE LOLF
(hr/yr) (MWh/yr) (occ./yr)
Area 1 2.012 228.013 0.4549
Area 2 2.041 226.795 0.4607
Area 3 1.972 217.444 0.4530
System 5.888 672.252

Table 4.16. Reliability Indices for the

Four-Area System
LOLE LOEE LOLF
(hr/yr) (MWh/yr) (occ./yr)
Area 1 0.860 94.473 0.2030
Area 2 2.010 222.325 0.4518
Area 3 0.835 86.500 0.2028
Area 4 1.978 218.994 0.4467
System 5.568 622.291

Table 4.17. Reliability Indices for the

Five-Area System
LOLE LOEE LOLF
(hr/yr) (MWh/yr) (occ./yr)
Area 1 0.836 91.757 0.1951
Area 2 0.854 87.702 0.2050
Area 3 0.812 83.489 0.1957
Area 4 0.831 87.104 0.1971
Area 5 0.303 30.104 0.0808
System 3.557 380.156
GENERATING SYSTEM ADEQUACY ASSESSMENT 109

stated as follows:
1. Select a system state S= (SI, S2, ... ,Sn) by sampling techniques,
where Si is the state of the ith component. The set of n components
includes the generating units in each area, the load levels of each
area, and all the tie lines.
2. Evaluate a reliability index R(S) for system state SeG by a linear
programming model, where G denotes the set of all sampled system
states and R(S) represents the selected reliability index for the overall
system or each area.
3. Calculate the expected value of R(S) by

E(R) = I R(S) n(S) (4.30)

SeG N
where N is the number of samples and n(S) is the number of occur-
rences of system state S.
The sampling technique used to select a generating unit state is described
in Section 4.3.1. After all the generating unit states in each area are deter-
mined by sampling, a generation capacity level for the area can be obtained.
A similar sampling technique can also be used to select the tie line states in
a multi-area system. Selection of a load level state and the linear program
model for multi-area reliability evaluation are discussed in the following
sections.

4.5.2. Load Model of a Multi-Area System

The load model of a multi-area system involves two phases. In the first
step, the 8760 hourly load points representing a year are grouped into several
clusters and the mean values of the load points for each area in each cluster
obtained by the K-mean algorithm. In the second step, the area load uncer-
tainty and correlation can be incorporated by a normal distribution sampling
technique and a correlation sampling technique.

(a) The K-Mean Algorithm.(II,12) The K-mean algorithm described

here is an extention to a multi-area system of the cluster technique for the
single-area system given in Section 4.3.3 . It therefore has the similar basic
steps:
Step 1: Select initial values of cluster means Mij, where i andj denote
cluster i and area j.
110 CHAPTER 4

Step 2: Calculate Euclidean distances from each hourly load point to

each cluster mean by

(4.31)

where Dki denotes the Euclidean distance from the kth load
point to the ith cluster mean, L kj the kth load value in the jth
area, and N A is the number of areas.
Step 3: Regroup the points by assigning them to the nearest cluster
and calculate the new cluster means by

Mij= I Lkj/NI i (4.32)

keIC

where NIi is the number of load points in the ith cluster and
IC denotes the set of the load points in the ith cluster.
Step 4: Repeat Steps 2 and 3 until all cluster means maintain
unchanged between iterations.
The cluster means Mij are then used as the load levels for each area in
each cluster. Each load level represents NIi load points in the corresponding
cluster in a mean value sense. Correlation between mean values of area load
points in each cluster is captured in the K-mean algorithm. The captured
correlation is approximate since it reflects correlative relation between cluster
means. After the reliability indices for each cluster are calculated, annual
reliability indices can be obtained by weighting the indices for each cluster
by (NI i /8760).

(b) Area Load Uncertainty and Correlation. Load uncertainty

and correlation always exist in an actual power system and it has long been
recognized that they can have a great impact in power system reliability
evaluation.
Load uncertainty can be modeled using a normal distribution and there-
fore it is necessary to draw normally distributed random numbers in order
to simulate area load uncertainty. A normal distribution has two parameters:
mean value and standard deviation. The mean values are the cluster means
obtained by the K-mean algorithm and the standard deviation is assigned
according to the perceived load forecast uncertainty, such as 5%. The normal
distribution sampling techniques given in Section 3.4.5 can be used to create
normally distributed random numbers.
If all the area loads are completely dependent, the cluster means of all
the area loads are the same for a cluster. In this case, only one normally
distributed random number is required for each cluster in order to determine
GENERATING SYSTEM ADEQUACY ASSESSMENT 111

the load states of all areas. If all area loads are not completely dependent
but are correlated to some extent, it is necessary to generate a normally
distributed random vector for each cluster, in which each component corre-
sponds to the load state of a particular area and whose components satisfy
the correlation between the loads of different areas in the same cluster. The
correlation matrix corresponding to each cluster can be calculated as follows.
According to the statistical definitions of variance, covariance, and cor-
relation coefficient, the element of the correlation matrix corresponding to
area n and area m for the ith cluster is

(4.33)

The covariance matrix C corresponding to the correlation matrix p can

be obtained by

(4.34)

where (J'j is the standard deviation for the ith area load uncertainty. The
same standard deviation is usually assumed for all area loads. In this case,
equation (4.34) is simply written as
C= (J'2 P (4.35)
After the covariance matrix C between area loads for each cluster is
obtained, the following load correlation sampling technique can be used to
simulate area load uncertainty and correlation.
Let H be an NA (number of areas) dimensional normally distributed
random vector with mean vector B and covariance matrix C. Let G be an
NA dimensional normally distributed random vector whose components are
independent of each other and each component has a mean of zero and a
variance of unity. Linear combination of normal distributions is still a nor-
mal distribution. There exists therefore a matrix A which can create the
following transformation relationship between Hand G (\6) :
H=AG+B (4.36)
The mean value vector B and the covariance matrix C of H can be
calculated from equation (4.36) as follows:
E(H) =AE(G) + B=AO+ B=B (4.37)
E[(H-B)(H-B)T] =E(AGGTAT) =AAT = C (4.38)
112 CHAPTER 4

Equation (4.38) gives the relation between matrix A and matrix C. On the
other hand, covariance matrix C is a nonnegative definite symmetric matrix
and therefore matrix C can be triangularized into the unique lower triangular
matrix multiplied by its transposed matrix. Consequently, in the following
equation,
(4.39)
where A is a lower triangular matrix. As a result, the following recursive
formulas can be derived from equation (4.39) to calculate the elements of
matrix A from those of matrix C:

A - Cil (l~i~NA) (4.40)

il - (Cll )I/2

Aii= (Cii- iII A~)1/2 (l<i~NA) (4.41)

k~1

Ci j - ,,}-I
L..k~1
A ik}k
A
A··=------ (l <j<i~NA) (4.42)
Ij A··
1J

The area load correlation sampling technique can be summarized in the

following steps:
Step 1: Calculate the area load covariance matrix C according to equa-
tions (4.33) and (4.34) from the information obtained by the
K-mean algorithm.
Step 2: Calculate the low triangular matrix A according to equations
(4.40)-(4.42).
Step 3: Draw an NA dimensional, normally distributed random vector
G. All its components have a mean of zero and a variance of
unity.
Step 4: Create an NA dimensional correlative normally distributed
random vector H from equation (4.36) in which B is a mean
value vector whose components are the cluster means. Both
area load uncertainty and the correlation defined by matrix C
are included in the random vector H.

4.5.3. Linear Programming Model for

Multi-Area Reliability Evaluation
After the area generating capacity levels, area load states, and tie line
states are selected using the sampling techniques described earlier, it is neces-
sary to evaluate the reliability indices of the selected multi-area system state.
GENERATING SYSTEM ADEQUACY ASSESSMENT 113

All areas can be divided into two sets of supporting and supported areas.
In the supporting set, the available area generating capacity is larger than
the area load. A "generator variable" is defined for each area in the support-
ing set. The upper limit of each "generator variable" is the difference between
the available area generating capacity and the area load. In the supported
set, the available area generating capacity is smaller than the area load. A
"required load" is defined for each area in the supported set, which equals
the difference between the area load and the available area generating capac-
ity. A "fictitious generator variable" is also defined for each area in the
supported set. When the "required load" at each area cannot be completely
satisfied due to the insufficient total available generating capacity in the
supporting set or/and limits of tie line capacities, the "fictitious generator
variables" can provide the unsatisfied parts of the "required load" such that
power balance in each area is always guaranteed. Essentially, the "fictitious
generator variables" are load curtailment variables in the supported areas
and therefore the upper limit of each "fictitious generator variable" is
assigned as the relative "required load."
The "generator variables," the "fictitious generator variables," the
"required loads," and tie lines constitute a new small "generation-transmis-
sion" system. The basic objective is to minimize the total load curtailment
while satisfying the power balance at each node (area) and upper limits of
the tie line capacities, the "generator variables," and the "fictitious generator
variables." The following linear programming model can be used for this
purpose:
min L a,GFi (4.43)
ieND

subject to
L TPj+GP/=O (ieNG) (4.44)
j~i

L TPj+ GFi= DP/ (ieND) (4.45)

j-+i

ITPjl ~TjX UeNT) (4.46)

O~GP/~SPi (ieNG) (4.47)

O~GFi~DP/ (ieND) (4.48)

where GP/ and GF/ denote the "generator variables" and the "fictitious
generator variables" at the ith node, respectively; ''j-+i'' indicates that j
belongs to the line set in which all lines are connected to node i; TPj is the
tie line power on line j and it is positive when entering node i and negative
114 CHAPTER 4

when issuing from node i; DPj is the "required load" at the ith node; SPj
is the upper limit of the ith "generator variable"; Tjax is the capacity limit
ofthejth tie line; NG, ND, and NT are the sets of the "generator variables,"
the "fictitious generator variables," and tie lines, respectively; a j is the
weighting factor for the ith "fictitious generator variable."
There is a wide variety of possible supporting policies in multi-area
generating system reliability assessment. One advantage of a linear program-
ming model such as that described is that variable supporting policies can
be easily incorporated. Different supporting policies and their effects are
discussed in Section 4.6. A priority order supporting policy which is most
commonly used, is considered first. This can be incorporated by assigning
different values of weighting factors aj. The supported area with the first
priority has a maximum value of a j, the supported area in the next priority
has a second maximum value of aj, etc. These values of aj are specified in
terms of their relative magnitudes and therefore it is easy to select these
values.
Another advantage of the linear programming model is that the LOEE
sensitivity indices of both system and area to area generation and tie line
capacities can be calculated. These sensitivity indices can provide informa-
tion about which tie line or which area generation capacity should be
reinforced from an overall system and each area point of view. It should be
noted that it is not possible to obtain the area sensitivity indices merely by
using the concept of dual variables. Calculation of the area sensitivity indices
can be explained as follows. For a standard linear programming problem
(4.49)
subject to
AX=b (4.50)
X~O (4.51)
its optimal and feasible solution can be expressed by
Xb=B-Ib (4.52)
where B is the optimal basis at the optimal and feasible solution and X b is
the basic variable subvector. The value of B will remain unchanged if the
right-hand term b has a sufficiently small change. Therefore
IlXb =B- l l1b (4.53)
which means

a(Xb)j= (B- 1) .. (4.54)

aboJ IJ

where (B-I)ij is the element of B- 1•

GENERATING SYSTEM ADEQUACY ASSESSMENT 115

Equation (4.54) indicates that the elements in the optimal basis are the
sensitivities of the basic variables to the elements in the right-hand term of
the constraints. Area sensitivity indices to area generations and tie line capac-
ities can therefore be obtained by solving the linear programming model
expressed by equations (4.43)-(4.48). The basic concepts and solution tech-
niques oflinear programming are given in Appendix D.1 for the convenience
of the reader.

4.5.4. Case Studies

(a) The Test System. A four-area system shown in Figure 4.27 is
considered as the test system. Area 1 is the basic IEEE RTS. The annual
peak load is 2850 MW. Areas 2, 3, and 4 are three modified RTSs. The
modifications are as follows:
Area 2: Three 197-MW generators are removed and the annual peak
load is reduced to 2280 MW.
Area 3: One 800-MW generator with forced unavailability FU=0.18
is added and the annual peak load is increased to 3575 MW.
Area 4: One 800-MW generator with FU =0.18 and one 600-MW gen-
erator with FU=0.15 are added and the annual peak load is
increased to 4180 MW.

Figure 4.27. A four-area generating system.

116 CHAPTER 4

Table 4.18. The Tie Line Data of

the Test System
From area Capacity
No. to area (MW) FU
1 1-2 100 0.0005
2 1--4 200 0.0005
3 2-3 100 0.0005
4 3-4 200 0.0005
5 1-3 200 0.0005
6 2--4 200 0.0005

The tie line data are given in Table 4.18.

When the area loads are considered to be completely dependent, the
same annual load curve data of the IEEE RTS given in Appendix A.l are
applied to the four areas. When correlation between area loads (noncom-
pletely dependent) is considered, the area load curve in Area I remains
unchanged and the area load curves in Areas 2, 3, and 4 were obtained by
displacing the hourly load values of Area 2 ahead by 4 hr, those of Area 3
ahead by 2 hr, and those of Area 4 backward by 2 hr, respectively.
In order to include generating unit derated states, the 350-MW and the
two 400-MW generators were derated to 50% capacity with the derated state
probabilities given in Table 4.10. The state probabilities of the derated
models are such that the derating-adjusted two-state model data are identical
to those given in Appendix A.I. The specified supported priority order is:
Area 4-Area 3-Area 2-Area 1.

(b) LOLE and LOEE Indices. The following eight cases were
studied:
Case 1: Completely dependent area loads, no load uncertainty, and no
generating unit derated states.
Case 2: Correlation between area loads, no load uncertainty, and no
generating unit derated states.
Case 3: Completely dependent area loads, load uncertainty (5% stan-
dard deviation), and no generating unit derated states.
Case 4: Correlation between area loads, load uncertainty (5% standard
deviation), and no generating unit derated states.
Cases 5 to 8 correspond to Cases I to 4, respectively, with the difference
that derated states of 50% capacity for the 350-MW and the two 400-MW
generators in each area are included.
One 20-step load model was created using the K-mean algorithm to
represent the annual load curve in the cases of completely dependent area
loads while four 20-step load models for the four areas were created when
GENERATING SYSTEM ADEQUACY ASSESSMENT 117

correlation between area loads was considered. The stopping rule used in
the studies is that the simulation in one load level step ends when either
16,000 samples of the system states have been simulated or the coefficient
of variation for the system LOEE is less than 0.1. The results for these
studies are shown in Tables 4.19-4.22.
It can be seen from the results in Tables 4.19 to 4.22 that the indices
in the cases of completely dependent area loads are larger than those in the
cases of correlation between area loads under the same conditions of load
uncertainty and/or generating unit derated states. Area load uncertainty can

Table 4.19. LOLE Indices for Cases 1 to 4 (hr/yr)

Case 1 Case 2 Case 3 Case 4
Area 1 0.5200 0.1856 1.5071 0.5683
Area 2 0.6572 0.2879 1.2968 0.5493
Area 3 4.6784 3.2533 6.8250 4.6746
Area 4 6.7150 5.6493 9.9241 8.1607
System 11.6947 9.0490 17.1283 13.3346

Table 4.20. LOEE Indices for Cases 1 to 4 (MWh/yr)

Case 1 Case 2 Case 3 Case 4
Area 1 48.5333 11.2750 212.6279 61.2335
Area 2 73.9207 18.0256 1628383 32.2251
Area 3 736.5811 388.3348 1222.5161 648.8776
Area 4 1309.9360 1064.9319 1972.3239 1501.6836
System 2168.9714 1482.5678 3570.3071 2244.0203

Table 4.21. LOLE Indices for Cases 5 to 8 (hr/vr)

Case 5 Case 6 Case 7 Case 8
Area 1 0.1626 0.0413 0.8033 0.2745
Area 2 0.5423 0.2879 1.0867 0.4558
Area 3 2.1571 1.0869 3.7754 2.2162
Area 4 3.1756 2.4964 5.1385 4.3451
System 5.7332 3.8505 9.3108 6.9754

Table 4.22. LOEE Indices for Cases 5 to 8 (MWh/vr)

Case 5 Case 6 Case 7 Case 8
Area 1 18.5329 2.4349 115.8170 26.6675
Area 2 53.7787 18.9414 133.9687 29.9503
Area 3 298.2609 106.2488 608.0114 262.7329
Area 4 532.6486 391.9146 1010.1765 694.6565
System 903.2213 519.5398 1867.9736 1014.0071
118 CHAPTER 4

Table 4.23. lOEE Sensitivity Indices of System and Areas to Area

Generation and Tie line Capacities in Case 1 (MWh/yr/MW)
Areal Area 2 Area 3 Area 4 System
Area 1 -0.5154 -0.0087 -0.1747 -0.0699 -0.7688
Area 2 -0.0175 -0.6552 -0.0437 -0.0349 -0.7513
Area 3 -0.1048 -0.0262 -4.6738 -0.0699 -4.8747
Area 4 -0.0699 -0.1048 -0.5416 -6.7093 -7.4256
Line 1 -0.3407 -0.5591 -0.0437 -0.0699 -1.0134
Line 2 -0.2097 -0.0786 -0.3145 -6.4035 -7.0063
Line 3 -0.0874 -0.5591 -4.4554 -0.1l36 -5.2154
Line 4 -0.0087 -0.0699 -3.6953 -6.1239 -9.8979
LineS -0.1485 -0.0087 -4.1758 -0.0000 -4.3331
Line 6 -0.0437 -0.4543 -0.4630 -6.6132 -7.5741

have a great impact on the calculated reliability indices and always provides
a more pessimistic appraisal of multi-area system reliability. Utilization of
derated state models for large-capacity generating units provides a more
optimistic estimation of multi-area system reliability compared to the utiliza-
tion of a derating-adjusted two-state model.

(c) Sensitivity Indices. The LOEE sensitivity indices for Cases 1

to 4 are listed in Tables 4.23 to 4.26. It should be noted that Area No. or
"System" in the first row corresponds to incremental variations of LOEE
in the sensitivity indices, while Area No. or Line No. in the first column
corresponds to incremental variations of area generations or tie line capaci-
ties. For example, the value -0.1747 in the second row and the fourth
column indicates that if the generation of Area 1 is increased by 1 MW, the
LOEE index of Area 3 is decreased by 0.1747 MWh/year.

Table 4.24. lOEE Sensitivity Indices of System and Areas to Area

Generation and Tie line Capacities in Case 2 (MWh/yr/MW)
Area 1 Area 2 Area 3 Area 4 System
Area 1 -0.1835 -0.0000 -0.0961 -0.0349 -0.3145
Area 2 -0.0000 -0.2796 -0.0000 -0.0000 -0.2796
Area 3 -0.0612 -0.0000 -3.2498 -0.0000 -3.3109
Area 4 -0.0349 -0.0349 -0.2883 -5.6435 -6.0016
Line 1 -0.1835 -0.2621 -0.0000 -0.0000 -0.4455
Line 2 -0.0786 -0.0349 -0.1922 -5.5212 -5.8269
Line 3 -0.0612 -0.2621 -3.2323 -0.0349 -3.5905
Line 4 -0.0175 -0.0349 -2.7693 -5.4600 -8.2817
Line 5 -0.0612 -0.0000 -3.0489 -0.0000 -3.1100
Line 6 -0.0350 -0.2184 -0.2883 -5.6260 -6.1676
GENERATING SYSTEM ADEQUACY ASSESSMENT 119

Table 4.25. LOEE Sensitivity Indices of System and Areas to Area

Generation and Tie Line Capacities in Case 3 (MWh/yr/MW)
Area I Area 2 Area 3 Area 4 System
Area I -1.5026 -0.0524 -0.4543 -0.4106 -2.4199
Area 2 -0.1310 -1.2929 -0.2184 -0.3320 -1.9743
Area 3 -0.3145 -0.1398 -6.8228 -0.2359 -7.5130
Area 4 -0.2970 -0.2883 -1.1532 -9.9241 -11.6626
Line 1 -0.8649 -0.8387 -0.0961 -0.1747 -1.9743
Line 2 -0.4717 -0.2009 -0.5766 -8.8059 -10.0551
Line 3 -0.1660 -0.7688 -5.8706 -0.2184 -7.0237
Line 4 -0.1223 -0.2097 -4.7349 -8.6312 -13.6981
Line 5 -0.4543 -0.0437 -5.3639 -0.0349 -5.8968
Line 6 -0.1747 -0.5242 -0.8125 -9.2077 -10.7191

Table 4.26. LOEE Sensitivity Indices of System and Areas to Area

Generation and Tie Line Capacities in Case 4 (MWh/yr/MW)
Areal Area 2 Area 3 Area 4 System
Area 1 -0.5678 -0.0000 -0.0262 -0.0524 -0.6465
Area 2 -0.0000 -0.5416 -0.0000 -0.0262 -0.5678
Area 3 -0.1310 -0.0262 -4.6738 -0.0262 -4.8572
Area 4 -0.0349 -0.1048 -0.1922 -8.1594 -8.4914
Line 1 -0.5329 -0.5416 -0.0000 -0.0262 -1.1007
Line 2 -0.3145 -0.1048 -0.1660 -7.7925 -8.3778
Line 3 -0.1310 -0.4892 -4.5078 -0.0524 -5.1805
Line 4 -0.1136 -0.0786 -4.2020 -7.8275 -12.2217
Line 5 -0.3494 -0.0262 -4.4816 -0.0000 -4.8572
Line 6 -0.0349 -0.3757 -0.1922 -8.0721 -8.6749

It can be seen from the results in Tables 4.23 to 4.26 that although the
values of the sensitivity indices are different in the various cases considering
completely dependent area loads or correlation between area loads and area
load uncertainty or no load uncertainty, their ranking order in these cases
are basically the same. On the other hand, the ranking order of the sensitivity
indices for the overall system and each area are different. From the overall
system point of view, the best choice is to reinforce Tie-line 4, followed by
Tie-line 6 and the generating capacity in Area 4. From an Area 4 point of
view, the best choice is to reinforce the generating capacity in Area 4, fol-
lowed by Tie-lines 6 and 2. From an Area 3 point of view, the best choice
is to reinforce the generating capacity in Area 3, followed by Tie-lines 3 and
5. From an Area 2 point of view, the best choice is to reinforce the generating
capacity in Area 2, followed by Tie-lines 1 and 3. From an Area 1 point of
view, the best choice is to reinforce the generating capacity in Area 1, fol-
120 CHAPTER 4

lowed by Tie-line 1. Increasing the generating capacity of each area naturally

has the greatest impact on improving the reliability of each area itself. It is
interesting to note that for the overall system, the first choice should be to
reinforce two tie lines instead of increasing the generating capacity of the
multi-area system.

4.6. DIFFERENT SUPPORTING POLICIES IN

MULTI-AREA GENERATING SYSTEM
ADEQUACY ASSESSMENT
A power system pool or interconnected system configuration is usually
represented by a group of areas each of which is associated with a specific
utility company or a jointly owned generation facility. It is therefore impor-
tant to calculate area indices which indicate area reliability. In order to
obtain realistic area indices, a supporting policy must be clearly specified.
Different supporting policies can lead to different area indices and therefore
a different appreciation of area reliability. This section discusses how to
incorporate different supporting policies into the linear programming model
presented in Section 4.5.3 and examines their effects in multi-area generating
system adequacy assessment. (17)

4.6.1. Incorporation of Different Supporting

Policies
There is a wide variety of supporting policies. A firm interchange con-
tract can be considered by adjusting the area load levels and the relative tie
line capacity. The following two basic emergency action philosophies can
be used in a loss-of-Ioad situation.
• The load loss sharing philosophy: Each area shares the unserved
demand according to a supporting policy, recognizing tie line
constraints .
• The no-load-loss sharing philosophy: Each area attempts to meet its
own demand. If there is excess capacity, it is supplied to supported
areas in accordance with an agreed supporting policy.
The basic emphasis here is not on the fundamental difference between
these two philosophies but on the effects due to possible variations of sup-
porting policies which can be used in both philosophies. Different supporting
policies will lead to different area reliability evaluation even if the same
GENERATING SYSTEM ADEQUACY ASSESSMENT 121

basic philosophy is used. The linear programming model presented earlier

corresponds to the general no-load-loss sharing philosophy. In the case of
the loss-load sharing philosophy, the required linear programming model
has the same mathematical form but it is necessary to define "generator
variables" and "required loads" at both supporting and supported area
sets. The following discussion therefore applies to both philosophies. Four
supporting policies are specifically considered but additional supporting
policies can be introduced by similar techniques.

(a) Priority Order Policy. This supporting policy has been

described in Section 4.5.3. A priority order can be considered by assigning
different values of the weighting factors ai in equation (4.43).

(b) Shortest Distance Policy. This supporting policy states that

excess capacities in supporting areas should be supplied to the nearest sup-
ported areas first. It also states that line power losses caused by supporting
power flowing on tie lines are minimized. In order to incorporate this
supporting policy, lengths of tie lines must be included in the input data.
The objective function given in equation (4.43) is changed into equation
(4.55) :

(4.55)
jeNT ieND

where pj denotes the perceived length (in kIn or mile) of the jth tie line and
the value of a satisfies

a»max {Pj}
jeNT

Equations (4.44) and (4.45) guarantee power balance at each area. Selecting
a large value of a in equation (4.55) guarantees that load curtailments will
be zero if avoidable or minimized if unavoidable. Introduction of the first
term in equation (4.55) enables the supporting power to trace the shortest
path.

(c) Priority Order Plus Path Order Policy. This supporting

policy includes two conditions: (a) excess capacities in a supporting area
set are supplied to supported areas according to a specified priority;
(b) when supporting power cannot directly reach the privileged supported
122 CHAPTER 4

Figure 4.28. Explanation of supporting policy 3.

area due to tie line capacity limits, it should be supplied to supported

areas according to the path order. A simple example is given to illustrate
this policy. As shown in Figure 4.28, Area 1 has excess capacity of 150 MW
while Areas 2 and 3 have shortage capacities of 200 MW and 30 MW,
respectively.
The capacity limit of each tie line is 100 MW. Area 2 is assumed to
have the supported privilege. According to Condition (a), 100 MW flows
directly from Area 1 to Area 2. According to Condition (b), another 50 MW
flows from Area 1 to Area 3 and then 30 MW are left for Area 3 while the
remaining 20 MW supports Area 2.
In order to incorporate this supporting policy into the linear program-
ming model, the tie line set NT is divided into three subsets of NTI, NT2,
and NT3. The lines connecting two supporting areas belong to NTI, those
connecting one supporting area and one supported area belong to NT2, and
those connecting two supported areas belong to NT3. The objective function
given in equation (4.43) is changed into equation (4.56):
(4.56)
jENTI kENT2 mENT3 iEND

The coefficients in the new objective function (4.56) meet the following four
requirements:
GENERATING SYSTEM ADEQUACY ASSESSMENT 123

2. The value of Pk (keNT2) are specified in such a way that Pk of

the lines connecting the first priority supported area have minimum
values, those connecting the second priority supported area have the
second minimum values, etc.
3. The values of Pm (me NT3) satisfy

min {Pm} > max {Pk}

meNT3 keNT2

4. The values of ai are specified using the same principle as in the

priority order policy and satisfy

min {at}» max {Pm}

ieND meNT3

The priority order plus path order policy can be implemented automatically
in the linear programming approach by selecting the coefficients which meet
the four requirements.

(d) Proportional Policy. This supporting policy includes two

aspects:
1. In those states where the total excess capacity in the supporting areas
is larger than the total shortage capacity in the supported areas, the shortage
capacity of each supported area can be compensated despite the supporting
priority and path if the tie lines are capable of transmitting the supporting
power. In other words, the area load curtailment depends only on the tie
line capacity limits. The objective function given in equation (4.43) is
changed into equation (4.57):

mina L GFi (4.57)

ieND

This means that each supported area has an equal opportunity for load
curtailment in the objective function. Whether load curtailments occur
depends only on the constraints expressed by equation (4.46).
2. In those states where the total excess capacity in the supporting areas
is smaller than the total shortage capacity in the supported areas, load
curtailments are unavoidable. A proportional curtailment principle can be
designed. The supporting-supported ratio principle is used here and other
possible proportional principles can be treated similarly. Each load curtail-
ment variable GFi is divided into two subvariables which are expressed by
GFil and GFa and have different weighting factors in the objective function
and different upper limits in the constraints. The linear programming model
124 CHAPTER 4

given by equations (4.43) to (4.48) is changed into

min a L GFil + L p,GF a (4.58)

ieND ieND
subject to

(ieNG) (4.59)
j-+i

2
L TPj + L GFik=DPi (ieND) (4.60)
j .... i k=1

ITPA ~ Tjax UeNT) (4.61)

O~GPi~SPi (ieNG) (4.62)

O~GFil~DPiR (ieND) (4.63)

O~GFa~DPi(1- R) (ieND) (4.64)

where R denotes the supporting-supported ratio which is defined as

L' NOSPi
R = --'..'e=..;..:'----_ (4.65)
LieNDDPi
GF il corresponds to the portion which can be compensated when tie line
capacity limits are not reached, and GFa the portion which may be supple-
mentarily compensated in the case that GFil cannot be compensated in some
areas because of tie line capacity limits; a and Pi satisfy
a > max {Pi}
ieND
Such a selection guarantees that the total excess capacity attempts to
compensate each supported area in terms of the proportion R. If this cannot
be reached because of tie line limits, the surplus capacity can compensate
parts reflected in the proportion (1- R).

4.6.2. Case Studies

Studies were conducted using two test systems. Test System 1 is shown
in Figure 4.27 and all data are given in Section 4.5.4. Test System 2 is
obtained by modifying Test System 1 in such a way that it has relatively
low generation reserve margin and relatively high tie line capacity limits:
GENERATING SYSTEM ADEQUACY ASSESSMENT 125

Table 4.27. The Tie Line Data for the Two Systems
Capacity (MW)
From area Length
No. to area (km) System 1 System 2 FU
1 1-2 100 100 100 0.0005
2 1-4 250 200 300 0.0005
3 2-3 100 100 200 0.0005
4 3-4 250 200 300 0.0005
5 1-3 150 200 200 0.0005
6 2-4 300 200 300 0.0005

• Modification 1: The annual peak loads of Areas 1 and 2 are

increased to 3000 MW and 2500 MW, respectively .
• Modification 2: The tie line capacities are modified as shown in
Table 4.27.
The following four cases were studied for both test systems. The reli-
ability index selected for these studies is the Loss Of Energy Expectation
(LOEE).
Case 1: Completely dependent area loads and no load uncertainty
Case 2: Correlation between area loads and no load uncertainty
Case 3: Completely dependent area loads and load uncertainty (5%
standard deviation)
Case 4: Correlation between area loads and load uncertainty (5% stan-
dard deviation)
In each case, the four supporting policies presented in the previous
subsection were considered. The specified supported priority is: Area 4-
Area 3-Area 2-Area 1. The stopping rule for the simulation is the same as
that used in Section 4.5.4.
It can be seen from the results in Tables 4.28 to 4.35 that considering
load uncertainty provides a higher estimate of system and area reliability
indices while considering correlation between area loads can lead to a lower
estimate of these indices. This impact is greater on Test System 2, which has
a lower generation reserve margin compared to Test System 1. Different
supporting policies create the same system indices, but they can provide
different area reliability indices depending on tie line parameters and config-
urations. The area indices due to different supporting policies are quite close
for Test System 1 and have relatively large differences for Test System 2.
This is caused by the fact that Test System 2 has higher tie line capacities
and therefore different supporting situations are less constrained by the tie
lines. When the tie lines have low capacities, such as in the case of Test
126 CHAPTER 4

Table 4.28. LOEE Indices in Case 1 for Test System 1 (MWh/yr)

Supporting policies
(a) (b) (c) (d)
Area 1 48.5333 15.8760 18.9144 37.2703
Area 2 73.9207 61.0366 61.1564 58.2003
Area 3 736.5811 662.6406 674.6348 665.7415
Area 4 1309.9360 1429.4180 1414.2658 1407.7592
System 2168.9714 2168.9714 2168.9714 2168.9714

Table 4.29. LOEE Indices in Case 2 for Test System 1 (MWh/yr)

Supporting policies
(a) (b) (c) (d)
Area 1 11.2750 1.0111 1.0111 6.5628
Area 2 18.0256 16.6377 16.6377 15.0702
Area 3 388.3348 353.2820 361.5459 361.0461
Area 4 1064.9319 1111.6367 1103.3728 1099.8884
System 1482.5678 1482.5678 1482.5678 1482.5678

Table 4.30. LOEE Indices in Case 3 for Test System 1 (MWh/yr)

Supporting policies
(a) (b) (c) (d)
Area 1 212.6279 91.9890 109.3441 136.7954
Area 2 162.8382 101.7723 121.9547 110.0646
Area 3 1222.5161 1099.6311 1155.6234 1098.3940
Area 4 1972.3239 2276.9138 2183.3840 2225.0520
System 3570.3071 3570.3071 3570.3071 3570.3071

Table 4.31. LOEE Indices in Case 4 for Test System 1 (MWh/yr)

Supporting policies
(a) (b) (c) (d)
Area 1 61.2335 28.6084 28.6084 41.7482
Area 2 32.2251 24.4907 24.4907 24.2376
Area 3 648.8776 614.1002 621.5178 623.6267
Area 4 1501.6836 1576.8206 1569.4031 1554.4074
System 2244.0203 2244.0203 2244.0203 2244.0203
GENERATING SYSTEM ADEQUACY ASSESSMENT 127

Table 4.32. LOEE Indices In Case 1 for Test System 2 (MWh/yr)

Supporting policies
(a) (b) (c) (d)
Area 1 123.3989 63.7031 98.3037 75.6751
Area 2 128.4102 94.0824 109.1272 93.0787
Area 3 337.1798 236.1089 317.9130 281.0001
Area 4 333.2816 528.3761 396.9266 472.5166
System 922.2706 922.2706 922.2706 922.2706

Table 4.33. LOEE Indices in Case 2 for Test System 2 (MWh/yr)

Supporting policies
(1) (2) (3) (4)
Area 1 33.1458 15.0634 22.5880 18.1044
Area 2 27.1487 22.5329 21.4571 23.1804
Area 3 114.9670 77.9003 103.6966 90.9659
Area 4 166.8760 226.6410 194.3958 209.8867
System 342.1375 342.1375 342.1375 342.1375

Table 4.34. LOEE Indices in Case 3 for Test System 2 (MWh/yr)

Supporting policies
(a) (b) (c) (d)
Area 1 498.1291 292.8050 424.1649 315.0927
Area 2 448.1004 317.5350 405.0695 326.6522
Area 3 830.8882 617.5696 833.0615 758.6989
Area 4 854.4456 1403.6537 969.2673 1231.1194
System 2631.5635 2631.5635 2631.5635 2631.5635

Table 4.35. LOEE Indices in Case 4 for Test System 2 (MWh/yr)

Supporting policies
(a) (b) (c) (d)
Area 1 137.7941 68.8414 76.9299 75.3196
Area 2 55.9805 34.7817 40.8426 39.6034
Area 3 202.4308 167.8020 195.3536 186.5768
Area 4 292.8550 417.6354 375.9344 387.5604
System 689.0602 689.0602 689.0602 689.0602
128 CHAPTER 4

System I, the range of possible supporting scenarios becomes relatively nar-

row and therefore different supporting policies do not create large differences
in the area indices.

4.7. CONCLUSIONS
This chapter discusses the application of Monte Carlo methods in both
single-area and multi-area generating system adequacy assessment. Two
basic approaches-the state duration sampling method and the state sam-
pling method-are presented. In the case of the state duration sampling
method, the load model is relatively simple and is the direct chronological
hourly load curve. The major focus is therefore on generating unit modeling.
The crucial aspect in this method is how to obtain an operating cycle of a
particular generating unit model by sampling. In the case of the state sam-
pling method, however, the generating unit model is relatively simple as the
state of a generating unit can be simulated by a uniformly distributed ran-
dom number. Emphasis is therefore placed on the load model in this method.
The cluster technique can be used to create a multistep load level model for
both single-area and multi-area systems. A normal distribution sampling
technique and a correlation sampling technique can be used to simulate load
uncertainty and correlation between area loads.
In the case of a multi-area system, it is necessary to model the assistances
between areas and the tie line constraints. Two techniques in the form of a
maximum flow algorithm and a linear programming model are presented
for this purpose. One of the advantages of the linear programming model
is that the LOEE sensitivity indices of both system and area to area genera-
tion and tie line capacities can be obtained. These sensitivity indices provide
information about which tie line or which area generating capacity should
be reinforced. A further advantage is that a wide variety of supporting
policies can be readily incorporated into the model. Four supporting policies
and their effects on multi-area system reliability evaluation are presented
and discussed. Readers can use similar techniques to consider other possible
supporting policies.
Selection of a stopping rule is an important factor in Monte Carlo
simulation. The stopping criterion can be the pre specified number of
samples, or the coefficient of variation tolerance, or a combination of both.
A large prespecified number of samples or/and a small tolerance coefficient
can provide relatively high accuracy in reliability indices with an attendant
increase in CPU time. A reasonable stopping criterion should be specified
for a particular system in order to provide a compromise between the accur-
GENERATING SYSTEM ADEQUACY ASSESSMENT 129

acy and computing time. This may need to be examined in some detail for
the configuration in question.

4.8. REFERENCES
1. R. Billinton, "Bibliography on Application of Probability Methods in the Evaluation of
Generating Capacity Requirements," IEEE/PES Winter Meeting, 1966, paper 31 CP 66-
62.
2. R. Billinton, "Bibliography on the Application of Probability Methods in Power System
Reliability Evaluation," IEEE Trans. on PAS, Vol. 91, 1972, pp. 649-660.
3. IEEE Committee Report, "Bibliography on the Application of Probability Methods in
Power System Reliability Evaluation, 1971-1977," IEEE Trans. on PAS, Vol. 97, 1978,
pp. 2235-2242.
4. R. N. Allan, R. Billinton, and S. H. Lee, "Bibliography on the Application of Probability
Methods in Power System Reliability Evaluation, 1977-1982," IEEE Trans. on PAS,
Vol. 103, 1984, pp. 275-282.
5. R. N. Allan, R. Billinton, S. M. Shahidehpour, and C. Singh, "Bibliography on the Applica-
tion of Probability Methods in Power System Reliability Evaluation, 1982-1987," IEEE
Trans. on PS, Vol. 3, No.4, 1988, pp. 1555-1564.
6. R. Billinton and R. N. Allan, Reliability Evaluation of Power Systems, Plenum, New York,
1984.
7. R. Billinton and R. N. Allan, Reliability Assessment of Large Electric Power Systems,
Kluwer, Dordrecht, 1988.
8. L. Gan, "Multi-Area Generating System Adequacy Assessment by Monte Carlo Simula-
tion," M.Sc Thesis, University of Saskatchewan, 1991.
9. IEEE Task Group, "A Four-State Model for Estimation of Outage Risk for Units in
Peaking Service," IEEE Trans. on PAS, Vol. 91, 1972, pp.618-627.
10. R. N. Allan, R. Billinton, and N. M. Abdel-Gawad, "The IEEE Reliability Test System-
Extension to and Evaluation of the Generating System," IEEE Trans. on PWR, Vol. 1,
No.4, 1986, pp. 1-7.
11. M. S. Aldenderfer and R. K. Blashfield, Cluster Analysis, Sage, Newbury Park, 1984.
12. H. Spath, Cluster Analysis Algorithms for Data Reduction and Classification of Objects,
Halsted, New York, 1980.
13. R Billinton and L. Gan, "A Monte Carlo Simulation Model for Adequacy Assessment of
Multi-Area Generating Systems," Proceedings of the Third International Conference on
Probabilistic Methods Applied to Electric Power Systems, lEE, London, 3-5 July, 1991.
14. R. Billinton and L. Gan, "Use of Monte Carlo Simulation in Teaching Generating Capacity
Adequacy Assessment," IEEE Trans. on PS, Vol. 6, No.4, 1991, pp. 1571-1577.
15. R. Billinton and W. Li, "A Monte Carlo Method for Multi-Area Generation System
Reliability Assessment," IEEE Trans. on PS, Vol. 7, No.4, 1992, pp. 1487-1492.
16. Wan Zikun, Elements of Probability Theory and Its Application, Science Publishing House,
Beijing, 1979.
17. W. Li and R. Billinton, "Incorporation and Effects of Different Supporting Polices in
Multi-Area Generation System Reliability Assessment," IEEE Trans. on PS, Vol. 8, No.3,
1993, pp. 1061-1067.
 5
Composite System
Adequacy
Assessment

5.1. INTRODUCTION
The basic objective of composite generation and transmission system
adequacy assessment (HL2) is to evaluate the ability of the system to satisfy
the load and energy requirements at the major load points. This evaluation
domain involves the joint reliability problem of generating sources and trans-
mission facilities and is sometimes called bulk system analysis. Although
this activity came much later than comparably significant developments in
HLI evaluation, considerable effort has been devoted to this area over the
last 25 years. (1-7)
As in HLl evaluation, there are two fundamental approaches to com-
posite system adequacy assessment: analytical enumeration and Monte
Carlo simulation. The analytical enumeration techniques have been devel-
oped mainly in North America.(8-15) Monte Carlo simulation techniques
were first advanced in Europe by EDF (France) and ENEL (Italy) and then
were further developed by work in Brazil and in North AmericaY6-30) A
Monte Carlo based computer program named MECORE has been devel-
oped at the University of Saskatchewan and enhanced at BC Hydro. This
chapter illustrates applications of Monte Carlo methods in composite system
adequacy assessment. The MECORE program utilizes the techniques
presented in this chapter and has been applied to reliability assessment of
practical composite systems.
Adequacy assessment at HL2 is a complex task which involves the two
aspects of system analysis and practical considerations in selecting a system
state. The required system analyses are load flow calculations, contingency
131
132 CHAPTER 5

analysis and ranking, rescheduling of generation, overload alleviation, and

load shedding philosophies, etc. These requirements are the same for both
analytical enumeration and Monte Carlo simulation. Selection of a system
state involves many practical considerations, such as derated states of gener-
ating units, common-cause outages of transmission lines, regional weather
effects, bus load uncertainty and correlation, reservoir operating rules,
dependent failures, and even complex operational considerations, etc. Obvi-
ously, these considerations are different for analytical enumeration and
Monte Carlo simulation. Generally, Monte Carlo methods are more flexible
when complex operating conditions and system considerations need to be
incorporated. It should be noted, however, that many issues associated with
system considerations have not yet been fully resolved. The important aspect,
therefore, is for a particular electric power utility to decide which factors
and effects are important in their case and then to use the most suitable
technique.
A system state sampling method for composite system adequacy evalua-
tion and basic system analysis techniques are presented in Section 5.2. Sev-
eral practical considerations in selecting system states and their effects on
composite system adequacy are discussed in Sections 5.3 to 5.7. The system
state sampling method cannot be used to directly calculate the frequency
index and only provides the Expected Number of Load Curtailments, which
is an approximation of the frequency index. Application of the system state
transition sampling technique to calculate the frequency index is described
in Section 5.8.
Two special issues are described in Sections 5.9 and 5.10. The first issue
is that of security considerations in composite system adequacy assessment.
The term "security" in this case is associated with steady state analysis and
introduces the concept of security constrained adequacy assessment. It is
therefore different from "security evaluation" as discussed in Section 2.1,
which is associated with system dynamic disturbances. The second issue is
noncoherence in composite system reliability evaluation.

5.2. SYSTEM STATE SAMPLING METHOD

AND SYSTEM ANALYSIS TECHNIQUES
The three basic simulation methods in reliability evaluation are
described in Section 3.6. In this section, the system state sampling method
is applied to composite system adequacy assessment. The most important
advantage of this sampling method is reduced computing time and memory
requirements. The number of components in composite system assessment
COMPOSITE SYSTEM ADEQUACY ASSESSMENT 133

is much larger than that in generation system adequacy assessment. Morever,

the system analysis associated with load flow calculation, contingency ana-
lysis, generation rescheduling, and overload alleviation requires a consider-
able amount of calculation.

5.2.1. Basic Methodology

The basic sampling procedure can be conducted by assuming that the
behavior of each component can be catagorized by a uniform distribution
under [0, 1]. In the case of a two-state component representation, the prob-
ability of outage is the component forced unavailability. It is also assumed
that component outages are independent events.
Let Sj denote the state of the ith component and FUj be its forced
unavailability. A random number generation method, such as the improved
prime number multiplicative congruential method, is used to draw a random
number Uj distributed uniformly under [0, 1]:

s.={01
I
(up state)
(down state)
if Uj~FUj
ifO~Uj<FUj
(5.1)

The state of the system containing t components including generating

units, transmission lines/transformers, etc., is expressed by the vector S:

S=(S), S2,"" Sj, ... , St)

When S equals zero, the system is in the normal state. When S is not equal
to zero, the system is in a contingency state due to component outage(s).
Assuming that each system state has the probability peS) and the index
function F(S), the mathematical expectation of the index function of all
system states is given by
E(F) = L F(S)P(S) (5.2)
SEG

where G is the set of system states.

Substituting the frequency of the state S in sampling for its probability
peS) yields

E(F) = L F(S) n(S) (5.3)

SEG N
where N is the number of samples and n(S) is the number of occurrences
of state S.
134 CHAPTER 5

Load curtailment occurs in relatively few system states and therefore it

is necessary to calculate E(F) only for these states. The set of system states
G can be divided into four subsets G1, G2, G3, and G4 and displayed as
follows:

where G1 is the normal state subset (no load curtailment), G1 is the contin-
gency state subset, G2 is the subset composed of the contingency states which
definitely have no load curtailment, G2 is the subset composed ofthe contin-
gency states which may have load curtailment, G3 is the subset composed
of the contingency states which have no load curtailment after rescheduling
the generation, and G4 is the subset composed of the contingency states
which still have load curtailment after rescheduling the generation.
After a system state is obtained by Monte Carlo sampling, it is first
necessary to judge whether or not this state belongs to G1. If the state
belongs to G1, it is necessary to determine by means of contingency analysis
whether or not it belongs to G2. If the state belongs to G2, it is necessary
to reschedule the generation and determine the curtailed load using the
minimization model of load curtailment. If the resulting curtailed load is
zero, the state belongs to G3, and if the resulting curtailed load is not zero,
then the state belongs to G4. Only those system states belonging to subset
G4 contribute to the unreliability indices.

5.2.2. Contingency Analysis For Composite

Systems
There are many contingency analysis techniques for composite systems.
A linearized load flow based contingency analysis approach is introduced
in the following. It matches the minimization model of load curtailment
described in the next subsection in computational accuracy. A more accurate
AC load flow based contingency analysis method is given in Appendix C.4.
Sampling all the system component states results in a specified system
state. If this is a normal state, then no load curtailment exists. If this is a
contingency state, load curtailment may be required. In an actual power
system, however, the possibility of load curtailment due to low-level contin-
gencies and the likelihood of high-level contingencies are relatively rare,
COMPOSITE SYSTEM ADEQUACY ASSESSMENT 135

therefore most of the drawn contingency states belong to G2 with only a

few belonging to G2. It is possible to judge whether or not a system state
belongs to G2 by means of a simple calculation instead of solving the load
flow equations. Once it has been determined that a state belongs to G2, it
makes no contribution to the unreliability indices and it is possible to proceed
immediately to the next sampling. It is necessary to reschedule generation
only for those states belonging to G2 and, if unavoidable, to shed loads by
solving the minimization model of load curtailment.
A load flow solution of the normal system state can be obtained which
includes generation outputs and line power flows. The direct analysis of
possible load curtailment for any drawn contingency state can be conducted
as follows:
(a) Generating Unit Contingencies. In general, it can be
assumed that a generator bus i has several generators denoted by set N;,
with the upper limit of each generator output being Pk (keN;) and its actual
power output in the normal state being Pile. When a generating unit is in a
contingency state, the corresponding Pk becomes P'tk where P'tk is the derated
output limit for a generator in a derated state and zero for a generator in
the full forced outage state. (Sampling and effects of generating unit derated
states are discussed in Section 5.4.) If the spare generation capacity at each
generator bus can compensate for the unavailable capacity at this bus, i.e.,
for all generator buses,
L (P'tk-P;k)~O (5.4)
keN,

then the generating unit contingencies lead to no load curtailment. In this

case, it is necessary to examine whether or not transmission component
outages bring about line overloads. If equation (5.4) is not satisfied for at
least one generator bus, then the system contingency state belongs to G2
and it is necessary to solve the minimization model to determine if the state
is in subset G3 or G4.
(b) Transmission Component Outages. If the redistribution of
load flow due to the removal of lines creates no overload in other lines under
the condition that equation (5.4) for all generator buses is satisfied, then the
contingency state belongs to G2. If this is not the case, then it belongs to
G2. In order to make this discrimination, it is therefore necessary to calculate
line flows following the removal of transmission components.
The following relationship can be obtained using the linearized load
flow formulation:
(5.5)
136 CHAPTER 5

Here, PG and PO are the generation output vector and the load power
vector, respectively; Si is the system state vector where, when j=O, SO
denotes the normal state and, whenj= 1, 2, ... , Sl, S2, ... ,denote the con-
tingency states due to single, double, ... , transmission component outages;
T(Si) are the line flow vectors under state Si; A(Si) is the relation matrix
between line flows and power injections under state Si. The mth row of
A(Si) can be calculated as follows:

Am(Si) Zr(Si);mZiSi ) (5.6)

where rand q denote the two bus numbers of the mth line; Xm is the reactance
of the mth line; Z(Si) is the bus impedance matrix of the system under state
Si in which the resistances of all lines are neglected; Zr(Si) and Zq(Si) are
the rth and qth rows of Z(Si), respectively.
The bus impedance matrices after the removal of specified lines can be
calculated directly from the bus impedance matrix of the normal state:
(5.7)
where
(5.8)
Here, X is a diagonal matrix whose dimension is the same as the number of
the outaged lines and whose diagonal elements are the reactances of the
outage circuits; M is a submatrix composed of the columns corresponding
to the outage lines of the bus-line incidence matrix.
The line flows of any contingency state involving single or multiple line
outage events can be calculated from Z(So) of the normal state and bus
power injections using equations (5.5)-(5.8). When equation (5.4) is satis-
fied, the power injections at all buses remain constant and the following
simple formulas can be derived to calculate the line flows for single line
outage states(3l):

(5.9)

(m=Fk) (5.10)

I: _ (1- BkD kk)l1bk

Jkk- (5.11)
(1 + I1bkD kk )Bk
D mk =(Mm)Tz(sO)Mk (5.12)
D kk =(Mk)TZ(So)M k (5.13)
COMPOSITE SYSTEM ADEQUACY ASSESSMENT 137

Here, k denotes any outage line and m any line in the system; Tm(So),
Tk(So), and Tm(SI) are the power flows in lines m and k under the normal
and the single transmission component outage states, respectively; Bm and
Bk are the mutual admittances of line m and line k, respectively; Abk is the
change value of Bk after removal of one or several circuits in line k; M m
and Mk are the column vectors of M corresponding to, respectively, line m
and line k.
The line power flows under single transmission component outage states
can be calculated directly using equations (5.9)-(5.13) from the matrix Z(So)
and the line power flows T(So) of the normal state. The likelihood of single
transmission component outages is much greater than those of multiple
transmission component outages. This set of formulas therefore create a
considerable decrease in the computational requirements. It should be noted
that equations (5.9)-(5.13) are obtained under the condition that the bus
loads remain unchanged. If the method of sampling load states is used, the
load level under a contingency state is determined randomly and therefore
these equations do not apply.

5.2.3. Linear Programming Optimization

Model
A considerable number of contingency states in which no load curtail-
ment exists will have been excluded after the contingency analysis of all the
drawn contingency states. For those contingency states belonging to G2, in
which generation outputs at some buses cannot be maintained due to gener-
ating unit contingencies (outage or derated) and/or there exist some line
overloads due to transmission component outages, generation outputs
should be rescheduled to maintain generation-demand balance and alleviate
line overloads and, at the same time, to avoid load curtailment if possible
or to minimize total load curtailment if unavoidable. The following minimi-
zation model of load curtailment can be used for this purpose:

min I Ci (5.14)
ieNC

subject to
T(Si) =A(Si)(PG+ c- PD) (5.15)

I PGi+ I Ci= I PDi (5.16)

ieNG ieNC ieNC

PGmin :::;; PG :::;;PGmax (5.17)

138 CHAPTER 5

O~C~PO (5.18)
IT(Sj)1 ~ T max (5.19)
where T(Sj), A(Sj), PG, PO are as defined in the previous subsection;
PGmin , PGmax , and T max are the limit vectors, respectively, of PG and T(Sj);
C is the load curtailment vector; NC and NG are the sets of all load buses
and all generator buses, respectively. The objective of this model is to mini-
mize the total load curtailment while satisfying the power balance, the lin-
earized load flow relationships, and the limits of line power flows and
generation outputs.
Linear programming theory and computational experience indicates
that the above model has multisolutions and therefore cannot be used to
calculate realistic bus indices. An acceptable load curtailment philosophy
should be incorporated in the minimization model. There is a wide range of
possible philosophies which can be utilized. The following are two basic
philosophies:
1. Loads are curtailed at buses which are as close to the elements on
outage(s) as possible.
2. Loads are classified according to their importance. The least impor-
tant load should be curtailed first, then the next least important, and
at the last the most important load.
These two load curtailment philosophies can be incorporated by modi-
fying the above minimization model to create the following model:

(5.20)

(i= 1, ... ,L) (5.21)

MS
~ PGi + ~ ~ CIj= ~ POi (5.22)
ieNG ieNC j=! ieNC

(5.23)

(ieNC;j= 1, ... , MS) (5.24)

(5.25)
where Land NS are the numbers of the lines and buses, respectively, and
other notations are as defined earlier. In the modified model, the load
curtailment variable at each bus is divided into MS subvariables (in general,
COMPOSITE SYSTEM ADEQUACY ASSESSMENT 139

MS = 2 or 3), while ai are the load percentages associated with each subvari-
able and Pi are the weighting factors. The least important load corresponds
to the smallest Pi and the most important load to the largest Pi. Introduction
of subvariables and their weighting factors allows load curtailment philos-
ophy 2 to be realized automatically in the resolution of the model.
Quantities W; are the weighting factors associated with each bus load.
The buses closest to the elements on outage(s) have relatively small Wi
and those far from outage(s) have relatively large Wi. The values of Wi
for each bus load are variable and depend on outage location(s) in the
particular contingency system state. Introduction of W; allows load curtail-
ment philosophy 1 to be realized automatically in the resolution of the
model.
It should be noted that the values of Pi or Wi need only to be specified
in terms of their relative magnitudes and therefore it is easy to select these
values. The selection of one or both will depend on the actual philosophy
used by the system. The model is solved by combining the linear program-
ming relaxation technique with the dual simplex method. (32) These two
approaches are given in Appendix 0.1. Only the active line power constraints
are introduced into equations (5.21) and (5.25), creating a small-scale linear
programming problem which can be solved quite quickly.

5.2.4. Basic Case Studies

In this subsection, the IEEE RTS(33) is used to conduct basic case
studies. The single line diagram of the system and the line and generator
data are given in Appendix A.I. Annualized system and bus indices are first
illustrated. Annualized indices are those calculated at the annual system
peak load level and expressed on a base of one year. Annual indices consider-
ing the annual load curve and results associated with this and other factors
are discussed in Sections 5.3 to 5.7. All the studies in this chapter were
conducted on a VAX-6330 computer except for the cases indicated. The
following adequacy indic~s are calculated. A comprehensive interpretation
of these indices is provided in Section 2.6.2.

ENLC Expected Number of Load Curtailments (occ.jyr)

EDLC Expected Duration of Load Curtailments (hrjyr)
PLC Probability of Load Curtailments
EONS Expected Demand Not Supplied (MW)
EENS Expected Energy Not Supplied (MWhjyr)
BPII Bulk Power Interruption Index (MW jMW-yr) _
BPECI Bulk Power/Energy Curtailment Index (MWhjMW-yr)
140 CHAPTER 5

BPACI Bulk Power-supply Average MW Curtailment Index (MW /

disturbance)
MBPCI Modified Bulk Power Curtailment Index (MW/MW)
SI Severity Index (system minutes/yr)

(a) Comparison Between the Monte Carlo and Enumeration

Methods. The annualized system indices of the IEEE RTS considering
two-state models for all generating units using Monte Carlo simulation and
state enumeration are given in Table 5.1. In the case of the enumeration
method, line outages and combined generating unit and line outages have
been considered up to the 3rd level while generating unit outages have been
considered up to the 4th and 5th levels, respectively.
The results in Table 5.1 indicate that in the case of the Monte Carlo
simulation method, 10,000 samples are sufficient for this system. The results
obtained from 10,000 samples have only about 6% differences compared to
those obtained from 100,000 samples and approach those obtained from the
enumeration method when line outages are considered up to the 3rd level
and generating unit outages up to the 5th level. The coefficient of variation
for the EDNS is 0.04 in the case of 10,000 samples. The CPU times listed
indicate that in order to obtain basically similar results, the Monte Carlo
simulation method requires much less computing time when compared to
the state enumeration method for the IEEE R TS. The state enumeration
approach used in this comparison was a relatively unsophisticated technique
which did not involve sorting generating unit outages. (34) It was included to
provide a direct comparison with the Monte Carlo approach.

Table 5.1. Annualized System Indices for the IEEE RTS Using Monte
Carlo and Enumeration Methods
Monte Carlo simulation State enumeration
Number of samples Generator outage level
Name of index 10,000 100,000 Up to 4th Up to 5th
ENLC 54.75342 57.92681 47.96653 54.21938
EOLC 698.88000 737.23102 657.78259 711.28967
PLC 0.08000 0.08439 0.07530 0.08142
EONS 13.97045 14.87208 11.99383 13.76009
EENS 122045.88281 129922.46875 104778.14063 120208.11719
BPII 3.33652 3.64042 2.61351 3.21786
BPECI 42.82311 45.58683 36.76426 42.17828
PBACI 173.67123 179.10896 155.28572 169.14433
MBPCI 0.00490 0.00522 0.00421 0.00483
SI 2569.38672 2735.20947 2205.85522 2530.69702
CPU time(min) 0.623 6.55 18.3 53.3
COMPOSITE SYSTEM ADEQUACY ASSESSMENT 141

Table 5.2. Annualized Bus Indices for the IEEE RTS

Bus No. PLC ENLC EDNS EENS
1 0.00200 1.46178 0.04144 361.98926
2 0.00220 1.57351 0.03966 346.47028
3 0.00240 1.88695 0.08150 711.98462
4 0.00280 2.04188 0.03983 347.95593
5 0.00390 2.66570 0.04306 376.17392
6 0.00530 3.75610 0.11766 1027.87903
7 0.00530 3.75610 0.02000 174.72641
8 0.00850 5.81365 0.23704 2070.79102
9 0.01140 7.98625 0.33950 2965.88159
10 0.01300 9.24670 0.47850 4180.18311
13 0.01980 13.91388 0.85216 7444.49316
14 0.02990 20.12966 0.96918 8466.77441
15 0.04090 26.91882 2.35796 20599.17383
16 0.04340 28.79770 0.85570 7475.41650
18 0.05480 37.99242 3.19185 27884.02734
19 0.07100 48.06261 2.30390 20126.89258
20 0.07980 54.56686 2.00151 17485.18555

(b) Annualized Bus Indices. The annualized bus indices of the

IEEE RTS using the two-state model for all generating units are given in
Table 5.2. Each load bus experiences load curtailments. This set of bus
indices is based on the following load curtailment philosophy: The bus loads
are divided into three groups. The first group corresponds to the top 20%
of the bus loads, the second group to the next 20%, and the third group to
the remaining 60010. When load curtailments are unavoidable, the loads in
the first group are curtailed first, then those in the second group, and at the
last those in the third group. This is conducted automatically by specifying
different weighting factors f3j in the objective function (5.20). The IEEE RTS
has been also evaluated using the model without a specific load curtailment
philosophy [equations (5.14) to (5.19)]. Several different solutions were
obtained by changing the order of the non basic variables entering the base.
This means that under the condition that all constraints are satisfied, there
is only one set of system indices but there are several sets of different bus
indices. Two sets of different bus indices obtained using the model without
a specific load curtailment philosophy are given in Tables 5.3 and 5.4. The
indices for the buses not listed in Tables 5.3 and 5.4 are zero. The system
indices associated with these two solutions are the same as those obtained
using the model with the specific load curtailment philosophy. It can be seen
that the bus inadequacies indicated by these two sets of bus indices are
contradictory. Therefore, the model without a specific load curtailment
philosophy cannot be used to calculate realistic bus indices.
142 CHAPTER 5

Table 5.3. The First Solution for Annualized Bus

Indices in the IEEE RTS Using the Model without
a Specific Load Curtailment Philosophy
Bus No. PLC ENLC EDNS EENS
3 0.00010 0.15839 0.00431 37.64702
10 0.00010 0.02817 0.01255 109.65305
15 0.00020 0.12536 0.00580 50.66939
16 0.00040 0.31162 0.02100 183.45644
18 0.01290 9.21853 1.44220 12599.09668
19 0.04340 28.79770 5.10252 44575.58984
20 0.07980 54.56686 7.37742 64449.15234

Table 5.4. The Second Solution for Annualized

Bus Indices in the IEEE RTS Using the Model
without a Specific Load Curtailment Philosophy
Bus No. PLC ENLC EDNS EENS
3 0.00010 0.15839 0.00431 37.64702
9 0.00020 0.12536 0.02340 204.42296
10 0.00280 2.04188 0.30535 2667.56055
13 0.03760 23.91431 3.59061 31367.58008
14 0.07960 54.26879 10.01209 87465.61719
15 0.00020 0.29807 0.03000 262.08063

5.3. INCORPORATION OF THE ANNUAL

LOAD CURVE
There are many practical factors to be considered in composite system
adequacy assessment. This section discusses the incorporation and resulting
effects of the annual load curve model. Generating unit derated states,
regional weather effects, common cause outages of transmission lines, and
bus load uncertainty and correlation are discussed in subsequent sections.

5.3.1. Multistep Model of the Annual Load

Curve
Both annualized and annual indices can be used to assess the adequacy
of a composite system. Annualized indices are calculated at a single load
level (normally at the system peak load level) and expressed on a base of
one year. Annual indices are calculated considering the variable load levels
throughout the year. Annualized indices can be employed in a wide range
COMPOSITE SYSTEM ADEQUACY ASSESSMENT 143

of studies, such as comparing the relative adequacies of different systems or

changes in system configuration or composition. Annual indices, however,
can be utilized for more comprehensive purposes, for instance, when reli-
ability worth is considered in power system planning. Calculation of annual
indices requires recognition of the various load levels that can occur during
a year. This can be done by a simple sampling of the chronological load
states or by utilizing a load model which eliminates the chronology and
aggregates the load states. Three possible methods for modeling an annual
load curve are described in Section 4.3.2. The cluster technique for a multi-
step load model is also given in Section 4.3.3. The utilization of a multistep
model of the annual load duration curve in a Monte Carlo simulation of
composite system adequacy evaluation is more effective from a computa-
tional point of view than individual sampling of the chronological load
states. This is mainly due to the following:
1. As noted in Section 4.3.2, combining Monte Carlo simulation of
system component states with enumeration of load level steps is a variance
reduction technique similar to stratified sampling. Many of the load points
in the flat segments of a load duration curve make almost the same contribu-
tion to the total indices. This is particularly true of the low load points,
which have relatively high probabilities of occurrence.
2. For a specific load level, the direct contingency analysis approach
described in Section 5.2.2 can be used to reduce the required computation
time.
3. The same system contingency state may be drawn many times in a
Monte Carlo simulation for system component outages. In these situations,
the system contingency state needs to be evaluated only once for a particular
load level but many times when sampling chronological load states.

5.3.2. Ratios of Generation-Transmission

Adequacy Indices
The accuracy of the load representation increases with the utilization
of more steps in the annual load duration curve. The computation time also
increases with the number of steps in the load model. The adequacy indices
for different composite systems, however, can have quite different sensitivi-
ties to the number of load level steps. If the adequacy indices of a composite
system are very sensitive to the load levels, then the number of load level
steps can have considerable impact on the annual indices. A set of ratios
of generation-transmission adequacy indices can be used to quantitatively
recognize the sensitivity of composite system adequacy indices to the number
of steps in the multistep load model.
144 CHAPTER 5

The adequacy indices of a composite generation and transmission sys-

tem depend on two basic aspects: the strength or weakness of the generation
system (reserve generation and forced unavailability of the generating units)
and the strength or weakness of the transmission system (transmission capac-
ity limits, forced unavailability of the transmission components, and the
system topology or configuration). The ratios of the contribution to the
composite system adequacy indices due to the generation system and the
contribution due to the transmission system can provide a general indication
of which segment is significant in regard to system adequacy. A set of Ratios
of Generation-Transmission Adequacy Indices (RGTAI) are defined as fol-
lows. In these definitions, the quantities without superscript * are the annual-
ized indices of a composite system at the peak load level while those with
* are the annualized contribution due to the generation system. The genera-
tion system contribution is determined at HLl, i.e., on a total generation-
total load basis.
(1) Ratio of Load Curtailment Probability (RLCP)

RLCP= PLC*
PLC-PLC*
(2) Ratio of the Number of Load Curtailment (RNLC)

RNLC= ENLC·
ENLC-ENLC*
(3) Ratio of Energy Not Supplied (RENS)

RENS EENS*
EENS-EENS*
(4) Ratio of Power-Interruption Index (RPII)

RPII= BPII*
BPII-BPII*
The definitions of indices PLC, ENLC, EENS, and BPII can be found
in Section 2.6.2. When the RGTAI values are greater than 1.0, the in-
adequacy of a composite system is largely caused by the generation system.
When the RGTAI values are smaller than 1.0, the inadequacy of the compos-
ite system is largely due to the transmission system. A basic requirement in
a composite system reliability study is to know if a composite system should
be reinforced and also to indicate which segment of it should be reinforced.
The RGTAI ratios provide useful indicators regarding both the need and
COMPOSITE SYSTEM ADEQUACY ASSESSMENT 145

the general functional areas of generation and transmission requiring

reinforcement.
The RGTAI values also quantitatively reflect the sensitivity of the calcu-
lated adequacy indices of a composite system to the load curve. The greater
the ratio values, the more sensitive to the load curve are the composite
system adequacy indices. This is due to the fact that load level variations
have a direct impact on the calculated adequacy indices of the installed
generating capacity but only a secondary effect on the transmission system
contribution to the composite system adequacy indices. This means that the
load level variations have much greater influence on the numerator values
of the ratios than the denominator values. In addition, load levels near the
annual peak load have relatively short durations while other lower load
levels have longer durations. It should be noted that the correlation between
the RGTAI values and the sensitivity of the composite system adequacy
indices to the load curve is not linear. The composite system adequacy
indices are quite sensitive to the load curve only when the RGTAI values
are much greater than 1.0.

5.3.3. Effect of the Number of Steps in a

Load Model
If the load duration curve is divided into a large number of steps, then
the calculated annual indices are closely representative of the actual load
model. This, however, can result in excessive computing time. It is only
necessary to use many steps in the load model to obtain accurate annual
indices in the case of a composite system which is sensitive to the load curve.
Fewer steps can be used in the case of a nonsensitive composite system with
an attendant decrease in required computing time.
In order to illustrate the difference between sensitive and nonsensitive
composite systems, two systems are considered. One is the IEEE RTS and
another is a model of the Saskatchewan Power Corporation System (SPCS).
The IEEE RTS has 24 buses (10 generator buses and 17 load buses), 38
lines/transformers, and 32 generating units. The annual system peak load
is 2850 MW and the total installed generating capacity is 3405 MW. The
SPCS has 45 buses (8 generator buses and 27 load buses), 71 lines/trans-
formers, and 29 generating units. The annual system peak load is 1802.5 MW
and the total installed generating capacity is 2530 MW. The single line dia-
gram of the SPCS and the line and generator data can be found in Reference
34. The two systems were assumed to have the same load profile. The indi-
vidual hourly load data are those of the IEEE R TS given in Appendix A.l.
146 CHAPTER 5

Table 5.5. Ratios of Generation-Transmission

Adequacy Indices for the Two Test Systems
RGTAI IEEE RTS SPCS
RLCP 265.67 0.31
RNLC 20.91 0.27
RENS 448.39 0.75
RPII 24.12 0.69

A load duration curve containing 8736 points can be created using these
data. The mean load of the 8736 load points is 61.44% of the annual system
peak load.
The RGTAI values for the IEEE R TS and the SPCS are shown in
Table 5.5. It can be concluded from these values that the IEEE RTS is very
sensitive to the number of steps in the load curve to which the SPCS model
is not sensitive. The annual indices for the two test systems have been calcula-
ted using three load models. In Modell, the load curve is divided into 70
steps with a step load increment of 1%. In Models 2 and 3, the load curve
is divided into 15 steps and 8 steps, respectively. The load increment of each
step is 5% and 10%, respectively. In each case, the final step includes all
load points lower than the load level of this step. The calculated annual
system indices are presented in Tables 5.6 and 5.7. The percentage values in
the brackets are the differences between the results for the IS-step and the
8-step model of the load duration curve compared with the results obtained

Table 5.6. Annual Indices Using the Three Load Models and Annualized
Indices at the Mean Load for the IEEE RTS
Annual indices
Annualized
70-step 15-step 8-step indices
Index model model model (mean load)
ENLC 0.80854 1.23661 (53%) 2.19343 (171%) 0.0
EDLC 10.23695 15.54475 (52%) 27.27785 (166%) 0.0
PLC 0.00117 0.00178 (52%) 0.00312 (167%) 0.0
EDNS 0.13137 0.21761 (66%) 0.42125 (221 %) 0.0
EENS 1147.61108 1901.03833 (66%) 3680.02637 (221%) 0.0
BPI! 0.03196 0.05286 (65%) 0.10255 (221%) 0.0
BPECI 0.40267 0.66703 (66%) 1.29124 (221 %) 0.0
BPACI 112.64307 121.81555 (8%) 133.25201 (18%) 0.0
MBPCI 0.00005 0.00008 (60%) 0.00015 (200%) 0.0
SI 24.16023 40.02186 (66%) 77.47424 (221 %) 0.0
CPU
(min) 15.97 4.48 2.78 0.29
COMPOSITE SYSTEM ADEQUACY ASSESSMENT 147

Table 5.7. Annual Indices Using the Three Load Models and Annualized
Indices at the Mean Load for the spes
Annual indices
Annualized
70-step IS-step 8-step indices
Index model model model (mean load)
ENLC 1.17838 1.21734(3%) 1.27841 (8%) 1.09548 (-7%)
EDLC 2.68576 2.84312 (6%) 2.98546 (11%) 2.27136 (-15%)
PLC 0.00031 0.00033 (6%) 0.00034 (10%) 0.00026 (-16%)
EDNS 0.01016 0.01077 (6%) 0.01173 (15%) 0.00920 (-9%)
EENS 88.72541 94.10423 (6%) 102.49805 (16%) 80.40615 (-9%)
BPII 0.02285 0.02400 (5%) 0.02570 (12%) .02151 (-6%)
BPECI 0.04922 0.05221 (6%) 0.05686 (16%) 0.04461 (-9%)
BPACI 34.95942 35.54316 (2%) 36.23811 (4%) 35.40000 (1%)
MBPCI 0.00001 0.0000 (0%) 0.00001 (0"10) 0.00001 (0"10)
SI 2.95344 3.13246 (6%) 3.41186 (16%) 2.67649 (-9%)
CPU
(min) 117.67 24.82 13.76 1.70

using the 70-step load model. The values in the last column are the annual-
ized indices calculated at the mean load level and the percentage differences
relative to the annual indices using the 70-step load model. The last row in
Tables 5.6 and 5.7 shows the CPU solution times.
It can be seen from Tables 5.6 and 5.7 that the number of steps in the
load duration curve has considerable influence on the calculated annual
indices for the IEEE R TS, which has very high RGTAI values. If the load
curve is divided into too few steps, the calculated annual indices can be quite
inaccurate for sensitive composite systems such as the IEEE R TS. This
system is so sensitive to the load level that the annualized indices at the
mean load level are effectively zero. The Saskatchewan Power Corporation
System which has low RGTAI values is not very sensitive to the number of
steps in the load duration curve. The 15-step approximate load model pro-
vides sufficiently accurate annual indices and even the 8-step approximate
load model provides relatively satisfactory results.
It is interesting to note that the annualized indices at the mean load
level for the SPCS, though optimistic, are reasonably clo..e to the calculated
annual indices obtained using the 70-step load model while the computing
times are much less. If computing time is limited or a problem, then it may
be acceptable in some system studies to replace annual indices by calculating
annualized indices at the mean load level. It can be seen from Tables 5.6
and 5.7 that this approach could be used in the SPCS but not in the IEEE
R TS. This approximation should be used with caution.
148 CHAPTER 5

Tables 5.8 and 5.9 show the contributions of each load level step to the
total annual indices using the IS-step load model for the two test systems.
These results further verify the previous analysis. In the case of the IEEE
R TS, the four load level steps higher than 80% of the annual peak load

Table 5.8. Contribution of Each Load Level Step to the Annual Indices
(in %) Using the 15-Step Load Model for the IEEE RTS
Load Load PLC EENS
level duration ENLC EOLC EONS BPII
(%) (%) (%) (%) (%) (%)
100.00 0.22 9.63 9.78 13.96 12.50
95.00 1.10 24.81 26.00 31.20 29.28
90.00 3.59 33.30 32.32 32.05 33.06
85.00 7.51 22.81 22.37 17.03 18.61
80.00 8.33 7.71 7.96 4.67 5.05
75.00 8.21 0.83 0.92 0.68 0.74
70.00 11.02 0.43 0.31 0.38 0.69
65.00 12.18 0.48 0.34 0.04 0.07
60.00 9.67 0.00 0.00 0.00 0.00
55.00 11.88 0.00 0.00 0.00 0.00
50.00 12.84 0.00 0.00 0.00 0.00
45.00 9.79 0.00 0.00 0.00 0.00
40.00 3.50 0.00 0.00 0.00 0.00
35.00 0.15 0.00 0.00 0.00 0.00
30.00 0.00 0.00 0.00 0.00 0.00

Table 5.9. Contribution of Each Load Level Step to the Annual Indices
(in %) Using the 15-Step Load Model for the SPCS
Load Load PLC EENS
level duration ENLC EOLC EONS PBII
(%) (%) (%) (%) (%) (%)
100.00 0.22 0.49 0.61 0.87 0.45
95.00 1.10 2.02 2.30 2.89 1.69
90.00 3.59 6.12 6.85 7.51 4.77
85.00 7.51 8.73 10.61 12.73 8.87
80.00 8.33 9.68 11.78 11.00 8.59
75.00 8.21 9.03 11.10 8.76 7.87
70.00 11.02 9.92 8.81 10.78 10.51
65.00 12.18 10.96 9.73 11.06 11.62
60.00 9.67 8.70 7.73 8.11 9.22
55.00 11.88 10.69 9.49 9.13 11.33
50.00 12.84 11.56 10.26 8.97 12.25
45.00 9.79 8.81 7.82 6.15 9.33
40.00 3.50 3.15 2.80 1.96 3.34
35.00 0.15 0.13 0.12 0.07 0.14
30.00 0.00 0.00 0.00 0.00 0.00
COMPOSITE SYSTEM ADEQUACY ASSESSMENT 149

contribute more than 90% of the annual indices. The duration of these four
steps is only 12.42% of the year. The load level steps lower than 65% of the
annual peak load make virtually no contribution to the calculated annual
indices. The duration covered by these steps is 47.83% of the year. In the
case of the spes, the contributions of each load level step to the total
calculated annual indices are generally related to the durations of each load
level step and do not directly depend on the load levels. All load level steps
higher than 30% of the annual peak load contribute to the total calculated
annual indices.

5.3.4. Effect of a Nonuniform Load

Increment Step Model
The effect of the high load levels in the IEEE R TS suggests that if the
multistep model of the load duration curve has relatively small load incre-
ments at the high load levels and relatively large load increments at the low
load levels, then the accuracy of the calculated annual indices will be
increased. The load curve was divided into 15 steps with a load increment
for each step of 2% for the first 10 steps and 5% for the following 4 steps.
The last step includes all load points lower than the load level of this step.
The annual indices of the two test systems were recalculated and are shown
in Table 5.10. The percentage values in brackets give the differences in the
results when using the I5-step nonuniform load increment model compared
with those obtained using the 70-step load model. It can be seen, by compar-

Table 5.10. Annual Indices of the Two

Test Systems Using the 15-Step Non-
uniform Load Increment Model of the
Load Curve
IEEE RTS SPCS
Index annual indices annual indices
ENLC 0.92368 (14%) 1.18735 (1%)
EOLC 11.70195 (14%) 2.73774 (2%)
PLC 0.00134(15%) 0.00031 (0010)
EONS 0.15568 (19%) 0.01104 (1%)
EENS 1360.02258 (19%) 96.47681 (9%)
BPII 0.03785 (18%) 0.02475 (8%)
BPECI 0.47720 (19%) 0.05352 (9%)
BPACI 116.79818 (4%) 37.56713 (7%)
MBPCI 0.00005 (00/.) 0.00001 (0010)
SI 28.63250 (19%) 3.21143 (9%)
150 CHAPTER 5

ing the results in Table 5.10 with those given in Tables 5.6 and 5.7 obtained
using the IS-step uniform load increment model, that using the nonuniform
load increment model considerably improves the accuracy of the calculated
annual indices for the IEEE R TS. It does not result in significant differences
in the SPCS study. The use of a nonuniform load model should therefore
be considered when analyzing a system with high RGTAI values.

5.4. GENERATING UNIT DERATED STATES

5.4.1. Sampling Multiderated States
Large-size generating units can operate in one or more derated states
and many electric power utilities are now using multistate models. In the
enumeration method, the recognition of derated states in the modeling of
large generating units can create a considerable increase in the number of
generation contingency states and therefore result in a significant increase
in overall solution time. The conventional approach in both generating
capacity and composite generation and transmission system reliability
studies is to use a two-state model for generating units and to replace the
forced unavailability by a derating-adjusted forced unavailability. It has
been shown in Chapter 4, however, that the use of a two-state equivalent
can result in a pessimistic appraisal of generating capacity adequacy. This
is also the case in composite system reliability assessment. One of the advan-
tages of the system state sampling method is that multistates of generating
units can be incorporated in the analysis without a significant increase in
computing time.
The state sampling technique including one derated state is given in
equation (4.12). This technique can be extended to the case of multiderated
states. Let Sj denote the state of the ith generating unit and FU j be its forced
unavailability. The probabilities of two derated states are expressed by PDl j
and PD7j. A uniformly distributed random number Uj is drawn in [0, I] for
the ith generating unit,

o (up state) if Uj~PDlj+ PD2 j + FUj

S; = {I (down state) ifPDl j+ PD2j~ Uj<PDl j+ PD2j+ FU j
(5.26)
2 (derated state 2) ifPDI;~ Uj<PDl j+ PD2 j
3 (derated state I) ifO~ Uj<PDl j

More derated states can be simulated similarly.

COMPOSITE SYSTEM ADEQUACY ASSESSMENT 151

5.4.2. Case Studies

In order to illustrate the effect of multiderated-state generating unit
representations, two test systems (IEEE R TS and RBTS) are examined. The
RBTS is a small educational test system developed at the University of
Saskatchewan. (35) The complete data for the two systems are given in Appen-
dix A. In the case of the IEEE RTS, the two 400-MW and the 350-MW
generating units have been given a derated state at 50010 capacity. The derat-
ing data are the same as those given in Tables 4.2 and 4.10. In the case of
the RBTS, the two 40-MW thermal units have been given a more detailed
state representation. Figure 5.1a shows a four-state representation, while
Figure 5.1b, c, and d provide three quasi-equivalent models. It has been
assumed that there are no transitions between the derated states and the
down state. In both systems, the state probabilities of the derated models
are such that the derating-adjusted two-state model data are identical to the
original two-state model data given in Appendix A.
The annualized system indices of the IEEE R TS considering the three-
state and the derating-adjusted two-state model for the 400-MW and 350-
MW generating units are given in Table 5.11. The percentage values in
parentheses are the differences in the indices using the derating-adjusted
two-state model as compared to those obtained using the three-state model.
It can be seen that the derating-adjusted two-state model leads to a pessimis-
tic evaluation of system adequacy.
The annualized system indices and bus indices of the RBTS considering
the two-state, the two three-state, and the four-state models for the 4O-MW
thermal generating units are given in Tables 5.12 and 5.13, respectively. The
percentage values are the differences in the indices using the two-state and
the two three-state models compared to those obtained using the four-state
model. The results in Tables 5.12 and 5.13 indicate that if the four-state
model for the 40-MW generating units is replaced by the three-state model
(b) in which the lower-capacity derated state is eliminated, then the results
are more optimistic (except for the average index BPACI). If the three-state
model (c), in which the larger-capacity derated state is eliminated, is used,
then the results are more pessimistic (except for BPACI). These replace-
ments, however, are unnecessary in the presented method because the com-
puting time using the four-state model is almost the same as that using the
three-state models. If the derating-adjusted two-state model for the 40-MW
generating units is used to replace the four-state model or the three-state
models, the results are considerably more pessimistic (except for BPACI),
although less computing time is needed. The quasi-equivalent models gen-
erally result in greater differences in bus indices than those obtained on a
system basis. The impact is quite different, however, at different buses.
 ....
~

20MW
Derated
PDE1 = 0.01
40MW 40MW
Up (a) Up
PUP = 0.93 PuP = 0.96

(b) 40MW (c)

Up
PUP = 0 97

OMW
Down o
PON =003 J:
»
(d) ~
m
::D
Figure 5.1. Two-, three-, and four-state models for a 4O-MW thermal generating unit in the RBTS. U1
COMPOSITE SYSTEM ADEQUACY ASSESSMENT 153

Table 5.11. Annualized System Indices

for the IEEE RTS
Index Three-state model Two-state model
ENLC 47.38287 54.75342 (16%)
EOLC 630.73920 698.88000 (11%)
PLC 0.07220 0.08000(11%)
EONS 10.32736 13.97045 (35%)
EENS 90219.83594 122045.88281 (35%)
BPH 2.49359 3.33652 (34%)
BPECI 31.65608 42.82311 (35%)
BPACI 149.98528 173.67123 (16%)
MBPCI 0.00362 0.00490 (35%)
SI 1899.36487 2569.38672 (35%)
CPU
(min) 0.740 0.623

Table 5.12. Annualized System Indices for the RBTS

Four-state Three-state Three-state Two-state
Index model (a) model (b) model (c) model (d)
ENLC 4.33496 4.11613 (-5%) 4.59188 (6%) 5.27252 (22%)
EOLC 76.87680 69.01440 (-10"10) 85.43808 (11%) 91.20384 (19%)
PLC 0.00880 0.00790 (-10"10) 0.00978 (11%) 0.01044 (19%)
EONS 0.09605 0.09015(-6%) 0.10189 (6%) 0.12429 (29%)
EENS 839.11414 787.57202 (-6%) 890.13239 (6%) 1085.81873 (29%)
BPH 0.30090 0.29236 (-3%) 0.31019 (3%) 0.38169 (27%)
BPECI 4.53575 4.25715 (-6%) 4.81153 (6%) 5.86929 (29%)
BPACI 12.84121 13.13995 (+2%) 12.49707 (-3%) 13.39275 (4%)
MBPCI 0.00052 0.00049 (-6%) 0.00055 (6%) 0.00067 (29%)
SI 272.14511 255.42876 (-6%) 288.69156 (6%) 352.15738 (29%)
CPU
(min) 0.326 0.352 0.307 0.187

5.5. REGIONAL WEATHER EFFECTS

5.5.1. General Concepts

The failure rate of an outdoor component such as a transmission
line is a function of the weather environment to which the line is exposed.
The failure rate of a transmission line can be much higher in adverse
weather than in normal weather, and if related transmission components
154 CHAPTER 5

Table 5.13. Annualized Bus Indices for the RBTS

Bus Four-state Three-state Three-state Two-state
Index No. model (a) model (b) model (c) model (d)
PLC 2 0.00012 0.00012 (0%) 0.00016 (33%) 0.00024 (100%)
3 0.00152 0.00142 (-7%) 0.00152 (0%) 0.00246 (62%)
4 0.00162 0.00162 (0%) 0.00166 (3%) 0.00242 (49%)
5 0.00716 0.00614 (-14%) 0.00832 (16%) 0.00898 (25%)
6 0.00870 0.00780 (-10%) 0.00968 (11%) 0.01034 (19%)
ENLC 2 0.06915 0.06915 (0%) 0.08805 (27%) 0.14550 (110%)
3 0.82640 0.78659 (-5%) 0.82422 (0%) 1.26890 (54%)
4 0.71569 0.72192 (1%) 0.74030 (3%) 1.11126 (55%)
5 2.82644 2.56493 (-9%) 3.15742 (12%) 3.82090 (35%)
6 4.22666 4.00783 (-5%) 4.48357 (6%) 5.16381 (22%)
EDNS 2 0.00048 0.00048 (0%) 0.00064 (33%) 0.00080 (66%)
3 0.01656 0.01582 (-5%) 0.01640 (-1%) 0.02618 (58%)
4 0.01242 0.01208 (-3%) 0.01252 (1%) 0.01940 (56%)
5 0.01250 O.oI 148 (-8%) 0.01394 (12%) 0.01688 (35%)
6 0.05409 0.05029 (-7%) 0.05839 (8%) 0.06103 (13%)
EENS 2 4.19328 4.19328 (0%) 5.59104 (33%) 6.98879 (67%)
3 144.66393 138.19930 (-5%) 143.26617 (-1%) 228.70407 (58%)
4 108.50108 105.53084 (-3%) 109.37469 (1%) 169.47839 (56%)
5 109.19910 100.28857 (-8%) 121.77868 (12%) 147.46265 (35%)
6 472.55692 439.36014 (-7%) 510.12170 (8%) 533.18488 (13%)

are exposed to the same adverse weather condition, then the phenomenon
known as adverse weather failure bunching can occur.(3) It is therefore
necessary to recognize these effects in order to incorporate weather consid-
erations in the transmission system model for composite system adequacy
evaluation. Markov state weather models, (36,37) have been developed, which
basically assume that the entire transmission system is exposed to the
same weather conditions. This obviously may not be valid for an actual
transmission system and particularly one spread out over a large geograph-
ical area. The exponential distribution assumptions required to create a
stationary Markov process also may not be valid in connection with
adverse weather modeling and can become unduly restrictive. The Monte
Carlo based method presented here can recognize regional weather aspects
including those situations in which transmission lines traverse several
geographical regions encompassing different weather conditions. Exponen-
tial distribution assumptions of weather durations are not required in
the proposed method. This therefore relaxes the traditional assumption
associated with a Markov process. The method can be used to analyze
practical large-scale power systems.
COMPOSITE SYSTEM ADEQUACY ASSESSMENT 155

~ ___U_p__
~(S)
~I~. ____:_: _:__~·~1___ D_O_W_N__
~(S)
~
Figure 5.2. Transmission line model considering weather conditions.

5.5.2. Basic Transmission Line Model

Recognizing Weather Conditions
This transmission line model is shown in Figure 5.2. This is not the
conventional two-state model because the failure rate, repair rate, and prob-
abilities of residing in the up state and down state are functions of weather
states expressed by S in Figure 5.2. It is convenient in terms of data collection
to divide weather into two conditions: normal and adverse. It is, however,
possible that one segment of a transmission line is exposed to normal weather
while its other part is exposed to adverse weather. Therefore, a specific
transmission line can reside in multiple and quite different weather environ-
ments. The model shown in Figure 5.2 is a multiple-state model discrimi-
nated by weather states S. In this model, transitions between weather
conditions are not specifically shown. Failure bunching of multiple facilities
depends on the probability of adverse weather and the failure rates and
repair rates of those transmission lines in adverse weather, but not specifi-
cally on the transitions between weather conditions.
It should be appreciated that the weather and therefore the stress levels
that transmission facilities can encounter are in fact associated with a con-
tinuous distribution. The difficulties of collecting the required data are
enormous and therefore, it is reasonable in the first instance to recognize
only two steps, i.e., normal and adverse. In terms of the model, however,
there are no practical difficulties associated with incorporating as many steps
or stress levels as the data are capable of supporting. In regard to the state
representation, assume that S = 0 indicates the normal weather condition and
S = I indicates the adverse weather condition. A transmission line therefore is
associated with two failure rates 11,(0) and 11,(1) and two repair rates p(O)
and p(1). At the present time, most data collection schemes do not recognize
11,(0) and 11,(1) but are only responsive to the average value II, of the failure
rate expressed in failures per calendar year. Using the concept of expectation,
11,= (1- U)A(O) + UA(I) (5.27)
where U is the fraction of time that a line spends in adverse weather.
156 CHAPTER 5

The proportion of failures F occurring in adverse weather reflects the

degree of potential failure bunching. If quantities U and F are known, A(O)
and A( I) can be evaluated from:
A(O) =A(l- F)/(l- U) (5.28)
and
A(1)=AF/U (5.29)
In the following case studies, U and F are specified rather than A(O)
and A(l). The effects of different degrees of failure bunching on the system
behavior can be illustrated by considering different values of F between 0
and 1.

5.5.3. Sampling Regional Weather States

The physical area of a given transmission system can be divided into
several regions according to their geographical and meteorological character-
istics. Figure 5.3 shows a representative geographical five-region division.
Region-divided weather states can be recognized in the following ways: a
single region encounters adverse weather; two adjoint regions exist in
adverse weather; three adjoint regions encounter adverse weather; etc. In
the case of a five-region division, the possible maximum number of region-
divided weather states is sCo + SCI + sC2 + sC3 + sC4 + sCs = 32. The possible
maximum number of region-divided weather states increases considerably
as the number of regions increases. The geographical range covered by the
same weather environment is generally relatively large compared to the area

.....
~

(4) /
/ (5)

Figure 5.3. A geographic five-region division.

COMPOSITE SYSTEM ADEQUACY ASSESSMENT 157

P1
+
Pi PN-1 PN
I I I
~I I I
0.0 1.0
Figure 5.4. Sampling of region-divided weather states.

of the total transmission system, and therefore it is unnecessary to divide a

system into a great number of regions. In those cases in which the number
of regions is relatively large, many reSion-divided weather states do not exist
in practice. For example, in Figure 5.3, the probability that Regions (1) and
(5) are simultaneously exposed to adverse weather and Regions (2), (3), and
(4) to normal weather may be effectively zero.
Given the criteria for discriminating between normal and adverse
weather, the probabilities of each region-divided weather state can be
obtained from the available meteorological records. The number of regions
used may depend more on the existence of adequate data than on the actual
transmission systems. Sensitivity analysis of the system adequacy may assist
in deciding whether to pursue the collection of additional data. Assume that
there are N region-divided weather states whose probabilities are not zero
and that their probabilities are expressed by PI, P2 , ••• , P N • Let
Pi, .... , P N - I correspond to the states in which at least one region
undergoes adverse weather and P N to the state of all regions in normal
weather. Obviously, PN is much larger than the summation of all Pi (i=
1, ... ,N -1). The values of PI, ... , PN are set sequentially in interval [0, 1]
as shown in Figure 5.4. A uniformly distributed random number between
[0, 1] can be drawn. If the random number falls in the region corresponding
to Pi, the transmission system is in the ith region-divided weather state for
this sample. It can be seen that the sampling efficiency is the same regardless
of the number of region-divided weather states, i.e., a large number of region-
divided weather states does not materially affect the computational
requirements.

5.5.4. Determination of Transmission Line

Forced Unavailability and Repair Time
with Regional Weather Effects
After obtaining a sample of the region-divided weather states, the forced
unavailabilities FU and repair times r of transmission lines in this weather
state can be determined. If the entire transmission line is in the region
158 CHAPTER 5

exposed to adverse weather, its transition rates are .:\.(1) and Jl(I) and thus
FU=':\'(I)/[':\'(I) + p(1)] and,= I/Jl(I). If the entire transmission line is in
the region exposed to normal weather, its transition rates are .:\.(0) and Jl(O)
and thus FU = .:\.(0) I[ .:\.(0) + Jl (0)] and , = 11Jl (0). If the transmission line
traverses several regions undergoing different weather conditions, the follow-
ing method can be used to calculate its FU and ,.
Assume that a transmission line traverses Regions (1) and (2), where
Region (1) is in adverse weather and Region (2) in normal weather, and
that R is the percentage of the line length in Region (1). Considering that
the failure frequency of a line is approximately equal to its failure rate and
the two sections of the line in Regions (1) and (2) are in series, the FU of
the line is given by

(5.30)

where

FU 1 =A(I)R/[A(I)R+ Jl(I)] and FU2 =A(0)(1- R)/[(A(O)(1- R) + Jl(O)]

The contribution due to the FU 1 FU2 term is quite small and, in prac-
tice, it can be neglected in equation (5.30). The equivalent repair time of
the line is given by

,= )_A(-,-l,-)R_+_,....:...(O"":,,,)A(--....:O--,-)....:...(1_-_R--,-)
--....:'(,-,1
(5.31)
A( I)R + A(O)(1- R)

where ,(O)=I/p(O) and ,(I)=I/p(1).

The calculation is similar in those cases in which a transmission line
traverses more than two regions. After the FU and, of each line associated
with a weather state sample are determined, the basic evaluation method
can be utilized to assess the adequacy of the composite system.

5.5.5. Test System and Data

In order to illustrate the basic procedure, case studies were conducted
using the educational test system (RBTS). In addition to the basic system
data given in Appendix A.2, the data required to incorporate weather effects
using the regional division shown in Figure 5.5 are as follows.
The overall area encompassed by the transmission system is divided
into the three regions shown in Figure 5.5. Lines 2 and 3 are in Region (1)
and Lines 6, 7, 8, and 9 are in Region (3). 30% of Line 1 is in Region (1)
COMPOSITE SYSTEM ADEQUACY ASSESSMENT 159

L1
Bus 1...,.+..,I;;-l--=..::.....----;;J,--l~,. Bus 2

Figure 5.5. Geographic three-region division for the RBTS.

and 70% in Region (2). 20% of Lines 4 and 5 are in Region (3) and 80% in
Region (2). The number of region-divided weather states is 8. The weather-
related probabilities associated with these regions are given in Table 5.14 in
which (0) denotes the state of three regions in normal weather, (1) or (2)
or (3) the state of only one region in adverse weather, (1) + (2) or (1) + (3)

Table 5.14. Region-Divided

Weather-State Probabilities
Weather state Probabilities
(0) 0.968
(I) 0.004
(2) 0.004
(3) 0.004
(1)+(2) 0.006
(1)+(3) 0.006
(2) + (3) 0.006
(1)+(2)+(3) 0.002
Total 1.000
160 CHAPTER 5

or (2) + (3) the state of two regions in adverse weather, and (1) + (2) + (3)
the state of all three regions in adverse weather.
The percentage U of the adverse weather duration for each line can be
calculated from the probabilities of region-divided weather states and the
regions traversed by transmission lines. Generally, the values of Ufor differ-
ent lines may differ. In the case of the data given above, however, all lines
have the same U value, i.e., O.oI8. Repair is generally more difficult during
adverse weather than during normal weather and, in many cases, repair
cannot proceed or commence until an adverse weather period ends. It has
therefore been arbitrarily assumed that the expected repair time of a line in
adverse weather is 1.5 times that in normal weather.

5.5.6. Case Studies

(a) Comparison between Different Weather-Related
Representations. The effect on the adequacy indices of different weather-
related representations can clearly be seen in the following analysis. In these
studies, the proportion of failures in adverse weather, F, is assumed to be
0.4. Given the values of F and U, values of .t(0) and .t(1) can be calculated
from the values of.t given in the basic data in term of equations (5.28) and
(5.29). A comparison between the results obtained considering weather
effects and those obtained without considering weather effects clearly illus-
trates their contributions. Two distinct weather models were considered: in
Case 1, regional weather differences are recognized using the data of Table
5.14 and, in Case 2, it is assumed that the entire transmission system is
exposed to common weather conditions. In both cases, the probabilities of
the entire transmission system being in the normal weather environment are
the same. The annualized and annual system and bus indices for the RBTS
in the three cases are given in Tables 5.15 to 5.18. The percentage values in
parentheses are the differences in the indices considering weather effects
compared to those without weather effects.
It can be seen that the basic two-state model which does not recognize
weather effects can lead to underestimation of the inadequacy indices,
especially for the annual indices of some buses. The assumption that the
entire transmission system is exposed to a common weather condition can
lead to overestimation of the failure bunching effect. Weather effects are
generally larger for the annual indices than for the annualized values. This
is basically due to the fact that generation outages have a relatively large
impact on the system adequacy at higher load levels than at lower load
periods and are not affected by weather conditions. The impact of transmis-
sion components is not quite as variable and responds to the adverse weather
COMPOSITE SYSTEM ADEQUACY ASSESSMENT 161

Table 5.15. Annualized System Indices with and without

Weather Effects for the RBTS
Without
weather Recognition of Entire system
Index effects regional weather in common weather
ENLC 5.27252 6.20917 (18%) 8.40546 (59%)
EOLC 91.20384 96.44544 (6%) 106.40448 (17%)
PLC 0.01044 0.01104 (6%) 0.01218 (17%)
EONS 0.12429 0.13624 (10"10) 0.16476 (33%)
EENS 1085.81873 1190.15723 (10"10) 1439.33862 (33%)
BPII 0.38169 0.48665 (28%) 0.78044 (104%)
BPECI 5.86929 6.43328 (10"10) 7.78021 (33%)
MBPCI 0.00067 0.00074 (10"10) 0.00089 (33%)
SI 352.15738 385.99692 (10"10) 466.81250 (33%)
CPU
(min) 0.187 0.294 0.303

Table 5.16. Annualized Bus Indices with and without Weather

Effects for the RBTS
Without
Bus weather Recognition of Entire system
Index No. effects regional weather in common weather
PLC 2 0.00024 0.00024 (0"10) 0.00024 (0"10)
3 0.00246 0.00280 (14%) 0.00346 (41%)
4 0.00242 0.00246 (2%) 0.00270 (12%)
5 0.00898 0.00922 (3%) 0.00964 (8%)
6 0.01034 0.01086 (5%) 0.01196 (16%)
ENLC 2 0.14550 0.14823 (2%) 0.15264 (5%)
3 1.26890 1.82331 (44%) 3.06077 (141%)
4 1.11126 1.18205 (6%) 1.67099 (50"10)
5 3.82090 4.21072 (10"/0) 5.13438 (34%)
6 5.16381 5.98878 (16%) 8.06009 (56%)
EONS 2 0.00080 0.00080 (0"/0) 0.00080 (0"/0)
3 0.02618 0.03105 (19%) 0.04248 (62%)
4 0.01940 0.02004 (3%) 0.02323 (20"/0)
5 0.01688 0.01860 (10"10) 0.02128 (26%)
6 0.06103 0.06575 (8%) 0.07698 (26%)
EENS 2 6.98879 6.98879 (0"10) 6.98879 (0"10)
3 228.70407 271.23013 (19%) 371.06458 (62%)
4 169.47839 175.06944 (3%) 202.89746 (20"10)
5 147.46265 162.48865 (10"10) 185.90128 (26%)
6 533.18488 574.37982 (8%) 672.48651 (26%)
162 CHAPTER 5

Table 5.17. Annual System Indices with and without

Weather Effects for the R BTS
Without
weather Recognition of Entire system
Index effects regional weather in common weather
ENLC 1.27622 1.61910 (27%) 2.62248 (105%)
EDLC 12.35270 14.34242 (16%) 19.09538 (55%)
PLC 0.00141 0.00164 (16%) 0.00219 (55%)
EDNS O.oI 786 0.02128 (19%) 0.02953 (65%)
EENS 155.99113 185.91623 (19%) 257.94995 (65%)
BPII 0.08771 0.11599 (32%) 0.20105 (129%)
BPECI 0.84320 1.00495 (19%) 1.39432 (65%)
MBECI 0.00010 0.00012 (20%) 0.00016 (60%)
SI 50.59172 60.29716 (19%) 83.65944 (65%)

CPU (min) 2.75 4.39 4.42

Table 5.18. Annual Bus Indices with and without Weather

Effects for the R BTS
Without
Bus weather Recognition of Entire system
Index No. effects regional weather in common weather
PLC 2 0.00002 0.00002 (0%) 0.00002 (0%)
3 0.00007 0.00009 (29%) 0.00018 (157%)
4 0.00009 0.00010 (11%) 0.00017 (89%)
5 0.00017 0.00023 (35%) 0.00033 (94%)
6 0.00141 0.00163 (16%) 0.00213 (51%)
ENLC 2 0.01101 0.01093 (0%) 0.01098 (0%)
3 0.04219 0.08496 (101%) 0.26133 (519%)
4 0.04827 0.07341 (52%) 0.23894 (395%)
5 0.08410 0.18155 (116%) 0.38903 (363%)
6 1.26807 1.59510 (26%) 2.50759 (98%)
EDNS 2 0.00005 0.00005 (0%) 0.00005 (0%)
3 0.00046 0.00061 (33%) 0.00162 (252%)
4 0.00053 0.00062 (17%) 0.00153 (189%)
5 0.00055 0.00112 (104%) 0.00170 (209%)
6 0.01627 0.01889 (16%) 0.02462 (51%)
EENS 2 0.43561 0.43561 (0"10) 0.43561 (0%)
3 4.01239 5.31128 (32%) 14.16684 (253%)
4 4.66888 5.38681 (15%) 13.39328 (187%)
5 4.78188 9.79080 (105%) 14.87348 (211%)
6 142.09238 164.99171 (16%) 215.08070 (51%)
COMPOSITE SYSTEM ADEQUACY ASSESSMENT 163

failure bunching phenomenon. Weather impacts on the bus indices are not
the same at different load buses. At Bus 2, which is connected directly to
generators, the bus indices are not generally affected by transmission line
failures and therefore not by weather conditions. (The relatively insignificant
difference in ENLC is due to the fact that the repair time in adverse weather
is assumed to be larger than that in normal weather.) At those load buses
connected to generators through transmission lines, the bus indices are sensi-
tive to weather conditions but the degree of sensitivity varies from bus to
bus. At a very sensitive bus such as Bus 3, the bus indices obtained assuming
that the entire system is exposed to common weather conditions can be quite
pessimistic, particularly in the case of the annual bus indices.

(b) Effects of the Proportion of Failures in Adverse

Weather. The proportion of failures in adverse weather is not the same
in different geographical areas. The proportion reflects the degree of failure
bunching and therefore a large proportion value indicates that weather has
large effects on composite system inadequacy indices. The variations in annu-
alized and annual EENS system indices with the proportion of failures in
adverse weather Fare shown in Figures 5.6 and 5.7, respectively. The varia-
tions in other system indices are similar. The EENS values in the figures are

4.5
entire system in common weather
4.0 recognizing regional weather

-
:::i
.s
3.5

en
3.0
..
....."
..........
z '
w 2.5
w ..
...........
....
-~
". .........
2.0 '
.........,
1.5
--.......
-...--..-......-........-.~.......
1.0
0.0 0.2 0.4 0.6 0.8 1.0
proportion of failures in adverse weather (F)
Figure 5.6. Variations in the annualized EENS system index with the proportion of failures
in adverse weather.
164 CHAPTER 5

4.5
entire system in common weather ,
4.0 recognizing regional weather //
3.5 //
//
-
~
~
ci ....
3.0
C/) //
..-..-"
Z
w 2.5
w
..........
,/

2.0 ../
...............
.,'"
1.5 .......
..................
1.0
0.0 0.2 0.4 0.6 0.8 1.0
proportion of failures in adverse weather (F)
Figure 5.7. Variations in the annual EENS system index with the proportion of failures in
adverse weather.

expressed in per unit of the annualized or annual EENS obtained without

considering weather effects. The upper curve corresponds to the cases of the
entire system in common weather conditions and the lower one to the cases
of recognizing regional weather aspects. It can be seen that the proportion
of failures in adverse weather has larger effects on annual system indices
than on annualized system indices and a large value of F can result in
increased overestimation of inadequacy with the assumption that the entire
system is in a common weather condition.

5.6. TRANSMISSION LINE COMMON

CAUSE OUTAGES

5.6.1. General Concepts

A common cause outage is an event having an external cause with
multiple failure effects where the effects are not consequences of each other.
The most obvious example of a common cause outage is the failure of
a transmission tower supporting two or more transmission circuits. The
COMPOSITE SYSTEM ADEQUACY ASSESSMENT 165

probability of a common cause outage can be many times larger than that
of a similar event due to two or more simultaneous independent outages. The
effect of the common cause outage on bus reliability indices can therefore be
significant compared with the effect of second- and higher-order independent
outages.
It should be noted that a common cause outage and a common eviron-
ment outage are completely different concepts. The weather effects discussed
in the previous section are associated with a common environment outage
but not a common cause outage. Although a common weather condition
affects the failure rates of the components, the actual failure process of
overlapping outages still assumes the component failures to be independent
with the independent failure rates enhanced because of the common
environment.
Two of the most useful models for representing a second-order overlap-
ping failure event including common cause failures are shown in Figure 5.8
in which Ac represents the common cause failure rate. The difference between
these two models is that one has a single down state (Figure 5.8a) and the
other has two separate down states: one associated with independent fail-
ures, the other associated with common cause failures (Figure 5.8b).
In the system state sampling method, transmission line states are mod-
eled using two-state (up and down) random variables. Transmission line
common outages can be considered in such a way that a fictitious or phys-
ically existing third component, not included in the system, can outage both
of these components if it should fail. Transmission line common outages

1 1
1U 1U
2U 2U

2 2
10 1U 10 10 1U
2U 20 2U 20 20

10
20

(a) (b)
Figure 5.8. Common cause failure models.
166 CHAPTER 5

therefore can be simulated by a separate two-state random variable using

one of the two models shown in Figure 5.8. When this separate random
number falls into the segment in the interval [0,1] corresponding to the
failure probability of the third component, the common outage event is
considered to occur in that sampling.

5.6.2. Case Studies

The likelihood of multiple transmission line failures can be enhanced
by common cause failures. A common cause failure, as noted earlier, is a
fundamentally different event from that associated with weather-related fail-
ure bunching. It should be clearly appreciated, however, that the common
cause event may itself be weather related. If this is the case, suitable data
can be inserted in the modeling process to recognize the variable nature of
the common cause failure phenomenon. In order to illustrate the effects of
common cause failures and compare these effects with those of weather, the
RBTS was used to conduct the following studies. Line pair (2,3) and Line
pair (4,5) in Figure 5.5 are assumed to be susceptible to common cause
failures using the data given in Appendix A.2. The calculated results obtained
considering only common cause failures and considering both common cause
failures and weather effects are given in Tables 5.19 to 5.22. The percentage
values in parentheses are the differences in the indices compared to those
obtained without considering any specified effect (see "without weather
effects" columns in Tables 5.15 to 5.18).

Table 5.19. Annualized System Indices Considering Common Cause

Failures and Weather Effects for the RBTS
Only Common cause failures Common cause failures
common cause and recognizing and entire system
Index failures regional weather in common weather
ENLC 5.73303 (9%) 6.63230 (26%) 8.82859 (67%)
EDLC 93.30048 (2%) 98.36736 (8%) 108.32640 (19"10)
PLC 0.01068 (2%) 0.01126 (8%) 0.01240 (19"/0)
EDNS 0.12989 (5%) 0.14138 (14%) 0.16990 (37%)
EENS 1134.74438 (5%) 1235.06079 (14%) 1484.24219 (37%)
BPII 0.43915 (15%) 0.53981 (41%) 0.83360 (118%)
BPECI 6.13375 (5%) 6.67600 (14%) 8.02293 (37%)
MBPCI 0.00070 (5%) 0.00076 (13%) 0.00092 (37%)
SI 368.02518 (5%) 400.56024 (14%) 481.37582 (37%)
CPU
(min) 0.205 0.320 0.339
COMPOSITE SYSTEM ADEQUACY ASSESSMENT 167

Table 5.20. Annualized Bus Indices Considering Common Cause

Failures and Weather Effects for the RBTS
Only Common cause failures Common cause failures
Bus common cause and recognizing and entire system
Index No. failures regional weather in common weather
PLC 2 0.00024 (0%) 0.00024 (0%) 0.00024 (0"10)
3 0.00270 (10"10) 0.00302 (23%) 0.00368 (50"/0)
4 0.00242 (0"/0) 0.00246 (2%) 0.00270 (12%)
5 0.00922 (3%) 0.00944 (5%) 0.00986 (10"10)
6 0.01058 (2%) 0.01108 (7%) 0.01218 (18%)
ENLC 2 0.14886 (2%) 0.15159 (4%) 0.15600 (7%)
3 1.71941 (36%) 2.23644 (76%) 3.47390 (174%)
4 1.12462 (1%) 1.19541 (8%) 1.68435 (52%)
5 4.28141 (12%) 4.63384 (21%) 5.55750 (45%)
6 5.62432 (9%) 6.41191 (24%) 8.48322 (64%)
EONS 2 0.00080 (0"/0) 0.00080 (0"10) 0.00080 (0"10)
3 0.03026 (16%) 0.03479 (33%) 0.04622 (77%)
4 0.01940 (0"/0) 0.02004 (3%) 0.02323 (20"10)
5 0.01752 (4%) 0.01920 (14%) 0.02188 (30"/0)
6 0.06191 (1%) 0.06655 (9%) 0.07778 (27%)
EENS 2 6.98879 (0"/0) 6.98879 (0"10) 6.98879 (0"10)
3 264.34692 (16%) 303.90277 (33%) 403.73721 (77%)
4 169.47839 (0"/0) 175.06944 (3%) 202.89746 (20"10)
5 153.05426 (4%) 167.73074 (14%) 191.14337 (30"/0)
6 540.87604 (1%) 581.36859 (9%) 679.47522 (27%)

Table 5.21. Annual System Indices Considering Common Cause

Failures and Weather Effects for the RBTS
Only Common cause failures Common cause failures
common cause and recognizing and entire system
Index failures regional weather in common weather
ENLC 1.29873 (2%) 1.63977 (28%) 2.64315 (107%)
EOLC 12.45566 (1%) 14.43680 (17%) 19.18976 (55%)
PLC 0.00143 (1%) 0.00165 (17%) 0.00220 (56%)
EONS 0.01797 (1%) 0.02138 (20"/0) 0.02963 (66%)
EENS 156.96021 (1%) 186.80725 (20"/0) 258.84094 (66%)
BPII 0.08886 (1%) 0.11705 (33%) 0.20211 (130"10)
BPECI 0.84843 (1%) 1.00977 (20"10) 1.39914 (66%)
MBPCI 0.00010 (0"/0) 0.00012 (20"10) 0.00016 (60"10)
SI 50.90601 (1%) 60.58614 (20"/0) 83.94841 (66%)
CPU
(min) 3.02 4.59 4.64
168 CHAPTER 5

Table 5.22. Annual Bus Indices Considering Common Cause Failures

and Weather Effects for the R BTS
Only Common cause failures Common cause failures
Bus common cause and recognizing and entire system
Index No. failures regional weather in common weather
PLC 2 0.00002 (0%) 0.00002 (0"10) 0.00002 (0"10)
3 0.00008 (14%) 0.00010 (43%) 0.00020 (186%)
4 0.00009 (0%) 0.00010 (11%) 0.00017 (89%)
5 0.00018 (6%) 0.00023 (35%) 0.00033 (94%)
6 0.00142 (1%) 0.00163 (16%) 0.00214 (52%)
ENLC 2 0.01102 (0%) 0.01093 (0%) 0.01099 (0%)
3 0.06431 (52%) 0.10524 (149%) 0.28162 (568%)
4 0.04871 (1%) 0.07385 (53%) 0.23938 (396%)
5 0.09303 (11%) 0.19038 (126%) 0.39787 (373%)
6 1.28313 (1%) 1.60834 (27%) 2.52083 (99%)
EDNS 2 0.00005 (0"/0) 0.00005 (0%) 0.00005 (0%)
3 0.00053 (15%) 0.00067 (46%) 0.00168 (265%)
4 0.00053 (0"/0) 0.00062 (17%) 0.00153 (188%)
5 0.00056 (2%) 0.00114 (107%) 0.00172 (213%)
6 0.01629 (0%) 0.01891 (16%) 0.02465 (52%)
EENS 2 0.43561 (0%) 0.43561 (0%) 0.43561 (0"/0)
3 4.59373 (14%) 5.84676 (46%) 14.70232 (266%)
4 4.66888 (0"/0) 5.38681 (15%) 13.39328 (187%)
5 4.91365 (3%) 9.92182 (107%) 15.00449 (214%)
6 142.34833 (0%) 165.21625 (16%) 215.30522 (52%)

It can be seen that the effects on the system adequacy indices of common
cause failures are smaller than those of weather in the analysis given. A
single common cause failure event involves the two specified lines while an
adverse weather condition can involve at least one region or several regions
which mUltiple lines traverse. At the peak load, the common cause failure
of Line pair (2,3) creates the load curtailment of 0.23 p.u. and the common
cause failure of Line pair (4,5) does not create any load curtailment. The
load curtailment due to the common cause failure of Line pair (2,3) also
decreases as the load level decreases. When the load level is equal to or less
than 86% of the peak load, the common cause failure of Line pair (2,3) no
longer creates any load curtailment and therefore the common cause failures
have hardly any effect on the annual system indices of this system. The
differences are quite large in the case of bus indices. The load curtailment
indices at Bus 2 are not affected by common cause failures as this bus is
directly connected to a generator bus. The bus indices at Bus 4 are not
affected by common cause failures because this bus is connected to the
generator bus through Line pair (4,5) and the common cause failure of line
pair (4,5) provides no load curtailment. Common cause failures have only
COMPOSITE SYSTEM ADEQUACY ASSESSMENT 169

a little impact on Buses 5 and 6 and a quite large impact on Bus 3 which is
connected directly with the Line pair (2,3). The specific effect of common
cause failures depend mainly on the transmission system configuration while
weather effects mainly depend on geographical distribution situations and
probabilities of adverse weather conditions.
It can be also seen from the results in Tables 5.15 to 5.22 that the
percentage differences between the inadequacy indices obtained considering
both common cause failures and weather-related but independent line out-
ages and those without considering these effects basically equals the sum of
the percentage differences due to only weather effects and those due to only
common cause failures. This indicates that these two kinds of effects can be
considered to be independent.

5.7. BUS LOAD UNCERTAINTY AND

CORRELATION
Bus load uncertainty and correlation always exist in an actual power
system and it has been shown in Chapter 4 that total system load uncertainty
can have a very great impact in an appraisal of generating capacity adequacy.
It is therefore important to illustrate the effects of bus load uncertainty and
correlation in composite system adequacy assessment. Load uncertainty can
be modeled using a normally distributed random variable. In the conven-
tional method, when a state enumeration technique is used, the total system
load is usually simulated using a discrete interval normal distribution(3) and
bus loads determined according to their proportion in the total system load.
This implies that all bus loads are completely dependent. In an actual power
system, however, bus loads are not completely dependent nor independent
but are correlated to some degree. Correlation between bus loads can be
expressed by a covariance matrix or a correlation coefficient matrix. In this
section, a tabulating technique for normal distribution sampling is used to
simulate bus load uncertainty and a correlation sampling technique is utilized
to simulate bus load correlation including those situations in which bus loads
are completely dependent, or independent, or correlated to some degree.

5.7.1. Tabulating Technique for Bus Load

Normal Distribution Sampling
Bus load uncertainty can be modeled using a normal distribution and
therefore it is necessary to draw normally distributed random numbers in
order to simulate bus load uncertainty. The following tabulating technique
170 CHAPTER 5

for normal distribution sampling can be used for this purpose and includes
two procedures:

(a) Tabulating. The value interval [0,1] of a normal cumulative

probability distribution function is divided into M equal subintervals and
the midpoint value of each subinterval is used to represent all values of the
normal cumulative probability distribution function F in this subinterval.
Therefore the value of function F at the ith subinterval is
_ i-O.5
Fi(x) -
I
--M (i= I, ... , M) (5.32)

A normally distributed random number Xi corresponding to the value F(Xi)

can be calculated by means of the inverse transform method, i.e.,

(i= I, ... , M) (5.33)

The random number Xi can be calculated from the value of F(Xi) using
the approximate inverse function formulas of the normal cumulative prob-
ability distribution function given by equations (3.31) and (3.32) in Section
3.4.5.
When i = I, ... , M, the subinterval number i, the values of the normal
cumulative probability distribution function F(X i) and the values of the
normally distributed random number Xi can be used to form a table.

(b) Sampling. The congruential method is used to generate a

uniformly distributed random number Y between [0, 1], which corresponds
to a value of the normal cumulative probability distribution function. A
normally distributed random number Xi can be selected directly from the
formed table according to the subinterval number i in which Y is located.
The tabulating technique for normal distribution sampling is more rapid
in computation compared to the direct inverse transform method in which
inverse function values are calculated in each sample. Only M inverse func-
tion values are calculated in the tabulating procedure in advance regardless
of the number of samples. Essentially, this is a discretization of a continuous
cumulative probability distribution function and it creates two errors: a
truncation error and a discretization error. A truncation error is due to the
last subinterval and therefore, in order to avoid this error, the direct inverse
transform method can be used only for the last subinterval instead of using
the midpoint value to represent all values of the distribution function corre-
sponding to this subinterval. When M is large enough, the discretization
error is sufficiently small and negligible.
COMPOSITE SYSTEM ADEQUACY ASSESSMENT 171

5.7.2. Bus Load Correlation Sampling

Technique
If all bus loads are completely dependent, only one normally distributed
random number is drawn for all bus loads in order to determine a load
state. If all bus loads are independent of each other, it is necessary to indepen-
dently draw the different normally distributed random numbers for each
bus load. From a practical point of view, bus loads are not completely
dependent nor independent but there is some correlation between them. In
this case, it is necessary to generate a normally distributed random vector
in which each component corresponds to a bus load and whose components
satisfy the specified correlation.
In Section 4.5.2, a correlation sampling technique is used to incorporate
area load correlation in multi-area generating system adequacy assessment.
A similar technique can be applied to consider bus load correlation in com-
posite system adequacy assessment and the details of the formulas associated
with the technique are therefore not repeated. The bus load correlation
sampling technique is summarized in the following steps:
1. Draw a noncorrelative, normally distributed N-dimension random
vector G in which each component corresponds to a bus load. All
components have a mean value of zero and a variance of unity.
2. Calculate the lower triangular matrix A according to equations (5.34)
to (5.36):

(l5.i5.N) (5.34)

(l <i5.N) (5.35)

(1 <j< i5'.N) (5.36)

where C is the covariance matrix defining bus load uncertainty and

correlation.
3. Create a correlative, normal distribution N-dimension random
vector H from equation (5.37):
H=AG+B (5.37)
where B is the mean vector of the bus loads. Each component of H
still corresponds to a bus load.
172 CHAPTER 5

Generally, it can be assumed that all bus loads have the same forecast
uncertainty. In this case, there is a simple relationship between the covariance
matrix C and the correlation coefficient matrix p of the bus loads as follows:
C= a 2 p (S.38)
where a is the standard deviation of the bus load forecast.

5.7.3. Case Studies

Case studies have been conducted using the RBTS and the IEEE R TS.
In addition to the basic data given in Appendix A, complementary data,
which are standard deviations of load normal distributions and correlation
coefficients between bus loads, are provided in this subsection. Load forecast
uncertainty is relatively small for short lead times but can be quite significant
when looking further into the future. It is therefore necessary to consider
the effects of different standard deviations. Both annualized and annual
indices have been calculated. In calculating annual indices, a IS-step non-
uniform increment model of the load duration curve was used. The numbers
of system state samples at each load level for the RBTS and the IEEE R TS
are SO,OOO and 10,000, respectively. The coefficients of variation in the vari-
ous simulation cases for the two systems range from 0.02 to O.OS.

(a) Effect of Load Uncertainty. Two load standard deviations of

4% and 10% were considered and it has been assumed that all bus loads are
completely dependent. The results are shown in Tables S.23 to S.26. The

Table 5.23. Annualized System Indices with and without

Load Uncertainty for the RBTS
Load normal distribution
Without load
Index uncertainty SO of 4% SO of 10%
ENLC 5.27252 5.75975 (9"10) 9.89653 (88%)
EOLC 91.20384 98.19264 (8%) 230.28096 (152%)
PLC 0.01044 0.01124 (8%) 0.02636 (152%)
EONS 0.12429 0.14208 (14%) 0.34905 (181%)
EENS 1085.81873 1241.24011 (14%) 3049.33813 (181%)
BPII 0.38169 0.43261 (13%) 0.80923 (112%)
BPECI 5.86929 6.70941 (14%) 16.48291 (181%)
MBPCI 0.00067 0.00077 (15%) 0.00189 (182%)
SI 352.15738 402.56433 (14%) 988.97449 (181%)
CPU
(min) 0.187 1.01 1.08
COMPOSITE SYSTEM ADEQUACY ASSESSMENT 173

Table 5.24. Annualized System Indices with and without Load

Uncertainty for the IEEE RTS
Load normal distribution
Without load
Index uncertainty SO of 4% SO of 10%
ENLC 54.75342 65.23052 (19%) 96.02156 (75%)
EOLC 698.88000 887.57758 (27%) 1405.62231 (101%)
PLC 0.08000 0.10160 (27%) 0.16090 (101%)
EONS 13.97045 17.26523 (24%) 34.12679 (144%)
EENS 122045.88281 150829.07813 (24%) 298131.62500 (144%)
BPII 3.33652 4.04968 (21%) 7.44591 (123%)
BPECI 42.82311 52.92248 (24%) 104.60758 (144%)
MBPeI 0.00490 0.00606 (24%) 0.01197 (144%)
SI 2569.38672 3175.34863 (24%) 6276.45508 (144%)
CPU
(min) 0.623 1.54 1.67

Table 5.25. Annual System Indices with and without Load

Uncertainty for the RBTS
Load normal distribution
Without load
Index uncertainty SO of 4% SO of 10"10
ENLC 1.27622 1.31200 (3%) 1.44645 (13%)
EOLC 12.35270 12.99498 (5%) 15.98906 (30"10)
PLC 0.00141 0.00149 (6%) 0.00183 (30"10)
EONS 0.01786 0.01813 (2%) 0.02216 (24%)
EENS 155.99113 158.40466 (2%) 193.58031 (24%)
BPII 0.08771 0.08849 (1%) 0.09818 (12%)
BPECI 0.84320 0.85624 (2%) 1.04638 (24%)
MBPCI 0.00010 0.00010 (0"/0) 0.00012 (20"10)
SI 50.59172 51.37448 (2%) 62.78280 (24%)
CPU
(min) 2.75 15.07 15.18

annualized and annual system indices obtained without considering load

uncertainty are also given for comparison. The percentage values in the
parentheses are the differences between the system indices obtained consider-
ing a load normal distribution and those obtained without considering load
uncertainty.
The results in Tables 5.23 to 5.26 indicate that system reliability indices
considering load uncertainty are larger than those obtained without consid-
ering load uncertainty and the differences between them increase as the load
standard deviation increases. The effect of load uncertainty, however, is
different for different systems, and for annualized and annual indices. For
174 CHAPTER 5

Table 5.26. Annual System Indices with and without Load

Uncertainty for the IEEE RTS
Load normal distribution
Without load
Index uncertainty SO of 4% SO of 10%
ENLC 0.92368 1.17028 (27%) 2.54645 (176%)
EOLC 11.70195 14.59255 (25%) 33.75541 (188%)
PLC 0.00134 0.00167 (25%) 0.00386 (188%)
EONS 0.15568 0.19704 (27%) 0.60573 (289%)
EENS 1360.02258 1721.36877 (27%) 5291.66357 (289%)
BPII 0.03785 0.04831 (28%) 0.14241 (276%)
BPECI 0.47720 0.60399 (27%) 1.85672 (289"10)
MBPCI 0.00005 0.00007 (40"10) 0.00021 (320%)
SI 28.63205 36.23934 (27%) 111.40344 (289%)
CPU
(min) 6.23 22.23 26.39

example, in the case of the relatively small load standard deviation of 4%,
the differences in the annual indices between considering and not considering
load uncertainty for the RBTS are quite small and may be masked by
computational errors.

(b) Effect of Bus Load Correlation. The correlation coefficients

between bus loads can be calculated according to the statistical definition
of the correlation function if historical data of bus loads can be provided.
When the correlation is strong, the situation approaches the completely
dependent case and, when the correlation is weak, it approaches the indepen-
dent case. It is therefore necessary to examine the following three cases:
1. Independence between bus loads.
2. Some correlation between bus loads.
3. Complete dependence between bus loads.
Case studies were again conducted using the RBTS and the IEEE R TS.
It has been assumed that all bus loads have the same standard deviation of
10%. The correlation coefficient matrix between bus loads for the RBTS is
as follows:
Bus No. 2 3 4 5 6
2 1.0 0.0 0.0 0.0 0.0
3 0.0 1.0 0.8 0.0 0.0
4 0.0 0.8 1.0 0.0 0.0
5 0.0 0.0 0.0 1.0 0.8
6 0.0 0.0 0.0 0.8 1.0
COMPOSITE SYSTEM ADEQUACY ASSESSMENT 175

It has been assumed for the IEEE RTS that industrial customers are
located mainly in the 230-kV area and commercial/residential customers in
the 138-kV area. In general, commercial/residential customers are more
influenced by common factors such as common weather or similar power-
consuming behavior compared to industrial customers. The correlation
between commercial and/or residential customers is therefore stronger than
that between industrial ones. It was also assumed that there is a weak correla-
tion between industrial and commercial/residential loads. Based on the
above assumptions, the correlation coefficients between bus loads for the
IEEE R TS are as follows:

• The correlation coefficients between the bus loads in the 138-kV area
are 0.8.
• The correlation coefficients between the bus loads in the 230-kV area
are 0.4.
• The correlation coefficients between the bus loads in the 138-kVarea
and those in the 230-kV area are 0.2.

The annualized and annual system indices considering bus load

uncertainty and correlation for the RBTS and the IEEE R TS are given
in Tables 5.27 to 5.30, respectively. The results indicate that the composite
system inadequacy indices obtained considering independence and com-
plete dependence between bus loads are lower and upper bounds of those
obtained considering different degrees of correlation between bus loads.

Table 5.27. Annualized System Indices Considering Bus

Load Uncertainty and Correlation for the RBTS
SO of loolo
Without load Completely
Index uncertainty Independent Correlated dependent
ENLC 5.27252 6.77444 7.58001 9.89653
EOLC 91.20384 128.41920 158.12161 230.28096
PLC 0.01044 0.01470 0.01810 0.02636
EONS 0.12429 0.16772 0.21047 0.34905
EENS 1085.81873 1465.19690 1838.62903 3049.33813
BPII 0.38169 0.47524 0.55301 0.80923
BPECI 5.86929 7.91998 9.93853 16.48291
MBPCI 0.00067 0.00091 0.00114 0.00189
SI 352.15738 475.19894 596.31207 988.97449
CPU
(min) 0.187 1.17 1.25 1.08
176 CHAPTER 5

Table 5.28. Annualized System Indices Considering Bus Load

Uncertainty and Correlation for the IEEE RTS
SO of 10%
Without load Completely
Index uncertainty Independent Correlated dependent
ENLC 54.75342 61.04555 77.36040 96.02156
EOLC 698.88000 804.58563 1098.11523 1405.62231
PLC 0.08000 0.09210 0.12570 0.16090
EONS 13.97045 15.54398 22.73200 34.12697
EENS 122045.88281 135792.23438 198586.76563 298131.62500
BPII 3.33652 3.70869 5.16711 7.44591
BPECI 42.82311 47.64639 69.67956 104.60758
MBPCI 0.00490 0.00545 0.00798 0.01197
SI 2569.38672 2858.78369 4180.77344 6276.45508
CPU
(min) 0.623 1.69 1.81 1.67

Table 5.29. Annual System Indices Considering Bus Load

Uncertainty and Correlation for the RBTS
SO of 10"10
Without load Completely
Index uncertainty Independent Correlated dependent
ENLC 1.27622 1.32834 1.34947 1.44645
EOLC 12.35270 13.44740 13.94054 15.98906
PLC 0.00141 0.00154 0.00160 0.00183
EONS 0.01786 0.01930 0.02025 0.02216
EENS 155.99113 168.62752 176.87294 193.58031
BPII 0.08771 0.09143 0.09379 0.09818
BPECI 0.84320 0.91150 0.95607 1.04638
MBPCI 0.00010 0.00010 0.00011 0.00017
SI 50.59172 54.69001 57.36420 62.78280
CPU
(min) 2.75 16.18 17.72 15.18

The degrees of effects of correlation between bus loads are different for
annualized and annual indices. The variations in annualized and annual
EENS system indices with the bus load standard deviation for the RBTS
and the IEEE RTS are shown in Figures 5.9 to 5.12, respectively. The
EENS values in the 5gures are expressed in per unit of the annualized
or annual EENS obtained without considering load uncertainty. The
upper curve corresponds to complete dependence between bus loads and
COMPOSITE SYSTEM ADEQUACY ASSESSMENT 177

Table 5.30. Annual System Indices Considering Bus Load

Uncertainty and Correlation for the IEEE RTS
SD of 10"10
Without load Completely
Index uncertainty Independent Correlated dependent
ENLC 0.92368 1.13351 1.63481 2.54645
EDLC 11.70195 14.07685 20.44840 33.75541
PLC 0.00134 0.00161 0.00234 0.00386
EDNS 0.15568 0.18129 0.30653 0.60573
EENS 1360.02258 1583.71399 2677.87231 5291.66357
BPII 0.03785 0.04494 0.07612 0.14241
BPECI 0.47720 0.55569 0.93960 1.85672
MBPCI 0.00005 0.00006 0.00011 0.00021
SI 28.63205 33.34135 56.37626 111.40344
CPU
(min) 6.23 23.16 29.25 26.36

the lower one to independence between bus loads. The range between
these two curves therefore corresponds to different degrees of correlation
between bus loads. It can be seen that the variation ranges are quite
large with the exception of the annual system indices for the RBTS.

6.0
100% dependence between bus loads
5.0 independence between bus loads

-=-
-
::l
ci. 4.0

/
en
zw
w 3.0

,....---/
. /

----......
2.0

- -
.._....._... .._.....
.......

1.0
0.00 0.02 0.04 0.06 0.08 0.10 0.12
standard deviation
Figure 5.9. Variations in the annualized EENS with the bus load standard deviation for
the RBTS.
178 CHAPTER 5

6.0
100% dependence between bus loads
5.0 - - independence between bus loads

-
-
:i 4.0
ci.
C/)
Z
w 3.0
w
...............................
..
-------------------
2.0 '

1.0
0.00 0.02 0.04 0.06 0.08 0.10 0.12
standard deviation
Figure 5.10. Variations in the annualized EENS with the bus load standard deviation
for the IEEE RTS.

6.0
1000/0 dependence between bus loads
independence between bus loads
5.0

-
:i
-
4.0
ci.
C/)
Z
w 3.0
w

2.0

......................................................
1.0 ~~--~~~~~~~~
0.00 0.02 0.04 0.06 0.08 0.10 0.12
standard deviation
Figure 5.11. Variations in the annual EENS with the bus load standard deviation for
the RBTS.
COMPOSITE SYSTEM ADEQUACY ASSESSMENT 179

6.0~--------------------------------------'

----------- 1000/0 dependence between bus loads /

5.0 Independence between bus loads /

-
:::;
-z
4.0
ci / '"
rn ,/"
w 3.0 "
w /
/"

..............". "
2.0 .""
..".'"
......, .".'
........................................,.-
1.0
0.00 0.02 0.04 0.06 0.08 0.10 0.12
standard deviation
Figure 5.12. Variations in the annual EENS with the bus load standard deviation for the
IEEE RTS.

5.S. FREQUENCY INDEX AND SYSTEM

STATE TRANSITION SAMPLING
TECHNIQUE
5.S.1. Discussion on the Frequency Index
The system state sampling method can provide accurate probability
and energy-related indices but it cannot be used to directly calculate actual
frequency-related indices. Alternatively, it calculates Expected Number of
Load Curtailments, which is an approximation of the actual frequency index.
The approximation results from the following two factors.

(a) Transitions between System States are Ignored. The fre-

quency concept is based on transitions between system states. Consider the
system state transition of the two repairable component system shown in
Figure 5.13. It is assumed that component 2 down (State 3) and both compo-
nents down (State 4) are load curtailment states while component 1 down
(State 2) is a no-load-curtailment state. Only those transitions between States
1 and 3 and between States 2 and 4 contribute to the frequency index. The
180 CHAPTER 5

1U 2UL!. J.1 1 ~ 10 2U~

J.12 J.12

-- -------------- 1-------
BounJ!.lllryWall

12 12

~ li.
1U 20l! J.1 1 10 20

Figure 5.13. Transitions between system states for a two-repairable-component system.

transition between States 3 and 4 should not be included in the frequency

index. This transition, however, is included in the ENLC index obtained by
system state sampling. Consequently, this leads to overestimation of the
frequency index. If an additional enumeration procedure is used to recognize
all no-Ioad-curtailment states which can be reached from each selected failure
state in one transition, the actual frequency index can be calculated. How-
ever, this can result in a very large computational burden. If a failure state
is associated with n components, each of which is modeled by two states,
n + I additional adequacy evaluations including load flow calculations, cor-
rective actions, etc. are required for this failure state to update the frequency
estimate. If generating unit derated states are considered, the number of
additional evaluations increases dramatically. In the case of a large compos-
ite system, this task can be computationally unfeasible.

(b) Chronology of the Load Curve is Eliminated. Consider the

simple example shown in Figure 5.14a. It is assumed that the duration of a
failure event is t and the power with which the system can provide some
bus during the failure event is expressed by the horizontal line A. The load
curve at this bus is expressed by B. The frequency of load curtailment at
the bus should be three in the duration t in Figure 5.l4a. If chronology of
the load curve is eliminated, the corresponding load duration curve is
expressed by C in Figure 5.14b. The frequency of load curtailment in the
duration t is only one. This example indicates that elimination of chronology
of the load curve may lead to underestimation of the frequency index.
COMPOSITE SYSTEM ADEQUACY ASSESSMENT 181

Load

f--
6"B r---
A

Time
(a) t

Load

c
Ar---------+-------------~

~------------------------~~.TIme
(b) t

Figure 5.14. Effect of eliminating load curve chronology on the frequency index.

Note that total duration of load curtailment is the same and therefore the
probability of load curtailment and the energy not supplied are the same in
the two cases.
Annualized indices are usually calculated at the single peak load level
and therefore affected only by the first aspect. As a result, the annualized
ENLC index is a pessimistic estimation of the actual annualized frequency
index. Annual indices considering the annual load duration curve are affected
by both aspects. If the effect due to eliminating the chronology of the load
curve is larger than that due to ignoring transitions between system states,
the annual ENLC index can be an optimistic estimation of the actual annual
frequency index. Otherwise, it is a pessimistic estimation.
Three basic simulation approaches are illustrated in Section 3.6. These
techniques are designated as state sampling, state duration sampling, and
system state transition sampling. The state duration sampling technique uses
a chronological load curve and component state transition cycles. If this
technique is applied, the actual frequency index can be obtained. This
182 CHAPTER 5

technique has been utilized in generating system adequacy assessment in

Chapter 4. Its application to composite system adequacy assessment is a
much more difficult task. This is not only because one year has 8760 hourly
load points but also because the chronological load curves of each bus are
basically different. In composite system assessment, load curtailments occur
at buses and therefore chronological bus load curves should be used if the
state duration sampling method is applied. Evaluation of each system state
is associated with load flow calculations, contingency analysis, overload
alleviation, generation rescheduling, and other possible considerations such
as those given in Sections 5.4 to 5.7. This may not be computationally
feasible in the case of a practical composite system. A more difficult obstacle
is that of the data problem. At the present time, most utility companies
cannot provide data records of the chronological load curves for all or for
even major buses.
Application of the system state transition technique to composite system
adequacy assessment is presented in the following subsection. This technique
is combined with the multistep load model given in Section 5.3. The approxi-
mation associated with ignoring transitions between system states is removed
but the approximation due to eliminating chronology of load curves still
remains. As a result, an actual annualized frequency index can be calculated
and the annual frequency index is an optimistic estimation of the actual
annual frequency index.

5.8.2. System State Transition Sampling

Technique
This technique focuses on state transition of the whole system rather
than on component states or component state durations. It creates a system
state transition sequence without the need to sample up and down cycles of
components and storing chronological information on system states. The
generating unit derated states can be easily included without additional
calculations. An important restriction in the method is the assumption of
exponential distributions for all state residence times.

(a) Basic Principles. The system state transition sampling tech-

nique is described in Section 3.6.3. Two main principles associated with its
application are as follows:
1. Assume that a system contains m components and that the duration
of each component state follows an exponential distribution with parameter
Ai. The duration of the system state also follows an exponential distribution
COMPOSITE SYSTEM ADEQUACY ASSESSMENT 183

with the parameter A=LAi, i.e., it has the probability density function

/(/)= i~l Aiexp(- i~l Ail) (5.39)

2. Starting from any system state, a system containing m components

has m possible reached states. The probability that the system reaches one
of these possible states is given by

(5.40)

The probabilities of m possible reached states are successively placed in the

interval [0,1] as shown in Figure 3.11. A uniformly distributed random
number U between [0, 1] is generated and if U falls into the segment corre-
sponding to Pj, the transition ofthejth component leads to the next system
state. A long system state transition sequence can be obtained by drawing
a number of samples and the consequences of each system state can be
evaluated using the same system analysis techniques given in Section 5.2.

(b) Evaluation Procedure. Composite system adequacy assess-

ment using the system state transition sampling technique can be summarized
by the following steps:
1. The simulation process starts from the normal system state in which
all generating units and transmission components are in the up state.
2. Let Ai be the transition rate at which one component departs its
present state. If the component is in the up state, A.i is its failure rate, and
if the component is in the down state, Ai is its repair rate. If multistates of
a component are considered (such as derated states of a generating unit),
there are several departure rates. The duration of the system state has a
probability density function shown in equation (5.39). According to the
inverse transform method, the sampling value Dk of the duration of the
present system state (numbered by k) can be obtained by (see Section 3.4.4)
-In U'
Dk=-m- (5.41)
Li=lAi
where m is the number of departure rates corresponding to the present
system state k and U' is a uniformly distributed random number between
[0, 1].
3. If the present system state is a contingency state in which at least
one component is in the outage or the derated state, the system analysis
techniques, including the minimization load curtailment model given in
184 CHAPTER 5

Section 5.2, are used to evaluate the adequacy of this system state. If the
present system state is the normal state in which all components are in their
up states, go to Step 4 without utilizing the minimization model.
4. The adequacy indices are updated using equations (5.42) to (5.44)
given later. If the coefficient of variation of the EDNS (Expected Demand
Not Supplied) index is less than a given tolerance or the specified number
of system state samples is reached, the simulation process ends. If not,
proceed to Step 5.
5. A uniform distribution random number U is generated to determine
the next system state using the system state transition sampling technique
described earlier. Return to Step 2.

(c) Calculation of Adequacy Indices. The total load curtailment

Ck for the system state k can be obtained by solving the minimization model
shown by equations (5.20) to (5.25). If Ck equals zero, the system state is a
no-load-curtailment state. If not, it is a failure state. The three basic system
unreliability indices can be calculated using the following equations:

(1) Expected Demand Not Supplied (MW)-EDNS

NK
EDNS = L CkDk/TD (5.42)
k=1

where Ck (MW) and Dk (hr) are the system load curtailment and the duration
for system state k; NKis the number of load curtailment system states; TD
(hr) is the sum of the durations of all system states in a long system state
transition sequence.

(2) Expected Frequency of Load Curtailments (occ./yr)-EFLC

EFLC=NF' 8760/TD (5.43)

where NF is the number of occurrences of transition from a load curtailment

state to a no-load-curtailment state in the system state transition sequence.

(3) Probability of Load Curtailments-PLC

NK
PLC= L Dk/TD (5.44)
k=1

Bus indices can be obtained using similar equations since the minimiza-
tion model also provides load curtailments at each bus for all drawn system
states. Other unreliability indices can be calculated from these three basic
indices.
COMPOSITE SYSTEM ADEQUACY ASSESSMENT 185

5.8.3. Case Studies

The application of the method is illustrated using the IEEE R TS. Com-
parisons between the system state sampling and the system state transition
sampling techniques are shown. The studies include base and generating
unit derated cases. The data for both cases are the same as those given in
Sections 5.2.4 and 5.4.2, respectively. In the case of annual indices, the
annual load curve is considered by modeling the 8736 hourly load points
by a 70-step load level model. In all study cases, the coefficient of variation
for the EDNS index ranges from 0.03 to 0.05. Tables 5.31 and 5.32 show
the annualized system indices in the base and the generating unit derated
cases, respectively. Table 5.33 shows annual system indices for the base case.
The * flag indicates the frequency and duration indices. In the case of the
system state sampling method, the Expected Number of Load Curtailments
(ENLC), i.e., the sum of occurrences of load curtailment states, is calculated
as a surrogate for the actual frequency index (EFLC) and therefore the
corresponding ADLC (Average Duration of Load Curtailments in hr/dis-
turbance) is also an ENLC-based index.
It can be seen that although these two methods utilize completely differ-
ent sampling techniques, the unreliability indices that are not associated with
the frequency index are quite close. Additional calculations indicate that
when the number of samples is increased, these unreliability indices become
closer. On the other hand, however, the frequency and duration indices

Table 5.31. The Annualized System Indices for the

IEEE RTS Using Two Monte Carlo Methods for the
Base Case
System state
transition System ·state Difference
Index sampling sampling (in %)
PLC 0.08434 0.08000 5.1
EDLC 736.77460 698.88000 5.1
* EFLC 19.57152 54.75342 179.7
(ENLC)
* ADLC 37.64524 12.76413 194.9
EONS 14.44465 13.97045 3.3
EENS 126188.45313 122045.88281 3.3
BPECI 44.27665 42.82311 3.3
SI 2656.59863 2569.38672 3.3
CPU
(min) 1.51 0.62
186 CHAPTER 5

Table 5.32. The Annualized System Indices for the

IEEE RTS Using Two Monte Carlo Methods for the
Generating Unit Derated Case
System-state
transition System-state Difference
Index sampling sampling (in %)
PLC 0.06784 0.07220 6.4
EDLC 592.67310 630.73920 6.4
* EFLC 17.73042 47.38287 167.2
(ENLC)
* ADLC 33.42691 13.31154 151.2
EDNS 10.31458 10.32736 0.2
EENS 90108.20313 90219.83594 0.2
BPECI 31.61691 31.65608 0.2
SI 1897.01465 1899.36487 0.2
CPU
(min) 1.47 0.74

Table 5.33. The Annual System Indices for the IEEE

RTS Using Two Monte Carlo Methods for the
Base Case
System-state
transition System-state Difference
Index sampling sampling (in %)
PLC 0.00119 0.00117 1.7
EDLC 10.36991 10.23695 1.3
* EFLC 0.39417 0.80854 105.1
(ENLC)
* ADLC 26.30801 12.66103 108.2
EDNS 0.14673 0.13137 10.5
EENS 1281.79480 1147.61108 10.5
BPECI 0.44975 0.40267 10.5
SI 26.98515 24.16023 10.5
CPU
(min) 39.73 15.97

(EFLC or ENLC and ADLC) obtained using the two methods have very
large differences. This indicates that the utilization of the ENLC index as a
surrogate for the EFLC index can lead to a considerable overestimation.
Both the annual EFLC and ENLC indices were obtained using a nonchrono-
logical multistep load model.
COMPOSITE SYSTEM ADEQUACY ASSESSMENT 187

5.9. SECURITY CONSIDERATIONS

IN COMPOSITE SYSTEM
ADEQUACY EVALUATION
This section discusses security considerations in composite system
adequacy evaluation. These considerations are designated as security con-
strained adequacy evaluation. The analysis is therefore within the domain of
adequacy and is not "security evaluation" which is associated with dynamic
conditions and the effects of cascading sequences after failures. The system
operating state division used in the approach, however, does not restrict the
inclusion of transient or dynamic conditions.

5.9.1. System Model Including Security

Considerations
(a) Classification of System Operating States. In order to rec-
ognize security considerations in the evaluation of a composite system, the
total power network can be divided into several operating states in terms of
the degree to which adequacy and security considerations are satisfied. Fig-
ure 5.15 shows a possible classification of the system operating states.(38)
The definitions of the system operating states are as follows(39):
The Normal State: "In the normal state, all equipment and operation con-
straints are within limits, including that the generation is adequate to supply

r-- Normal

Restorative Alert
~
Extreme
!
Emergency +--
------ Emergency

Figure 5.15. Classification of system operating states.

188 CHAPTER 5

the load (total demand), with no equipment overloaded. In the normal state,
there is sufficient margin such that the loss of any elements, specified by
some criteria, will not result in a limit being violated. The particular criteria,
such as all single elements, will depend on the planning and operating philos-
ophy of a particular utility." It is clear that the system is both adequate and
secure in the normal state.

The Alert State: "If a system enters a condition where the loss of some
element covered by the operating criteria will result in a current or voltage
violation, then the system is in the alert state. The alert state is similar to
the normal state in that all constraints are satisfied, but there is no longer
sufficient margin to withstand an outage (disturbance). The system can enter
the alert state by the outage of equipment, by a change in generation sched-
ule, or a growth in the system load." In the alert state the system no longer
has sufficient margin to satisfy the security constraints.

The Emergency State: "If a contingency occurs or the generation and load
changes before corrective action can be (or is) taken, the system will enter
the emergency state. No load is curtailed in the emergency state, but equip-
ment or operating constraints have been violated. If control measures are
not taken in time to restore the system to the alert state, the system will
transfer from the emergency state to the extreme emergency state." In this
state both adequacy and security constraints are violated. This is a temporary
state which requires operator action because equipment operating con-
straints have been violated. The first objective will be to remove the viola-
tions without load curtailment, by such means as phase shifter adjustment,
redispatch, or startup of additional generation. If successful, this could lead
to the alert state, where further actions would still be necessary to achieve
the normal state. Such actions could include voltage reduction. On the other
hand, once the alert state is reached, it may be decided to take no further
control action.

The Extreme Emergency State: "In the extreme emergency state, the equip-
ment and operating constraints are violated and load is not supplied." In
this state, load has to be curtailed in a specific manner in order to return
from this state to another state.

(b) Steady State Security Constraints. The steady state security

constraints are basically the operating limits which have to be satisfied for
normal operation of the power system. The basic security constraints
in order to operate the system within an acceptable security domain are:
COMPOSITE SYSTEM ADEQUACY ASSESSMENT 189

(l) Voltage magnitude constraints: Operating limits are imposed on

the voltage magnitude of buses, i.e.,
(5.45)
(2) Branch flow constraints: These are the thermal capacity limits on
transmission lines and transformers. In some cases, the steady state stability
limits on transmission lines expressed by angle differences can also be trans-
formed into branch flow constraints.
ITI = Tmax (5.46)
where T represents the line flows on branches.
(3) Real power generation constraints: The real power generation con-
straints at the slack bus and generator buses are
(5.47)
(4) Reactive power generation constraints: The reactive power genera-
tion constraints at the slack bus and generator buses are
(5.48)
There may be other security constraints. For example, in the operating
guide of some utilities, transient stability or voltage stability limits can be
expressed as power flow limits on crucial branches or generation limit curves.
Therefore these stability limits can be introduced indirectly into adequacy
evaluation as security constraints.

5.9.2. System Analysis Techniques

System analysis techniques are associated with the following two
aspects:

(a) Contingency Analysis. In order to include bus voltage and

reactive constraints, AC load flow based contingency analysis has to be
utilized. The basic equations are the following fast decoupled load flow
equations:
[AP/V] = [B'][VAo] (5.49)
[AQ/V] = [B"][AV] (5.50)
where
[AP] = Vector of bus real power injection increments
[AQl = Vector of bus reactive power injection increments
190 CHAPTER 5

[AS] = Vector of bus phase angle increments

[A V] = Vector of bus voltage magnitude increments
[B'] = Constant bus admittance matrix for real power iteration
[B"] = Constant bus admittance matrix for reactive power iteration.
The contingency analysis commences with the normal decoupled load
flow solution and updates the changes in bus voltage magnitudes and angles
using the linearized equations, followed by the limit check for the respective
constraints for a contingency without solving the load flow equations. (40) The
more accurate AC load flow based contingency analysis method described in
Appendix C.4 can also be used.
(b) Corrective Actions. Corrective actions to alleviate security
constraint violations can be conducted by an Optimal Power Flow (OPF)
solution. Nonlinear OPF requires a large amount of CPU time and can
encounter convergency problems in some multicomponent failure situations.
The ability to include a high degree of accuracy in corrective calculations
will never override inherent uncertainties in the forecast data including load,
failure rates, and repair times and in Monte Carlo simulation. It is therefore
reasonable to use a linearized optimization model for corrective actions.
1. Linear programming model for load curtailments and generation
rescheduling. In the event of constraint violation(s), the system could be,
according to the definitions, in the emergency state or the extreme emergency
state depending on whether there exist load curtailments after corrective
actions have been taken. The main objective, in this regard, is to alleviate
the problem using rescheduling of generation without requiring load curtail-
ments. If it is not possible to overcome the difficulty by rescheduling the
generation, then load will be curtailed from different buses. This task can
be performed using the linear programming model given in equations (5.14)
to (5.19). If bus indices are also required, the model expressed by equations
(5.20) to (5.25) should be applied.
2. Increment-type linear programming model for voltage adjustment and
Q-load curtailment. In the case of bus voltage violations, these violations
may be alleviated by adjustments of generator bus voltages, reactive power
injections of reactive sources, and transformer taps. If it is not possible to
eliminate all the bus voltage violations by these adjustments, then reactive
load curtailments at some buses are unavoidable. Reactive load curtailments
should be minimized in this case. The following increment-type linear pro-
gramming model can be used for this purpose(41):
min I ailAV;1 + I pjAQj (5.51)
ieNG jeNC
subject to
[B"] [A V] = [AQ] (5.52)
COMPOSITE SYSTEM ADEQUACY ASSESSMENT 191

(iENGuNR) (5.53)
A~;;z.AVl UENC) (5.54)
AQ:"in ~ AQi ~ AQ:"ax (iENGuNR) (5.55)
O~AQj~AQjax UENC) (5.56)
where A Vand AQ are bus voltage and reactive power change vectors, respec-
tively. Equation (5.52) is in the form of the voltage-reactive power equation
in the decoupled load flow but the signs of the components of [B"] are
treated individually. Quantities AVfin, AVfax, Ag:nn, and Agrax are the
lower and upper limits of AVi and AQi, respectively, while AVl are voltage
violations at load buses; NG, NC, and NR are the sets of generator, load,
and static reactive source buses, respectively. Equation (5.53) indicates that
the generator and reactive source bus voltage adjustments should be within
the permissible changes, which are the differences between the bus voltage
limits and their actual values in the contingency state. Equation (5.54) indi-
cates that load bus voltage changes should be larger than their violations,
which are the differences between the actual load bus voltages in the contin-
gency state and their limits. When the linear programming relaxation tech-
nique is used, (32) only load bus voltage change variables corresponding to
voltage violated buses are introduced in the model at each iteration. Equa-
tion (5.55) indicates that reactive power adjustments at generator and reac-
tive source buses should be within the permissible changes, which are the
differences between the bus reactive power limits and their actual values in
the contingency state. Equation (5.56) indicates that when load bus reactive
power curtailments are unavoidable, these curtailments cannot exceed the
reactive loads themselves. Therefore l1(!J'ax are simply the bus reactive loads.
The effects of transformer tap changes can be implicitly included in [B"].
The main objective of the optimization model is to minimize total Q-Ioad
curtailments. The weighting factors pj therefore have to be larger than all
ai. The first term in the objective function provides the possibility that when
there is no need of Q-Ioad curtailment, the generator bus voltage adjustments
will be minimized.

5.9.3. Hybrid Methodology

A hybrid method consisting of the state sampling and the enumeration
procedures can be used to evaluate the system operating state probabilities.
The flowchart of the method is shown in Figure 5.16.
In this approach, the system state is determined using the state sampling
technique described in Section 5.2.1. If the sampled situation creates a system
192 CHAPTER 5

Figure 5.16. Flowchart for the hybrid method.

problem, the probability index of the emergency or the extreme emergency

state is enhanced depending on whether or not load curtailments occur after
corrective action. If the sampled situation does not create a system problem,
then the enumeration method is utilized to determine whether the sampled
COMPOSITE SYSTEM ADEQUACY ASSESSMENT 193

situation belongs to the normal or to the alert state. As soon as any system
problem is encountered with the enumerated outages, no more outages using
the enumeration method are considered and the index for the alert state is
enhanced. If none of the enumerated states creates any system problem, the
indices for the normal state are enhanced. In all the cases, the full sampling
probability (=l.O/sample number) is added to the appropriate system state.
If a sampled system state is a high-level contingency state and it still does
not create any system problem, this is defined as an "extreme sampling
situation." This is seldom an actual operating state. In this case, no further
enumerated outages are considered. An extreme sampling situation makes
a contribution to the "no problem" condition, which is a classification
beyond the four operating states in Figure 5.15. The probability index of
the extreme emergency state converges slower than that of other states. The
coefficient of variation for the probability index of the extreme emergency
state therefore should be used as a stopping rule in the Monte Carlo sampling
method.

5.9.4. Case Studies

The IEEE R TS and the RBTS were used to conduct case studies for
composite system adequacy assessment with security considerations. The
6-bus RBTS has been modified to a 5-bus system (MRBTS) having 8 lines.
This is done to remove the radial line between bus 5 and 6 as, in the original
case, the probability of the normal state will be always zero because the
outage of line 9 (single level contingency) will isolate bus 6 and result in the
curtailment of the 20 MW load at bus 6.
The results obtained at the annual peak load level are shown in Figures
5.17 and 5.18 for the two test systems. As can be seen from the two figures,
the indices converge to the corresponding analytical values as the number
of samples increases. The perturbations in the probabilities of the normal
and alert states for the IEEE R TS can be appreciated from the fact that on
the 1256th sample, the sampled situation is a single-level outage (line 27)
which creates a voltage problem. Therefore, according to the definition of
the normal state, this system has no normal state since it cannot withstand
all the single-level outages and thus the probability index of the normal state
goes down to zero. The indices of the emergency state are zero for both the
test systems for about the first 2000 samples, since the sampled situations
up to this sampling number create either no system problems or only system
problems which require load curtailments. The probability indices of the
operating states for various coefficients of variation are presented in Table
5.34 for the MRBTS and in Table 5.35 for the IEEE RTS.
 0.3 <'">
'02.0
...
1.0 Alert state ~
Normal state Emergency state
x

1.5
Simulation ~ 0.2 Simulation
~ 0.9 ~
~
:0 Analytical Analytical
.g 1.0
.0 ~'"
e'" a.. 0.1 Simulation ~
a.. 0.8 ,.,~
Analy1lcal
a.. 0.5 ~-........~""""______
0.0.)...1~~~~~--~-~~-~ ... .. .......... 10000
0.7 I 2000 o 2000 4000 6000 8000 10000 0.0 o......2000 4000 6000 8000
o 4000 6000 8000 10000
Number of samples Number of samples
Number"of samples

0.020 "'0 10.0

Extreme emergency state
x No problem
8.0
0.015
~, ~ 6.0
:0
fJ 0.010
'"
~ 4.0
~ a..
a..
0.005 Simulation
2.0
Analytical

0.000l'-~--~-----~----­ 0.0
o 2000 4000 6000 8000 10000 0 2000 4000 6000 8000 10000
Number of samples Number of samples
(")
:I:
:to
Figure 5.17. Probabilities of different operating states for the MRBTS. ~
m
:%I
U1
 0.3 1.0 '(~ 3.0
Normal state Alert state Emergency state
x

~ 0.2 Simulation 0.8 :>. 2.0

.1!'
ii Analytical Simulation
:5 j
1l 1l Analytical
e
[l. 0.1 £. 0.6 a.e 1.0 .~ I -__ Simulation I

Analytical
0.0+-1- - - ' - r -_ _ _ _ _ _ _ _ _ _ ~
O.Oj...l-...J-~--_-~-- _ _~ 0.4-1-1~-~-----~----~
o 2000 4000 6000 8000 10000 o 2000 4000 6000 8000 10000 o 2000 4000 6000 8000 10000
Number of samples Number of samples Number of samples

0.31 0.06
Extreme emergency state
No problem

~ 0.2 Simulation
~0.04 I Simulation
.5 Analytical
as - - - - - - Analytical
.0 .5 nn
e as
.0
a. 0.1
£. 0.02
0.0 0.00
0 2000 4000 6000 8000 10000 0 2000 4000 6000 8000 10000
Number of samples Number of samples

Figure 5.18. Probabilities of different operating states for the IEEE RTS. ...
CD
U1
196 CHAPTER 5

Table 5.34. Probability of Different States for

Various Coefficients of Variation for the MRBTS
Coefficients of variation
System states 0.10 0.05 0.035
Normal 0.809951 0.813126 0.813210
Alert 0.180468 0.177479 0.177860
Emergency 0.000264 0.000199 0.000170
Ext. emergency 0.008790 0.008796 0.008280
No problem 0.000527 0.000355 0.000430
Samples 11,376 45,132 100,000
CPU (sec) 9.21 36.92 83.14

Table 5.35. Probability of Different States for Various

Coefficients of Variation for the IEEE RTS
Coefficients of variation
System states 0.10 0.05 0.0326 0.0102
Normal 0.000000 0.000000 0.000000 0.000000
Alert 0.887443 0.876200 0.876600 0.876990
Emergency 0.000000 0.001200 0.001800 0.001740
Ext. emergency 0.071210 0.087572 0.085800 0.087310
No problem 0.026034 0.021593 0.022200 0.019930
Samples 1306 4168 10,000 100,000
CPU (min) 0.55 1.77 4.15 41.50

5.9.5. Application of Variance Reduction

Techniques
The variance indicates the variability of the estimator around the
expected value. Therefore large variance essentially implies poor accuracy.
A variance reduction technique (VRT) can be used to reduce the variance
of the estimator by replacing the original sampling procedure by a new
procedure that yields statistically the same expected value but with a smaller
variance. With proper application, a VRT can result in a higher precision,
e.g., a smaller confidence interval for the same number of samples, or alterna-
tively, achieve a prespecified precision with a smaller number of samples.
Five variance reduction techniques are described in Section 3.5. In principle,
these five techniques can be used in power system reliability evaluation.
Different VRTs have, however, completely different efficiencies in different
cases.
COMPOSITE SYSTEM ADEQUACY ASSESSMENT 197

Table 5.36. Probability of Different States

without and with VRT for the MRBTS
Stratified
after
System states NoVRT Antithetic sampling
Normal 0.812880 0.811100 0.812430
Alert 0.177720 0.179840 0.178170
Emergency 0.000180 0.000180 0.000179
Extreme emergency 0.008820 0.008560 0.008813
No problem 0.000340 0.000300 0.0000325

The antithetic variates and the stratification after sampling methods

were applied to study the effect of VRT on the probability indices of the
operating states. The concept of the antithetic variates method is given in
Section 3.5.5. The stratification after sampling method is basically the same
as the stratified sampling method given in Section 3.5.4, except that in this
case the number of observations in each stratum is not prefixed. It is noted
in Section 5.3.1 that the utilization of a multistep model of the annual load
duration curve to calculate annual indices is essentially similar to a stratified
sampling method, e.g., loads are stratified in terms of their levels. The low
load levels which make smaller contributions to the total indices can be
allowed to have a relatively smaller number of samples or a relatively large
variance. This method has been proven to be computationally efficient for
that case. In the following studies, however, only the peak load level is
considered. Stratification is applied to division of component outages. The
five strata are:

1. No outage.
2. Only line outages.
3. Only generator outages.
4. Generator and line outages.
5. Outages beyond the level considered.

The probabilities of the above five strata (corresponding to ~ in the

formulas given in Section 3.5.4) can be calculated from the failure and repair
rates of the components using an analytical method.
The two test systems (the MRBTS and the IEEE RTS) were again used
for this study. The total number of samples is 50,000 for the MRBTS and
10,000 for the IEEE RTS, respectively. The probability indices for the
MRBTS using the two variance reduction techniques are given in Table
5.36. The number of samples belonging to each stratum for the stratification
198 CHAPTER 5

Table 5.37. Number of Samples in Different

Operating States from Different Strata for the
MRBTS
Stratum
System states 2 3 4 Total
Normal 39,759 47 838 0 40,644
Alert 0 836 7872 178 8886
Emergency 0 5 0 4 9
Extreme emergency 0 2 422 17 441
No problem 0 0 0 17 17
Total 39,759 890 9132 216 49,997

Table 5.38. Probability of Different States

without and with VRT for the IEEE RTS
Stratified
System states NoVRT Antithetic after sampling
Normal 0.000000 0.000000 0.000000
Alert 0.876600 0.875700 0.866372
Emergency 0.001800 0.001900 0.001628
Extreme emergency 0.085800 0.084700 0.084255
No problem 0.022200 0.020700 0.020906

Table 5.39. Number of Samples in Different

Operating States from Different Strata for the
IEEE RTS
Stratum
System states 2 3 4 Total
Normal 2288 1 132 1 242
Alert 0 42 6225 77 6344
Emergency 0 5 0 13 18
Extreme Emergency 0 3 849 6 858
No problem 0 0 171 51 222
Total 2288 51 7377 148 9864

after sampling method are shown in Table 5.37. The similar results for the
IEEE RTS are given in Tables 5.38 and 5.39. Table 5.40 shows the CPU
times and the variances of the estimates of the extreme emergency state
probability. As expressed in equation (3.11), the efficiency of the simulation
method can be measured by the product of the variance and the CPU time.
It can be seen from Table 5.40 that the antithetic variate technique is more
efficient than the stratification after sampling method with respect to the
COMPOSITE SYSTEM ADEQUACY ASSESSMENT 199

Table 5.40. Variances of Estimates of

Extreme Emergency State Probability and
CPU Times
CPU time on
Method Variance Micro VAX 3600
MRBTS
without VRT 1.75E-7 38.86 sec
antithetic 8.50E-8 37.74 sec
stratification 1.68E-7 38.32 sec
IEEE RTS
without
VRT 7.84E-6 4.15 min
antithetic 3.69E-6 4.13 min
stratification 7.34E-6 4.15 min

calculation of probability indices for the extreme emergency state in both

test systems. In fact, the stratification after sampling method basically does
not provide improvement compared to the original sampling method without
VRT. It should be noted that the efficiency of the methods with respect to
other system operating states may lead to different conclusions.

5.10. NONCOHERENCE IN COMPOSITE

SYSTEM ADEQUACY ASSESSMENT
Coherence is an important concept in general engineering system reli-
ability evaluation. This concept states that if a component fails, system
reliability never improves. Conversely, if a lost component is recovered,
system reliability never decreases. The coherence assumption is widely used
in engineering system reliability assessment. It is valid in generating system
adequacy evaluation; however, it is not always true for a transmission or a
composite generation and transmission system. This is because reliability of
such a system does not only depend on failure or repair of components, but
also on other constraints such as load flow equations and the capacity limits
of transmission components.

5.10.1. Definitions and Measures of

Noncoherence
(a) Definitions of Noncoherence. A general definition of nonco-
herence is that the loss of a component creates a more reliable system state;
200 CHAPTER 5

conversely, the recovery of a lost component leads to a less reliable system

state. From a power system planning point of view, however, a looser defini-
tion can be used. The looser definition states that loss of a component does
not create a less reliable system state and recovery of a lost component does
not lead to a more reliable system state.
The noncoherence feature is a function of the system states. The effect
on system performance of the loss or recovery of a component depends on
not only the component itself but also the system state prior to its loss or
recovery. In other words, it is possible that the loss or recovery of a compo-
nent may be noncoherent for a system state but coherent for another system
state. It is necessary to relate noncoherence to failure levels. The following
definitions are therefore further suggested:
• If in the normal system state, loss of one or more components creates
a more reliable state than that normal state, the system is zero-level
noncoherent.
• If in a single-level contingency system state (a component has failed),
loss of one or more components creates a more reliable state than
that contingency state, the system is first-level noncoherent.
The second-level, third-level, etc., noncoherence of a system can be
defined similarly. When a system is zero-level coherent, it can be first-level
or higher-level noncoherent. Basically, a composite system is and should be
zero-level coherent because attempts are always made to avoid noncoherence
in designing it. It is possible, however, that a transmission or composite
system can be first- or higher-level noncoherent. This is because detecting a
first- or higher-level noncoherence event requires a very considerable number
of analyses for double and higher-level contingencies, while in most cases
only single-level contingency analysis is emphasized in utility system plan-
ning guidelines. This is also associated with the historical development pro-
cess of a system. Some parts of a network may also become "overlooped"
in some system operating states when other parts are reinforced, and such
situations cannot be found easily by traditional means. This case can happen
particularly in lower-voltage transmission systems (230 kV and below). Gen-
erally, second- and higher-level noncoherent events have relatively small
effects on total system reliability indices due to low probabilities of occur-
rence. First-level noncoherence, however, can create a very significant impact
in some cases.
(b) M~asure of Noncoherence. The following noncoherence
degree indices can be used to measure degree of noncoherence:
NCDE = (EENS - #EENS)/EENS (5.57)
NCDF = (EFLC- #EFLC)/EFLC (5.58)
COMPOSITE SYSTEM ADEQUACY ASSESSMENT 201

NCDP = (PLC - #PLC)/PLC (5.59)

where
EENS = Expected Energy Not Supplied
EFLC = Expected Frequency of Load Curtailments
PLC = Probability of Load Curtailments
NCDE = Noncoherence Degree of EENS
NCDF = Noncoherence Degree of EFLC
NCDP=Noncoherence Degree ofPLC.
In equations (5.57) to (5.59), the reliability indices with prefix # are
those excluding the effects of noncoherence while the reliability indices
without prefix # are those including the effects of noncoherence, Le.,
obtained using a reliability evaluation method without the coherence
assumption. Note that the reliability indices excluding the effects of nonco-
herence are not the same as those obtained using a reliability evaluation
method with the coherence assumption. They are completely different. In
order to calculate the reliability indices excluding effects of noncoherence,
the following steps should be conducted (first-level noncoherence events are
taken as examples) :
l. For each single component failure, open one or more transmission
components intentionally (this is referred to failure-dependent disconnec-
tion) and conduct overload and/or load curtailment analysis.
2. If an intentional opening leads to less overloads or load curtailments
compared to just the single component failure, this is a first-level noncoher-
ence event. Record all such events and intentionally disconnected corre-
sponding components.
3. Evaluate system reliability excluding the effects of noncoherence.
This means that each time a single component failure associated with nonco-
herence is selected by state enumeration or sampled by Monte Carlo simula-
tion, corresponding intentional disconnections are imposed. This requires
that a composite system reliability evaluation program has a function to
simulate dependent failures.
Calculation steps associated with higher-level noncoherence events are
similar, but require much more computational effort due to analyses of
very considerable contingencies. Considering that higher-level transmission
failures have small probabilities of occurrence, it is generally acceptable to
neglect effects of the second- and higher-level noncoherence events when
computational time is limited.
Equations (5.57) to (5.59) show three noncoherence degree indices asso-
ciated with the reliability indices EENS, EFLC, and PLC. Noncoherence
degree indices relating to other reliability indices can also be calculated in
202 CHAPTER 5

(1) ............

Figure 5.19. A three-bus system.

the same way. These indices range between 0.0 and 1.0. Essentially, they are
percentages of contributions of noncoherence with respect to total system
indices including effects of noncoherence. Generally, they cannot be equal
to 1.0 but can be zero. Large values of the indices denote severe noncoher-
ence. Zero is a critical value. It indicates that the system is coherent in terms
of the strict noncoherence definition. It can also be said, however, that the
system is noncoherent at index zero if the loosened noncoherence definition
is used. Different noncoherence degree indices may give different indications.
For example, it is possible that the NCDE index of a system can be close
to 1.0 while the NCDF or the NCDP of the same system can be close to
zero. This means that the system is quite noncoherent from the EENS index
point of view but almost coherent from the EFLC or the PLC index point
of view. This case can happen when intentional failure-dependent disconnec-
tions basically do not alleviate overloads or load curtailments but only
reduce the extent of overloads or load curtailments. Therefore coherence or
noncoherence is also a reliability-index-dependent concept.

5.10.2. A Simple Example

A simple example is used to illustrate the definitions and concepts given
in the Section 5.10.1. Figure 5.19 shows a three-bus system. Two generating
plants located in a generation center provide power to a load center at Bus
COMPOSITE SYSTEM ADEQUACY ASSESSMENT 203

2 and local loads at Bus 3. The load center is relatively far away from the
generation center while the two generating plants are quite close. Therefore,
the impedance of Line 2 is much smaller than that of Line I or Line 3. Data
are given in Tables 5.41 to 5.43. The base value is 100.0 MVA.
The contingency state enumeration technique is used to evaluate the
system reliability. Contingencies are enumerated to the third level, i.e., single,
double, or triple components can fail at the same time. It is unnecessary
to consider the fourth- and fifth-level contingencies because of very low
probabilities of occurrence. Out of the five first-level contingencies, only one
(loss of G2) requires load curtailments in order to alleviate overloads on
Line 2. In the case of the loss of G2, the load flow equation forces the power
flow on Line 2 to be higher than its rating. Four out of the ten second-level
contingencies and nine out of the ten third-level contingencies require load
curtailments. The results associated with load curtailment events are listed

Table 5.41. Bus Data of the Three-Bus System

Load
Bus P Q Specified voltage
No. Type (p.u.) (p.u.) (p.u.)
1 Swing 0.00 0.00 l.0
2 PQ 0.45 0.20
3 PV 0.45 0.20 l.0

Table 5.42. Branch Data of the Three-Bus System

Line R X Rating
No. From To (p.u.) (p.u.) (MVA)
1 1 2 0.008 0.034 50.0
2 1 3 0.001 0.003 50.0
3 2 3 0.008 0.034 50.0

Table 5.43. Component Failure Probabilities

Component Forced unavailability
Gl 0.05
G2 0.05
Line 1 0.001
Line 2 0.0005
Line 3 0.001
204 CHAPTER 5

Table 5.44. Results Associated with Load Curtailment Events

Element Load curtailment EENS
Event out (MW) PLC (MWh/yr)
I G2 25 0.047381345 10376.5146
2 Gl+G2 90 0.002493755 1966.0764
3 G2+L2 45 0.000023703 9.3437
4 G2+Ll 45 0.000047429 18.6965
5 Ll+L3 45 0.000000902 0.3556
6 Gl+G2+Ll 90 0.000002496 1.9680
7 Gl+G2+L2 90 0.000001248 0.9835
8 Gl+G2+L3 90 0.000002496 1.9680
9 Gl+Ll+L3 45 0.000000047 0.0185
10 Gl+L2+L3 45 0.000000024 0.0095
11 G2+Ll+L2 90 0.000000024 0.0189
12 G2+Ll+L3 45 0.000000047 0.0185
13 G2+L2+L3 45 0.000000024 0.0095
14 Ll+L2+L3 90 4.5E-I0 0.0004

in Table 5.44. The system states without load curtailments are not listed.
Among all these load curtailment contingencies, loss of G2 is a unique
noncoherence event. In the case of losing G2, if Line 3 is intentionally
opened, the overloads on Line 2 will disappear and consequently 25 MW of
active load curtailments at Bus 3 are no longer required.
Adding event contributions to the PLC or EENS yields a total system
PLC or EENS index which includes the effect of noncoherence. In order to
obtain the PLC or EENS index excluding the effect of noncoherence, it is
necessary to deduct the contribution due to loss of G2 (Event 1) from the
total PLC or EENS index. Generally, the deduction should be the difference
between the contribution due to the G2 loss event and that due to the event
of loss of G2 plus intentional opening of Line 3. In this particular case, loss
of G2 plus opening of Line 3 does not require any load curtailment. It can
be seen from Table 5.44 that although loss of G2 is a unique noncoherence
event, it is also a unique first level load curtailment event and its contribution
to the total indices is dominant. It leads to a high degree of system noncoher-
ence. Two noncoherence degree indices of the three-bus system are as
follows:
NCDE=0.838
NCDP = 0.948
It is important to emphasize that since a noncoherence event is system
state dependent, the intentional opening leading to a more reliable system
state does not mean that the opened component is surplus and should be
removed permanently. The intentional opening is only beneficial for this
COMPOSITE SYSTEM ADEQUACY ASSESSMENT 205

particular system state. In the above example, for instance, opening Line 3
is only beneficial for the loss of the G2 event. It cannot be concluded that
Line 3 should also be removed in other system states.

5.11. CONCLUSIONS
This chapter discusses the application of Monte Carlo methods in com-
posite system adequacy assessment. The main difference between Monte
Carlo simulation methods and state enumeration approaches is in the selec-
tion of the system states. Therefore, the system analysis techniques for each
selected state, such as contingency analysis techniques and optimization
models for corrective actions given in Sections 5.2 and 5.9, can also be
applied in enumeration approaches. Monte Carlo methods can be used to
consider complex practical factors associated with the selection of system
states. For example, regional weather effects and bus load uncertainty and
correlation described in Sections 5.5 and 5.7 may be very difficult, if not
impossible, to incorporate using enumeration approaches. When generating
unit derated states and transmission line common cause outages are consid-
ered, the system state sampling methods given in Sections 5.4 and 5.6 require
less calculations than an enumeration technique. If the annual load curve is
incorporated to calculate annual indices, however, combining the enumera-
tion of multistep load levels with Monte Carlo simulation of component
outages is more computationally efficient than sampling both load levels
and component outages. This hybrid method is given in Section 5.3 and is
essentially a variance reduction technique from a Monte Carlo simulation
point of view.
The major advantage of the system state sampling method is reduced
CPU time and memory requirements. This method can provide accurate
probability and energy/power related indices. It can, however, only provide
approximate frequency-related indices. This is because the transitions
between system states and the chronology of the annual load curve are
ignored. Theoretically, the system state duration sampling method can be
used to calculate actual frequency indices. In the case of composite systems,
however, there exist two obstacles. Considering chronological load curves
of all buses creates a very large computational burden, which may be unfeas-
ible at the present time for a practical composite system. Collecting chrono-
logical load data at all buses is also a very difficult task. The system state
transition sampling method described in Section 5.8 provides a compromise.
It creates a system state transition sequence but still uses the multistep load
level model without load chronology. This method therefore can be used to
calculate actual annualized frequency indices. If a multiload-Ievel transition
206 CHAPTER 5

model is utilized, the method can also be extended to include the effect of
load state transitions on the frequency indices. However, this is still an
approximation of actual annual frequency indices. There is a quite large
difference between a multiload-Ievel transition model and the actual chrono-
logical bus load curves.
Section 5.9 discusses security considerations in composite system
adequacy assessment. This is not "security evaluation" involving system
dynamic disturbances and a cascade collapse process. The basic idea is to
incorporate steady-state security constraints in evaluation of operating state
probability indices. In the operating practices of many utilities, transient
and voltage stability limits are usually expressed using limit curves in genera-
tion space or power flow limits of crucial lines. If these limits are introduced
in the simulation models, then transient and voltage stability are indirectly
considered in the reliability evaluation. This may be a good replacement for
detailed security evaluation, which reflects utility operational guidelines.
The definitions and indices of noncoherence in composite system
adequacy assessment are discussed in Section 5.10. Noncoherence is system
state dependent and reliability index dependent. Noncoherence of a compos-
ite system is not only theoretically meaningful, but also exists in actual power
systems and has been detected in the Be Hydro South Vancouver Metro
system.(42) A general composite system reliability evaluation program should
not be based on the automatic assumption of coherence.

5.12. REFERENCES
I. IEEE Committee Report, "Bibliography on the Application of Probability Methods in
Power System Reliability Evaluation," IEEE Trans. on PAS, Vol. 91, 1972, pp. 649-660.
2. R. N. Allan, R. Billinton, and S. H. Lee, "Bibliography on the Application of Probability
Methods in Power System Reliability Evaluation, 1977-1982," IEEE Trans. on PAS,
Vol. 103, 1984, pp. 275-282.
3. R. Billinton and R. N. Allan, Reliability Evaluation of Power Systems, Plenum, New York,
1984.
4. R. Billinton and R. N. Allan, Reliability Assessment of Large Electric Power Systems,
Kluwer, Boston, 1988.
5. M. Th. Schilling, R. Billinton, A. M. Leite de Silva, and M. A. EI-Kady, "Bibliography
on Composite System Reliability (1964-1988)," IEEE Trans. on PS, Vol. 4, No.3, 1989,
pp. 1122-1131.
6. L. Salvaderi, R. N. Allan, R. Billinton, J. Endrenyi, D. Mcgills, M. Lauby, P. Manning,
and R. J. Ringlee, "State of the Art of Composite System Reliability Evaluation," 1990
CIGRE Meeting, WG38.03, Paris, Sept. 1990.
7. IEEE Tutorial Course Text, Reliability Assessment of Composite Generation and Trans-
mission Systems, IEEE Publishing Services, 9OEH0311-1-PWR, 1990.
8. R. Billinton, "Composite System Reliability Evaluation," IEEE Trans. on PAS, Vol. 88,
1969, pp. 276-280.
COMPOSITE SYSTEM ADEQUACY ASSESSMENT 207

9. R. Billinton, "Elements of Composite System Reliability Evaluation," CEA Trans., Vol. 15,
Pt. 2, 1976, Paper No. 76-SP-148.
10. P. L. Dandeno, G. E. Jorgensen, W. R. Puntel, and R. J. Ringlee, "Program for Composite
Bulk Power Electric System Adequacy Assessment," Trans. of the lEE Conference on
Reliability of Power Supply Systems, lEE Conference Publication No. 148, Feb. 1977.
II. G. E. Marks, "A Method of Combining High Speed Contingency Load Flow Analysis
with Stochastic Probability Methods to Calculate a Quantitative Measure of Overall Power
System Reliability," IEEE Paper A78-053-01.
12. EPRI Report, "Transmission System Reliability Methods," Project RP 1530-1, EL2526,
1982.
13. R. Billinton and T. K. P. Medicheria, "Overall Approach to the Reliability Evaluation of
Composite Generation and Transmission Systems," lEE Proc., Vol. 127, Pt. C, No.2,
1980.
14. A. P. Meliopoulos, "Bulk Power Reliability Assessment with the RECS Program," Pro-
ceedings 1985 PICA Conference, pp. 38-46.
15. R. Billinton, "Composite System Adequacy Assessment-The Contingency Evaluation
Approach," IEEE Tutorial Course Text, 90EH0311-I-PWR, Feb. 1990.
16. J. C. Dodu and A. Merlin, "An Application of Linear Programming to the Planning
of Large Scale Power Systems: The MEXICO Model," Proceeding of the 5th PSCC,
Cambridge, England, Sept. 1975.
17. P. Noferi, L. Paris, and L. Salvaderi, "Monte Carlo Methods for Power System Reliability
Evaluation in Transmission and Generation Planning," Proceedings 1975 Annual Reli-
ability & Maintainability Symposium, Washington D.C., Jan. 1975, Paper 1294, pp. 449-
459.
18. P. Noferi and L. Paris, "Effect of Voltage and Reactive Power Constraints on Power
System Reliability," IEEE Trans. on PAS, Vol. 94, 1975, pp.482--490.
19. A. Merlin and P. H. Oger, "Application of a Variance Reducing Technique to a Monte
Carlo Simulation Model of Power Transmission Systems," pp. 3-35, 3--44, Workshop Pro-
ceedings: Power System Reliability-Research Needs and Priorities, EPRI Report WS77-
60, October, 1978.
20. R. Billinton and L. Salvaderi, "Comparison Between Two Fundamentally Different
Approaches to Composite System Reliability Evaluation," IEEE Trans. on PAS, Vol. 104,
1985, pp. 3486-3492.
21. S. H. F. Cunha, M. V. F. Pereira, G. C. Oliveira, and L. M. V. G. Pinto, "Composite
Generation and Transmission Reliability Evaluation in Large Hydroelectric Systems,"
IEEE Trans. on PAS, Vol. 104, No.2, 1985, pp. 2657-2663.
22. EPRI Report, Composite System Reliability Evaluation Methods, Report EL-5178, June
1987.
23. EPRI Report, Development of a Monte Carlo Composite System Reliability Evaluation
Programming, Report EI-6926, Aug. 1990.
24. R. Billinton and W. Li, "Hybrid Approach for Reliability Evaluation of Composite Genera-
tion and Transmission Systems Using Monte Carlo Simulation and Enumeration Tech-
nique," lEE Proc., Vol. 138, Pt. C, No.3, 1991, pp. 233-241.
25. R. Billinton and W. Li, "A Novel Method for Incorporating Weather Effects in Composite
System Adequacy Evaluation," IEEE Trans. on PS, Vol. 6, No.3, 1991, pp. 1154-1160.
26. W. Li and R. Billinton, "Effect of Bus Load Uncertainty and Correlation in Composite
System Adequacy Evaluation," IEEE Trans. on PS, Vol. 6, No.4, 1991, pp. 1522-1529.
27. R. Billinton and W. Li, "Consideration of Multi-State Generating Unit Models in Compos-
ite System Adequacy Assessment Using Monte Carlo Simulation," Canadian Journal of
Electrical and Computer Engineering, Vol. 17, No. I, 1992.
208 CHAPTER 5

28. R. Billinton and W. Li, "Composite System Reliability Assessment Using a Monte Carlo
Approach," The Third International Conference on Probabilistic Methods Applied to
Electric Power Systems, London, 3-5 July, 1991.
29. R. Billinton and W. Li, "Direct Incorporation of Load Variations in Monte Carlo Simula-
tion of Composite System Adequacy," Inter-RAMQ Conference for the Electric Power
Industry, Philadelphia, PA, Aug. 25-28, 1992.
30. R. Billinton and W. Li, "A System State Transition Sampling Method for Composite
System Reliability Evaluation," IEEE Trans. on PS, Vol. 8, No.3, 1993, pp. 761-770.
31. W. Li, "An On-Line Economic Power Dispatch Method with Security," Electric Power
Systems Research, Vol. 9, 1985, pp. 173-181.
32. W. Li, "A Successive Linear Programming Model for Real-Time Economic Power Dispatch
with Security," Electric Power Systems Research, Vol. 13, 1987, pp. 225-233.
33. IEEE Committee Report, "IEEE Reliability Test System," IEEE Trans. on PAS, Vol. 98,
1979, pp. 2047-2054.
34. R. Billinton and S. Kumar, "Effect of High-Level Independent Generation Outages in
Composite System Adequacy Evaluation," Proc. lEE, Vol. 134, No.1, 1987, pp. 17-26.
35. R. Billinton, S. Kumar, N. Chowdhury, K. Chu , K. Debnath, L. Goel, E. Khan, P. Kos,
G. Nourbakhsh, and J. Oteng-Adjei, "A Reliability Test System for Educational Purpose:
Basic Data," IEEE Trans. on PS, Vol. 4, 1989, pp. 1238-1244.
36. R. Billinton and K. E. Bollinger, "Transmission System Reliability Evaluation Using
Markov Processes," IEEE Trans. on PAS, Vol. 87, 1968, pp. 538-547.
37. K. A. Clements, B. P. Lam, D. J. Lawrence, T. A. Mikolinnas, A. D. Reppen, R. J.
Ringlee, and B. F. Wollenberg, "Transmission System Reliability Methods-Mathematical
Models, Computing Methods and Results," EPRI EL-2526, Power Technologie Inc.,
Schenectady, New York, July 1982.
38. M. E. Khan and R. Bil1inton, "A Hybrid Model for Quantifying Different Operating States
of Composite Power Systems," IEEE Trans. on PS, Vol. 7, No.1, 1992, pp. 187-193.
39. EPRI Report, "Composite System Reliability Evaluation: Phase l-Scoping Study," Final
Report, EPRI EL-5290, Dec. 1987.
40. M. E. Khan, "A Security Approach to Composite Power System Reliability Evaluation,"
Ph.D. Thesis, University of Saskatchewan, 1991.
41. R. Billinton and M. E. Khan, "A Security Based Approach to Composite System Reliability
Evaluation," IEEE Trans. on PS, Vol. 7, No.1, 1992, pp. 65-72.
42. W. Li, "Reliability Assessment of the South Metro System Planning Alternatives," BC
Hydro, RSP Technical Report, File 940-R-5, Aug. 10, 1992.
 6
Distribution System
and Station
Adequacy
Assessment

6.1. INTRODUCTION
This chapter discusses the application of Monte Carlo methods to distribu-
tion system and station adequacy assessment. The basic system analysis
logic associated with distribution system and station configuration adequacy
assessment is very similar. The main approach is to examine the existence
of continuity between the supply point(s) and the load points. Existence of
continuity is associated not only with failures of line components, but also
with the switching logic of breakers and section switches, protection device
actions, and operating guidelines for backup supply sources. This is obvi-
ously quite different from the system analysis principles associated with
generation adequacy or composite system adequacy assessment.
Both the state sampling and state duration sampling techniques can be
applied to distribution system and station configuration adequacy assess-
ment. The state duration sampling technique, however, is preferable as this
approach can be used to calculate actual frequency and duration indices
using the chronological component or system state transition cycles. Fre-
quency and duration indices are particularly important in distribution system
and station adequacy assessment. Reliability evaluation of a distribution
system or a voltage step-down station is usually associated directly with the
customer load point performance. Most customers are very concerned about
209
210 CHAPTER 6

how often outages are expected to occur and how long they are expected to
last.
Distribution system and station reliability assessment can be con-
ducted quite successfully using analytical methodsY-13) The concepts and
techniques associated with the analytical method are both useful and
helpful in understanding the application of Monte Carlo methods. Some
techniques are, in fact, a combination of the Monte Carlo method and
the analytical approach. The basic concepts of an analytical approach to
distribution system reliability assessment are first described in Section 6.2.
Application of the component state duration sampling method to distribu-
tion system and station configuration adequacy assessment is illustrated
in Sections 6.3 and 6.4, respectively. One advantage of the Monte Carlo
method is that it can provide information related to the probability
distributions of the reliability indices in addition to the mean or average
values.(14,IS) Simulation can also be used to recognize complex component
repair or restoration time distributions. These aspects are quite difficult, if
not impossible, to incorporate into an analytical method without assump-
tions or simplifications.

6.2. BASIC CONCEPTS AND ANALYTICAL

TECHNIQUES FOR DISTRIBUTION
SYSTEM ADEQUACY ASSESSMENT

6.2.1. General Concepts

Quantitative reliability evaluation of a distribution system can be
divided into the two basic segements of measuring past performance and
predicting future performance. (13) Most utilities collect data on past system
performance and display the results using a range of statistics. Some of
the basic indices used to assess past performance are the system average
interruption frequency index (SAIFI), the system average interruption dura-
tion index (SAlOl), the customer average interruption duration index
(CAIDI), the average service availability index (ASAI), the average service
unavailability index (ASUI), and the energy not supplied (ENS).(2) Pre-
dictive reliability assessment of a distribution system is usually concerned
with the performance at the customer load points. The basic indices normally
used are load point failure rate, average outage duration, and annual
unavailability. The basic indices are important with respect to a particular
load point, but they do not give an overall appreciation of the area or system
DISTRIBUTION SYSTEM AND STATION ADEQUACY ASSESSMENT 211

Table 6.1. Average Annual Canadian

Indices
SAIFI SAIDI
Year (int./syst. cust.) (hr/syst. cust.) lOR
1988 4.35 6.42 0.999267
1989 3.61 4.35 0.999503
1990 4.03 4.55 0.999480
1991 3.55 4.24 0.999516
1992 3.06 3.34 0.999619

performance. Additional indices can be calculated using these three basic

indices and the number of customers/loads connected at each load point.
These additional indices are weighted averages of the basic load point indices
and are basically identical to those which have been used for many years to
assess past system performance. The most common additional indices are
SAIFI, SAIDI, CAIDI, ASAI, and ASUI. The procedure for calculating
these indices is given in Section 2.6.3.
Canadian electric power utilities have been active in collecting service
continuity statistics for over 30 years and produced their first national com-
pilation in 1962. This work is now performed by the Canadian Electrical
Association (CEA). The average service availability index (ASAI) is known
as the index of reliability (lOR) in the CEA reporting system. Table 6.1
shows the aggregate Canadian values for SAIDI, SAIFI, and lOR for the
1988-1992 period. These data give a general indication of Canadian utility
performance over the 5-yr period. Distribution engineers are acutely con-
scious of the contribution made by the distribution facilities toward the
overall service to the customer and normally subdivide the interruptions into
categories which can be used to identify causes and the need for future
expenditures. This is illustrated in Tables 6.2 and 6.3, which show the average
cause contributions to the overall annual SAIDI and SAIFI statistics taken
from the 1988-1992 period. Values in parentheses are the percentage contri-
butions to the annual value. It can be seen from Tables 6.2 and 6.3 that the
category "loss of supply" contributes 23.5% and 17.1 % to the overall average
SAIFI and SAIDI values, respectively. This supports the general statement
that approximately 80% of all customer interruptions occur due to failures
in the distribution system. Most utilities compile service continuity statistics
on individual feeders, groups of feeders, districts and areas in their systems.
These data are used for a wide variety of managerial and engineering
functions. (6,\3)
Past performance statistics provide a valuable reliability profile of the
existing system. Distribution planning, however, involves the analysis of
future systems and the evaluation of system reliability when the configura-
212 CHAPTER 6

Table 6.2. Cause Contributions to the

Annual SAIFI (int./syst. cust.)
Cause Average
Unknown/other 0.32 (8.5)
Scheduled interruption 0.75 (20.0)
Loss of supply 0.88 (23.5)
Tree contact 0.25 (6.7)
Lightning 0.22(5.9)
Defective equipment 0.72 (19.2)
Adverse weather 0.21 (5.6)
Adverse environment 0.03 (0.8)
Human element 0.10 (2.7)
Foreign interference 0.27 (7.2)
Total 3.75

Table 6.3. Cause Contributions to the

Annual SAlOl (hr./syst. cust.)
Cause Average
Unknown/other 0.29 (6.4)
Scheduled interruption 0.69 (15.3)
Loss of supply 0.77 (17.1)
Tree contact 0.61 (13.5)
Lightning 0.32 (7.1)
Defective equipment 0.91 (20.2)
Adverse weather 0.59 (13.1)
Adverse environment 0.05 (1.1)
Human element 0.06 (1.3)
Foreign interference 0.22 (4.9)
Total 4.51

tion, or the operating conditions, or the protection scheme changes. This

task is known as predictive reliability assessment, in which the distribution
system reliability indices are calculated using component failure data.

6.2.2. Basic Distribution Systems

The distribution system is that portion of the electric power system
which links the bulk power source(s) to the consumer's facilities. Sub-
transmission circuits, distribution substation, primary feeders, distribution
transformers, secondary circuits, and consumer's connections form different
parts of what can generally be called a distribution system. Distribution
system reliability evaluation therefore consists of assessing how adequately
DISTRIBUTION SYSTEM AND STATION ADEQUACY ASSESSMENT 213

the different parts are able to perform their intended function. The distribu-
tion system is an important part of the total electric system as it provides
the link between the bulk system and the customer.
A distribution circuit normally uses primary or main feeders and lateral
distributors. A main feeder originates from the substation and passes
through the major load centers. The individual load points are connected
to the main feeder by lateral distributors with distribution transformers at
their ends. A main feeder is constructed using single, parallel, or meshed
circuits. Many distribution systems used in practice have a single-circuit
main feeder and are defined as radial distribution systems. There are also
many systems which, although constructed using meshed circuits, are opera-
ted as radial systems using normally open switches in the meshed circuit.
The main feeder, in some cases, may have branches to reach the widely
distributed areas.
Radial systems are popular because of their simple design and generally
low cost. These systems have a set of series components between the sub-
station and the load points. The failure of any of these components causes
outage of the load point(s). The outage duration and the number of cus-
tomers affected due to a component failure are reduced by using extensive
protection and sectionalizing schemes. Sectionalizing equipment provides a
convenient means of isolating a faulted section. The supply can then be
restored to the healthy sections, maintaining the service to some of the load
points, while the faulted component is repaired. The time taken by this type
of isolation and switching action is referred to as restoration time.
In some systems, there is provision for an alternate supply in the case
of a failure. This alternate source is used to supply that section of the main
feeder which becomes disconnected from the main supply after the faulted
section has been isolated. The alternate supply, however, may not always
be available and this factor should be included in the analysis. Fuses are
usually provided on the lateral distributors. Faults on a lateral distributor
or in a distribution transformer are normally cleared by a fuse and therefore
service on the main feeder is maintained. If the fuse fails to clear the fault
for some reason, the circuit breaker or the back-up fuse on the main feeder
acts to clear the fault. The faulted lateral distributor is then isolated and the
supply is restored to the rest of the system by switching action.

6.2.3. Analytical Techniques

It is useful to illustrate the basic analytical approach to distribution
system assessment before discussing the proposed Monte Carlo simulation
method. Analytical techniques can be readily used to calculate load point
214 CHAPTER 6

Alternate Supply

r---~~~-----B

'--'--.....;;;..;.;.~--
.......
I1 mile N/O

~c
Figure 6.1. A simple radial configuration.

and system average performance indices. (1-3) An important benefit of the

Monte Carlo method is that it can be used to provide probability distribu-
tions of these indices. The basic logic required to analyze the system failure
modes is the same for both the analytical approach and the Monte Carlo
method.
As noted earlier, the basic indices used to predict the reliability of a
distribution system are the average load point failure rate(Jt), the average
load point outage duration (r), and the average annual load point outage
time or unavailability (U). The system performance indices SAIDI, SAIFI,
CAIDI, etc. can be calculated from the three basic predictive indices. In
order to illustrate the basic analytical technique, an example calculation is
presented for the simple radial system shown in Figure 6.1. The following
section presents and compares the results of Monte Carlo simulation for the
same system. In this configuration, all switches except that connected to the
alternate supply are normally closed, and the feeder breaker and substation
supply bus are assumed to be fully reliable. In Case 1, the customer load
points A, B, and C are supplied from the primary main feeder by fused
laterals. The alternate supply shown in Figure 6.1 is assumed to be unavail-
able in Case I. The individual component data are as follows:
Primary main feeder
0.1 failures/circuit mile/yr, 3.0 hr average repair time
Primary lateral
0.25 failures/circuit mile/yr, 1.0 hr average repair time
Manual sectionalizing time for any switching action is 0.5 hr.
DISTRIBUTION SYSTEM AND STATION ADEQUACY ASSESSMENT 215

Table 6.4. Calculations for Case 1

Load point A Load point B Load point C
A- r AT A- r AT A- r AT
Component (f/yr) (hr/f) (hr/yr) (f/yr) (hr/f) (hr/yr) (f/yr) (hr/f) (hr/yr)
Main sections
2 miles 0.2 3.0 0.6 0.2 3.0 0.6 0.2 3.0 0.6
3 miles 0.3 0.5 0.15 0.3 3.0 0.9 0.3 3.0 0.9
1 mile 0.1 0.5 0.05 0.1 0.5 0.05 0.1 3.0 0.3
Lateral sections
3 miles 0.75 1.0 0.75
2 miles 0.5 1.0 0.5
1 mile 0.25 1.0 0.25
1.35 1.15 1.55 1.1 1.86 2.05 0.85 2.41 2.05

The basic analytical approach involves a failure mode and effect analysis
utilizing the following basic equations:
(6.1)
LA.iri
r=-- (6.2)
S L A.i
~=~~ ~~
This procedure for Case I is illustrated in Table 6.4. The results are summar-
ized in Table 6.5.
There are many configurations, particularly in rural locations, which
have a topology similar to that shown in Figure 6.1. The results shown in
Table 6.5 can be used to obtain the system performance indices. The calcula-
tion formulas for these indices are given in Section 2.6.3. Assume that there
are 250, 100, and 50 customers, respectively, at load points A, B, and C,
giving a total of 400 customers in the system:
Annual Customer Interruptions
= (250)(1.35) + (100)(1.1) + (50)(0.85) =490
Customer Interruption Duration
= (250)(1.55) + (100)(2.05) + (50)(2.05) = 695

Table 6.5. Calculated Indices for Case 1

Index A B C
A- (failures/yr) 1.35 1.10 0.85
r (hr/failure) 1.15 1.86 2.41
U(hr/yr) 1.55 2.05 2.05
216 CHAPTER 6

Table 6.6. Calculations for Case 2

Load point A Load point B Load point C
A r Ar A r Ar A r Ar
Component (f/yr) (hr/f) (hr/yr) (f/yr) (hr/f) (hr/yr) (f/yr) (hr/f) (hr/yr)
Main sections
2 miles 0.2 3.0 0.6 0.2 1.0 0.2 0.2 1.0 0.2
3 miles 0.3 0.5 0.15 0.3 3.0 0.9 0.3 1.0 0.3
I mile 0.1 0.5 0.05 0.1 0.5 0.05 0.1 3.0 0.3
Lateral sections
3 miles 0.75 1.0 0.75
2 miles 0.5 1.0 0.5
1 mile 0.25 1.0 0.25
1.35 1.15 1.55 1.1 1.5 1.65 0.85 1.24 1.05

System Average Interruption Frequency Index (SAIFI)

SAIFI = 490/400 = l.23 interruptions/system customer/yr
System Average Interruption Duration Index (SAIDI)
SAIDI = 695/400 = 1.74 hr/system customer/yr
Customer Average Interruption Duration Index (CAIDI)
CAIDI = 695/490 = l.42 hr/customer interrupted
Average Service Availability Index (ASAI)
ASAI = (400 x 8760 - 695) / (400 x 8760) = 0.999802.
The calculated values can be compared with measured values or, if
available, with standard indices for the system to determine if the configura-
tion is acceptable.
Similar calculations can be conducted to investigate the effect on the
load point and system indices of configuration and operating changes. The
calculations for two additional cases are given in Tables 6.6 and 6.8.
In Case 1, the alternate supply shown in Figure 6.1 is assumed to be
unavailable. This facility is available in Case 2, with an average switching

Table 6.7. Calculated Indices for Case 2

Index A B C
A (failures/yr) 1.35 1.10 0.85
r (hr/failure) 1.15 1.50 1.24
U (hr/yr) 1.55 1.65 1.05
System performance indices
SAIFI= 1.23 interruptions/system customer/yr
SAIDI= 1.51 hr/system customer/yr
CAIDI = 1.23 hr/customer interrupted
ASAI = 0.999827
DISTRIBUTION SYSTEM AND STATION ADEQUACY ASSESSMENT 217

Table 6.S. Calculations for Case 3

Load point A Load point B Load point C
A- r A.r A- r A.r A- r A.r
Component (f/yr) (hr/f) (hr/yr) (f/yr) (hr/f) (hr/yr) (f/yr) (hr/f) (hr/yr)
Main sections
2 miles 0.2 3.0 0.6 0.2 3.0 0.6 0.2 3.0 0.6
3 miles 0.3 0.5 0.15 0.3 3.0 0.9 0.3 3.0 0.9
I mile 0.1 0.5 0.05 0.1 0.5 0.05 0.1 3.0 0.3
Lateral sections
3 miles 0.75 1.0 0.75 0.75 1.0 0.75 0.75 1.0 0.75
2 miles 0.5 0.5 0.25 0.5 1.0 0.5 0.5 1.0 0.5
I mile 0.25 0.5 0.125 0.25 0.5 0.125 0.25 1.0 0.25
2.10 0.92 1.93 2.10 1.39 2.93 2.10 1.57 3.30

time of one hour. Tables 6.6 and 6.7 illustrate the effect of this alternate
supply on the calculated reliability indices. It can be seen that the load point
failure rates are not affected by the ability to backfeed from an alternate
configuration. This will apply in all cases in which the restoration of service
is done manually. If automatic switching is used and the customer outage
time is considered to be so short that the event is not classed as a failure,
then the overall failure rate will be reduced to a value close to the primary
lateral value. The ability to backfeed has a pronounced effect on the length
of the interruption, particularly for those customers at the extremities of the
primary main.
The load point failure rates are highly dependent upon the components
exposed to failure and the degree of automatic isolation of a failed compo-
nent in the network. Case 3 is a situation in which the alternate supply is
unavailable and each lateral is solidly connected to the primary main. In
this case, all load points will have the same failure rate, as any fault will
result in the feeder breaker tripping. The analysis for this case is shown in
Table 6.8. The load point and system results are given in Table 6.9.
Table 6.9. Calculated Indices for Case 3
Index A B C
A- (failures/yr) 2.10 2.10 2.10
r (hr/failure) 0.92 1.39 1.57
U(hr/yr) 1.93 2.93 3.30
System performance indices
SAIFI=2.1O interruptions/system customer/yr
SAIDI=2.35 hr/system customer/yr
CAIDI= 1.12 hr/customer interrupted
ASAI = 0.999732
218 CHAPTER 6

6.3. MONTE CARLO SIMULATION

APPROACH TO DISTRIBUTION
SYSTEM ADEQUACY ASSESSMENT

6.3.1. Probability Distribution

Considerations
The analytical techniques given in Section 6.2.3 can be readily used to
calculate the expected or average value of the particular reliability measure.
The expected values are extremely useful and are the primary indices of load
point adequacy. There is, however, an increasing awareness of the need
for information related to the variation of these reliability measures.(IS-17)
Probability distributions provide a practical means of describing this. This
information can prove useful in studies involving:
1. The determination of customer costs of interruptions using nonlinear
cost functions. (16)
2. The validation of reliability models and the applicability of reliability
data. The index probability distributions can prove useful in estimat-
ing the errors resulting from inaccurate data.
3. Comparisons between performance indices of different years or
different systems to determine the probability of their having differ-
ent average values. Such a comparison can assist planners to judge
whether differences in indices indicate real changes in performance
or are due to statistical variations.
4. The probability of the interruptions being longer than the Critical
Service Loss Duration Time or some other time of interest. (18-20) This
information is especially useful in the design of distribution systems
for industrial customers with critical processes or commercial cus-
tomers with nonlinear cost functions.

Studies of the distributions associated with the basic reliability indices

indicate that the load point failure rate is approximately Poisson-
distributed.(17,21) The failure probability therefore can be readily obtained
using the mean load point failure rate, since the Poisson distribution is a
single-parameter function. This information can be calculated directly using
the Poisson equation or from a set of published graphs.(22) It has been
noted that if the restoration times of components can be assumed to be
exponentially distributed, the load point outage duration is approximately
gamma-distributed and the desired probability information can be readily
calculatedY7,21) There are, however, many distribution systems for which
DISTRIBUTION SYSTEM AND STATION ADEQUACY ASSESSMENT 219

the gamma distribution does not adequately describe the load point outage
duration. These systems are typically those in which the restoration times
may be better described by nonexponential distributions, e.g., log-normal
repair or manual sectionalizing times. When the restoration times are
assumed to be nonexponential, the interruption duration cannot generally
be represented by a gamma distribution.{I3-15) Probability distributions for
the annual load point outage time, SAIDI, SAIFI, and CAIDI indices can
also not be represented by common distributions.

6.3.2. Component State Duration Sampling

Method
In distribution system reliability assessment, the three basic load point
indices are failure rate, average outage duration, and average annual outage
time. As noted in Section 3.6.2, the component state duration sampling
method can be used to calculate actual frequency and duration indices,
recognizing component state duration distributions, and to provide reli-
ability index probability distributions. The following two techniques based
on the component state duration sampling can be applied to distribution
system reliability assessment.
(a) Individual Component Operating Cycle Technique. A
customer connected to any load point in a radial distribution system requires
that all components between the load and the supply point be operating.
Consequently, the principle of series systems can be applied. Equations (6.1)
to (6.3) are based on this principle. The individual component operating
cycle technique is a combination of the component state duration sampling
and the analytical method given in Section 6.2.3. This technique includes
the following steps:
1. Artificial up-down operating histories of all line sections are
obtained by component state duration sampling.
2. The effects of the operating state (up or down) of each line section
on a load point state are examined individually, based on the system config-
uration. Possible modifications associated with manual manipulations are
then conducted. For example, when a load point supply can be restored by
sectionalizing equipment, the down-state duration of that load point caused
by a particular line section failure should be adjusted as
Td=min{T., T r } (6.4)
where Td is the down-state duration after the modification, Ts is the sampled
value of the down-state duration before the modification, and Tr is the
manual sectionalizing time.
220 CHAPTER 6

3. The average values of the three basic load point indices caused by
each single line section (subscript z) operating history can be calculated by
N
A·=- (6.5)
I LTu

LTd
r·=--
I N (6.6)

LTd LTd
Ui ~--=A.;ri (6.7)
LTu+LTd LTu
where L Tu and L Td are summations of all up times and all down times
during the total sample years, respectively; N is the number of transitions
between the up and down states during the total sample years. The calculated
values from these equations correspond to those in one row associated with
a particular line section in Table 6.4. The total load point indices correspond-
ing to all line sections can be obtained using the same calculation procedure
shown in Table 6.4, in terms of equations (6.1) to (6.3). The analytical
approach is obviously much faster than the Monte Carlo method if only
mean values of the reliability indices are required. The purpose behind utiliz-
ing the Monte Carlo method is to calculate index probability distributions
and possibly incorporate more complex configurations or operating logic.
4. The calculations associated with the probability distributions are
similar. The average values of the three basic indices in a single year due to
each single line section can be obtained by equations similar to (6.5)-(6.7).
The difference is that 'L Tu , 'L Td , and N correspond to each year instead
of the entire sampling span. Thus 'L Tu , 'L Td , and N should be expressed
as ('L TU)j, ('L Td)j and Nj where sUbscriptj denotes year j. Corresponding
Ai, ri, and Ui also should be expressed as Aij, ri]' and Uij (subscript ij indicates
line section i and year j). Equations (6.1) to (6.3) can be used to obtain the
total load point indices caused by all line sections in each sample year. The
probability distribution of a load point outage rate can be calculated using

p(k)=m(k) (k=O, 1,2, ... ) (6.8)

M
where M is the number of years in a whole sampling span and m(k) is the
number of years in which the load point outage rate equals k. For example,
when k = 1, m( 1) is the number of years in which the number of load point
failures is 1. Similar calculations can be conducted to obtain the probability
distributions ofload point outage duration (r) and load point unavailability
(U). Because r and U are continuous random variables, discrete class inter-
vals must be selected.
DISTRIBUTION SYSTEM AND STATION ADEQUACY ASSESSMENT 221

Time
Figure 6.2. A simulated system operating history.

(b) System Operating Cycle Technique. The individual compo-

nent operating cycle technique utilizes equations (6.1) to (6.3), which are
based on the principle of series systems and therefore only apply to simple
radial distribution systems. The system operating cycle technique can be
applied in more complex situations, such as combined configurations of
distribution systems and substation arrangements, or a distribution system
with parallel circuits or meshed loops. It includes the following steps:
1. Artificial up-down operating histories of all line sections are
obtained by the component state duration sampling.
2. A system operating history shown in Figure 6.2 can be obtained by
combining up-down operating histories of all line sections. In this process,
manual manipulations such as sectionalizing actions also should be incorpor-
ated. Each step shown in Figure 6.2 corresponds to a system operating
state, which denotes opening due to component failure(s) or closing due to
component repairs or manual switching.
3. Each system state is evaluated to obtain load point up-down operat-
ing histories. This is usually associated with checking connectivity from the
source point(s) to each load point.
4. The mean values and probability distributions of the three basic load
point reliability indices can be calculated using equations similar to (6.5)-
(6.8). The load point up-down operating histories are obtained from evaluat-
ing whole system states but not from each single line section operating
history. Consequently, utilization of equations (6.1) to (6.3) is no longer
required.

6.3.3. Load Point Index Distributions

A Monte Carlo based computer program has been developed at the
University of Saskatchewan to simulate the performance of any N-section
222 CHAPTER 6

Table 6.10. Load Point Indices Using

the Two Methods (Case 1)
Load point
Index A B C
l (failures/yr)
Analytical 1.35 1.10 0.85
Simulation 1.35 1.10 0.86
r (hr/failure)
Analytical 1.15 1.86 2.41
Simulation 1.17 1.87 2.42
U(hr/yr)
Analytical 1.55 2.05 2.05
Simulation 1.57 2.07 2.08

radial distribution system with loads connected to laterals or directly to the

primary mains.('4.'7) Any combination of exponential, normal, log-normal,
and gamma distributions can be used to simulate the failure, repair, manual
sectionalizing, alternate supply, and fuse times. Studies were performed using
the system shown in Figure 6.1. The mean values in Case I (base case)
obtained from the Monte Carlo simulation for a period of 5000 years and
those obtained from the analytical approach are summarized in Table 6.10.
In this simulation, all the failure and repair or restoration times are assumed
to be exponentially distributed. It can be seen that the results using the
Monte Carlo method and the analytical approach are basically the same. It
should be noted that the mean load point indices are unaffected by restora-
tion time distributions.

(a) Distributions of Load Point Failure Rate. The components

were assumed to be in their useful operating life and therefore the failure
times were exponentially distributed. The probability distribution ofthe load
point failure rate can be calculated using the Poisson distribution:

P(n) = (Mr e-}J (6.9)

n!
where n is the number of failures in time t.
The results of Monte Carlo simulation match those obtained by theor-
etical analysis. Figure 6.3 presents the distributions associated with the fail-
ure rates of load points A, B, and C in Cases 1, 2, and 3 for the system
shown in Figure 6.1. The distributions for Case I are identical to those for
Case 2, due to the fact that backfeeding does not alter the probability of
failure. In both cases, the distributions are noticeably different for the three
 C
0.4, 04 0.4 en
-t
Cases 1 &2 Case 3
Load pOint A
Load pomtA
Cases 1 &2
:!:!
Average = 1 3450 f/yr
Average = 2 0910 f/yr IX!
0.3 Load pOint B
0.3 0.3 Average = 1 1026 flyr
C
-t
~ ~ ~
is :c :cC1l 0
2 0.2 2 0 .2 o Simulated result ~ 0.2
z
e e • Analytical result c..
en
c.. c.. -<
o Simulated result en
0.1 I;; ~~~~~:~~~ :=:~I~t I 0.1 0.1 • Analvtlcal result -t
n nnn,.. n m
3:
00"
nII-I
II"
0
II"
1
II"
2 3
II"
4
II"
5""
0.0 II II"
0
II"
1 2
II" II"
3
II"
4
II"
5
0.0 11 ~
0
~
1
II"
2 3
II"
4
II"
5
II" »
Z
Failures (occ/yr) Failures (occ/yr) C
Failures (occ/yr)
en
-t
»
-t
0
Z
0.4, 0.5, 05, »
C
m
.... a5e J 0
0.4 Cases 1 &2 04 Load pOint C
0.3 Load point C
C
Average = 2 0910 flyr »
Average = 0 8594 flyr
~ ~ 0.3 ~ 03 0
:c :cC1l is -<
2 0 .2 .0 C1l
.0 »
e e 02 0 0.2 o Simulated result en
c.. c.. 0: • Analytical result en
o Simulated result m
0.1 en
• Analytical result
n 0.1 01 II ,...... en
3:
0.0 II - -
0
nnn-- 2
II"
3
II"
4"" ""
5
0.0 II

0
II"
nn
1
II"
2
II"
3
- II"
4
=-1'"
5
0.0 I

0
-
nnn
II"
1 2
II"
3
II" II"
4 5
- m
Z
-t
Failures (occ/yr) Failures (occ/yr) Failures (occ/yr)

Figure 6.3. Load point failure rate distributions.

N
N
W
224 CHAPTER 6

load points. At load point A, years with one failure occur most frequently
while, at load point C, years with no failures occur most frequently. The
distributions for Case 3 vary from those for Cases 1 and 2, but are identical
for each of the three load points. This is due to the fact that each load point
experiences a failure when anyone of the solidly connected laterals suffers
a fault.

(b) Distributions of Load Point Outage Duration. If the repair

and other restoration times of the components can be assumed to be expo-
nentially distributed, the load point outage duration can be approximated
by a gamma distribution.(21) This is confirmed by the results of Monte Carlo
simulation. Figure 6.4 shows the results for the outage durations of load
points A, B, and C for Cases 1, 2, and 3 when the restoration times are
exponentially distributed. These distributions can be reasonably described
by the gamma distribution (chi-square test at a significance level of 0.05).
As can be seen, the general shape of the distributions does not vary.
Although the gamma distribution can take on many different shapes, the
shape is always of this general form when it is the result of combinations
of exponential distributions. The gamma distribution becomes more or less
spread out, or more peaked, depending on the component average restora-
tion durations.
In many practical systems, the restoration times are not exponentially
distributed. Restoration times may be better described by nonexponential
distributions, such as log-normal repair times. The studies carried out indi-
cate that when the restoration times are assumed to be nonexponelltial, the
load point outage duration cannot generally be represented by a gamma
distribution.
It should be emphasized that the average values of the load point outage
duration indices are not affected by underlying distributions. A set of aver-
ages such as those calculated for the example system can have any set of
distributions associated with it.
The durations associated with repairs and other restoration activities
may often be well described by the log-normal distribution. Figure 6.5 shows
the simulated outage duration results, when restoration times are assumed
to be log-normally distributed with a standard deviation equal to one-third
of the mean. These load point outage duration distributions have a radically
different shape than those assuming exponential restoration times (Figure
6.4). In Case 1, the probability of outage duration for load point A appears
to be decreasing as the outage duration increases, except for a peak of two
bars wide. This peak is attributable to the large number of repairs of I-hr
average duration which are made on the first primary lateral. Due to it being
further down the line, load point C has a larger number of 3-hr repair
 0.35 0.35 035
0.30 0.30 C
030 Case 1
Case 2 Case 3 (jj
Load pOint A
0.25
I Average = 1 1709 hr
>-
0.25 Load pOint A
Average = 1 1707 hi
0.25 Load point A
Average = 0 9411 hr
-I
:xl
~ 0.20 ~ 0.20 ~
D .c 0.20 I:a
D 015 .g'" 0.15 D C
e'" 0.15 -I
a. a: e'"
a.
0.10 0.10 0.10 0
Z
0.05 0.05 0.05 C/)
-<
C/)
0.00 0.00 0.00
0 2 3 4 5 0 2 3 4 5 0 2 3 4 5 -I
m
Outage Duration (hr) Outage Duration (hr) Outage Duration (hr)
0.30 0.30
s:
0.30 »
Z
0.25 0.25 0.25 C
Case 1 Case 2 Case 3
Load pOint 8 C/)
Load pomt B Load pOint B
~ 0.20 Average = 1 8739 hr ~ 0.20 Average = 1 5363 hr ~ 0.20 Average = 1 4416 hI -I
:0
»
-I
~ 0.15 2l 0.15 ~ 0.15
D D 0
e e e
a. 0.10 z
a. 010 a. 0.10
»
C
0.05 0.05 005 m
-,-- ·l'--~- -'~- ,------.,--------r--~r_ 0
0.00 0.00 0.00 C
0 2 3 4 5 0 2 3 4 5 0 2 3 4 5 »
Outage Duration (hr) Outage Duration (hr) Outage Duration (hr) 0
0.30
-<
030 0.30 »
C/)
0.25 C/)
0.25 025 m
Case 1 Case 2 Case 3 C/)
Load pOint C Load pomte Load point C C/)
~ 0.20 Average = 2 4151 hr ~ 0.20 Average = 1 2597 hr Average = 1 6350 hr
.c ~ 0.20 ~
:0 s:m
2l 0.15 2l 0.15 0.15
e i n Z
e a. . 11 -I
a. 0.10 a. 010 0.10
005 0.05 0.05
0.00 II II [[ II II II II II II II [I II
0.00 0.00 , IQCJc::Jc:::::l N
0 2 3 4 5 0 2 3 4 5 0 2 3 4 N
5 U1
Outage Duration (hr) Outage Duration (hr) Outage Duration (hr)

Figure 6.4. Load point outage duration distributions for exponentially distributed repair times.
 0.30 0.30 0.30
0.25 0.25
Case 1 0.25
load pomt A Case 2
Load pOInt A Case 3 ~
~ 0.20 Average = 1 1668 hr ~ 0.20 Load pOint A 01
Average = 1.1560 hr ~ 0.20 Average = 0 9252 hr
~ 0.15 ~ 0.15 i5
CO 0.15
.0 .0
~ e
Cl. 0.10 £ 0.10 Cl.
0.10
0.05 0.05 0.05
0.00 I I I I I I IOooononCJ== 0.00 IOOOOnonC0= =
o 1 2 3 4 5 0.00 J II II II II II IDOo p n == =
I I I I1I I
o 2 3 4 5 o ~ 3 4 5
Outage Duration (hr) Outage Duration (hr) Outage Duration (hr)
0.30 0.30 0.30
0.25
Case 1
0.25 0.25
Case 2 Case 3
Load pomt B
~ 0.20 Load pomt B Load point B
Average = 1 8772 hr ~ 0.20 Average = 1 5084 hr ~ 0.20 Average = 1 4010 hr
:0 i5
.2l 0.15 .2l 0.15
e ~ 0.15
e .0
Cl. 0.10 Cl. e
0.10 Cl. 0.10
0.05
0.05 0.05
0.00 II I I I I I I I I I 100000=
234 5 o 2 3 4 5 2 3 4 5
Outage Duration (hr)
Outage Duration (hr) Outage Duration (hr)
0.30 0.30 0.30
0.25 0.25
Case 1 0.25
Case 2 Case 3
Load point C Load point C
~ 0.20 Average = 2 4096 hr 0.20 Load point C
Average = 1 2455 hr ~ 0.20 Average = 1 5825 hr
i5 ~
.2l 0.15 B 0.15 ~ 0.15
e co .0
Cl. 0.10 ~
Cl. 0.10 £ 0.10 C')
0.05 J:
0.05 0.05 »-a
0.00' II II II II II II II II II II II lor I~,------ --I
234 5 o 1 2 3 4 m
Outage Duration (hr)
2 3 4 5 :II
Outage Duration (hr) Outage DuratIon (hr) 01

Figure 6.5. Load point outage duration distributions for log-normally distributed repair times (SD = 1/3 mean).
DISTRIBUTION SYSTEM AND STATION ADEQUACY ASSESSMENT 227

average durations. The repairs to the mains in combination with repairs to

the third lateral result in a bimodal distribution with two peaks at durations
about 1 and 3 hr long. Load point B has a distribution that is a combination
of those at A and C.
The backfeeding in Case 2 reduces the number of failures with three-
hour repairs at load point C. The second mode in the distribution is elimin-
ated. The mode resulting from the lateral repairs is now more pronounced
for both load points Band C. When the laterals are connected solidly to
the mains as in Case 3, the first mode is even more pronounced for all three
load points. The predominant cause of outages are failures on the primary
laterals resulting in I-hr average repairs.
A visual inspection of the distributions in Figure 6.5 indicates that the
distributions are quite different from those in Figure 6.4 and that attempts
to predict the duration probabilities using the gamma distribution in the
cases of nonexponential restoration times could lead to large errors. Similar
results have been obtained from simulations in which restoration activities
are assumed to be gamma and normally distributed. The study results indi-
cate that no known distribution can universally describe the outage duration
distributions.
The distributions vary with the means and the standard deviations of
the component repair times. Figure 6.6 shows the outage duration distribu-
tions for load points A and C for the system in Case 1 with the assumptions
that the standard deviations for the component repair times are equal to m
(MTTR), m12, and m16. When the standard deviations are relatively large,
the contributing distributions overlap, with the result that the outage dura-
tion distribution is almost a monotonically decreasing function. As the stan-
dard deviations decrease, the contributing distributions become apparent
and the distribution is definitely multimodal. The sizes and the number of
primary main or lateral sections affects the mean, the standard deviation,
and possibly the type of the component repair time distributions. The load
point outage duration distributions are then indirectly affected by these
factors.

(c) Distributions of Load Point Annual Interruption

Time. The distributions of load point annual interruption time are depen-
dent on both the failure rate and outage duration distributions. This makes
it even more difficult to describe the interruption time distributions by known
functions. Figure 6.7 shows these distributions for load points A and C for
the system in Case 1.
The simulations that assume exponentially distributed component
repair times result in load point annual interruption time distributions that
contain a sharp peak for the interval indicating the number of years with
 N
N
co
0.35 0.35 0.35
0.30 030 0.30
0.25 0.25 0.25
~ ~
~0.20 B 0.20 B 0.20
.is co co
~ 0.15 .g 0.15 ~ 0.15
e a: a.
a. 0.10 0.10 010
0.05 005 0.05
0.00 I[JoOnClDc;ncn 000 Ii "Cle ... 0.00
o
I I I I.I I I
3 4 5 0 2 3 4 5 0 2 3 4 5
Outage Duration (hr)
~bDD ~
Outage Duration (hr) Outage Duration (hr)

0.30 0.30 0.30

0.25 0.25 0.25

0.20 ~ 020 ~0.20

~
~ 0.15 .g 0.15 .g.0 0.15
.0 .0
e e
£. 0.10 a. 010 a. 0.10

005 0.05
::: n III1 I I I IDOOO[JD]C]OD o 00 I II II bCli II II II II I'
o 2 3 4 5 2 3 4 5 o 2 3 4 5
Outage Duration (hr) Outage Duration (hr) Outage Duration (hr)
o
Figure 6.6. Load point outage duration distributions as a function of component repair time standard deviations for log-normally J:
:to
distributed repair times for Case 1.
~
m
:lI:I
01
 0
Ui
-I
::II
0.30 030 0.30
CD
0.25 C
0.25 0.25 -I
~0.20 ,.. 0.20 0.20
0
:E
Z
.~ (I)
~ 0.15 j 0.15 <
(I)
:g.c 0.15
e e -I
c.. 0.10 c.. 0.10 £. 0.10 m
0.05 0.05
3:
0.05
l>
0.00 Z
0 4 5 6 7 8 2 3 4 5 6 7 6 7 8
0
8 2 (I)
Outage Duration (hr) Outage Duration (hr) Outage Duration (hr)
~
::!
0
Z
l>
0.60 0.60 060 0
m
0.50 0.50 0.50 0
C
~ 0.40
l>
~ 0.40 ~0.40 (")
13 :0 <
.:g 0.30 .:g 0.30 :g 0.30 l>
e e .c (I)
c.. 0.20 c.. 0.20
e (I)
c.. 0.20 m
(I)
0.10 0.10 0.10 (I)
3:
0.00 000 m
0 2 3 4 5 6 7 8 2 3 4 5 6 7 8 0 2 3 4 5 6 7 8
Z
-I
Outage Duration (hr) Outage Duration (hr) Outage Duration (hr)

Figure 6.7. Load point annual interruption time distributions for Case I.
N
N
CQ
230 CHAPTER 6

no failures. The distributions are steadily decreasing with long tails. The
distributions are not of exponential or gamma form. The simulations that
assume log-normal distributions for the component repair times and expo-
nential distributions for the other times also result in load point annual
interruption time distributions with a sharp peak for the no-failure interval.
The log-normal simulations do not result in steadily decreasing distributions
but in distributions with multiple modes. The multiple modality is more
prominent when the repair times are assumed to have small standard devia-
tions. It is interesting to note that in all the three cases in Figure 6.7, the
number of years with zero hours of interruption is independent of the form
and the standard deviation of the component repair time distributions. This
independence occurs because the number of years with zero interruption
time is dominated by the failure rate distributions, which determine the
number of years in which no failures occur.

6.3.4. System Performance Index

Distributions
Section 6.2.3 notes that the system performance indices can be calcula-
ted directly from the basic load point indices. System performance indices
are in fact weighted averages of the load point indices. The distributions of
system performance indices therefore can also be obtained from those of
the load point indices by similar calculations. Figure 6.8 shows these distribu-
tions resulting from the simulations of the system in Case I with the assump-
tion that the component repair times are exponentially and log-normally
distributed.
The SAIDI distribution is dependent on the repair time distributions.
The number of customers at each load point and the average failure rates
are weighting factors that are independent of the associated distributions.
The SAIDI distributions in Figure 6.8 are more or less similar to the annual
interruption time distributions in Figure 6.7. This is because SAIDI is a
linear combination of annual interruption times. Additional studies indicate
that in a large system, the resemblance tends to decrease because of the
averaging effect of the large number of load points which are aggregated.
In this small system, the number of years with a SAIDI equal to zero is
relatively high (i.e., P[SAIDI = 0] is high). This might be expected in a small
or moderate-size system. In the case of a large system or region it is very
likely that there would be at least one interruption.
The SAIFI distributions are identical for the exponential and log-
normal (SD=m and SD=mj6) component repair times because SAIFI is
only dependent on the component failure time distributions and on the
 0.25 0.25 0.25
Exponential Repair Exponential ~epalr Exponential Repair
0.20 50.(') =m 0.20 So (,) =m 0.20 SO (,) =m
Ave,age = 1 224 Average = 1 332
~~'(r,.IOI) ~ ~ ~~ 5 0 (SAIFI) = 0 974 SO (CAlOI) = 1 199
~ 0.15 ~ 0.15 ~ 0.15
~ ~ ~
~ 0.10 .g 0.10 .g 0.10
!l. Ii: Ii:
0.05 0.05 0.05

0.00 1,Jl.u..JLJLJL.\LJLI!...JL.II.JILl-lL:::':.II-.!L-I!-J"-::~~ 0.00 -1-'.JI,1I-.J/......jJ....L.JJL...JI.--II-.lI--JL..JJ,-.J/......!I-JL...JI--.JI-...Il.....J 0.001 II II I II

0.0 1.0 2.0 3.0 O.D 1.0 2.0 3.0 0.0 1.0 2.0 3.0
SAlOl (hours' customer) SAIFI (interruption' customer) CAIOI (hours' interruption)
025 0.25 0.25
Lognormal Repair Lognormal Repa"
0.20 SO (,) =m 0.20 0.20 5.0 (,) =m
Ave,age = 1 761 Average = 1 346
5 0 (SAlOl) = 2 292 5 0 (CAIOI) = 1 257
~ 0.15 .~ 0.15 .~ 0.15
:c
~ ~ ~
~ 010 .g 0.10 .g 0.10
!l. Ii:
Ii:
005 0.05 0.05

0.00 ~I...JIJ....I.JL.!L-II.....IL..JL....JL...JL...lJ-Jl-I!-JL..JL....JL...JL...lL...l
00 1.0 2.0 3.0 1.0 20 3.0 1.0 2.0 3.0
SAlOl (hours' customer) SAIFI (Interruption' customer) CAIDI (hours 'Interruption)
0.25 0.25 1 0.25
Lognormal Repa"
Lognormal Repaw
5.0 (,) = ml6
Lognormal Repa"
0.20 50.(,) =m/6 0.20 SO (,) =m/6 0.20 Average = 1 333
Average = 1 757 Ave,age = 1 224 5 0 (CAIOI) = 0 589
5 0 (SAlOl) = 1 729 SO (SAIFI) = 0 974
.~ 0.15 ~0.15 ~ 0.15
~ :s
<II
ii
~ 0.10 .g 0.10 ~ 0.10
!l. !l.
Ii:
0.05 0.05 0.05

0.00:-;';--1....
1 .........lL-u..:';;-ll-ll...JLJL.\LJLJUJ-lJJ 0.00 ~,....JI....I ....JO-...............JI--"--"--"--"-.!I-.!I-I!-JI!-J'-....J.............. 0.00' = 0'
0.0 1.0 2.0 3.0 0.0 1.0 2.0 3.0 0.0 0.5 1.0 1.5
SAlOl (hours' customer) SAIFI (interruption' customer) CAIDI (hours' interruption)

Figure 6.S. System performance index distributions for Case 1.

232 CHAPTER 6

number of customers served at each load point. As in the case of SAIDI

distributions, the probability of SAIFI equaling zero diminishes in large
systems. The distribution becomes less like an exponential distribution and
more like one with a mode about the average.
The CAIDI distributions are nonlinearly related to both the failure and
repair times. This and other comparisons indicate that the standard devia-
tions of the underlying component distributions can affect the shape of the
CAIDI index distribution as much as or more than the form chosen for the
underlying distributions. For large systems the dispersion of the CAIDI
distribution also tends to decrease.
It is important to appreciate that the system performance indices of
SAIDI, SAIFI, and CAIDI are associated with probability distributions.
They can be expressed by their average or mean values but they are random
variables. It should be expected therefore that SAIDI, SAIFI, and CAIDI
statistics obtained from actual circuit or system operating behavior will vary
over time as each year is simply a snapshot of a continuum in time. The
ability to generate the index distribution by Monte Carlo simulation provides
the opportunity to appreciate and quantify the dispersion associated with
these important customer parameters.

6.4. STATION RELIABILITY ASSESSMENT

6.4.1. General Concepts
Substations and switching stations serve important functions in an elec-
tric power system. These facilities are energy transfer points between generat-
ing stations, transmission, subtransmission, and distribution systems. The
structure of reliability assessment of substation and switching stations is
basically similar and therefore the term station is used to describe both of
them in the following sections. Stations, in a general sense, provide the
connection between generating units and transmission systems (voltage step-
up substations), and between transmission or distribution systems and con-
sumer facilities (voltage step-down substations). Stations are functionally
designed with considerable flexibility in the arrangement of their basic com-
ponents such as circuit breakers, bus sections, and transformers. Factors
considered in selecting a specific station configuration include service security
and reliability, operating flexibility and simplicity, short-circuit current limit-
ations, protective relaying, equipment maintenance, future extensions and
modifications, standardization and cost, etc.
DISTRIBUTION SYSTEM AND STATION ADEQUACY ASSESSMENT 233

From a reliability evaluation point of view, a station divides "a point"

in a bulk power network into a number of dispersed points. Some are loca-
tions where energy enters and therefore can be termed "source points."
Some are locations where energy departs and can be termed "load points."
The basic principle of station reliability assessment, as in radial distribution
system assessment, is to evaluate connectivity from the source points to
the load points. Station reliability evaluation is more complex than radial
distribution system reliability evaluation, as it is associated with multiple
source points, multiple breakers, and complicated switching functions.
Station reliability indices can serve two basic functions: (1) They should
be in a form which can be incorporated in a composite system adequacy
assessment. A complex station arrangement can be represented by a single
bus with a detailed reliability model. Thus the effect of station configurations
can be included in composite system evaluation. (2) Station indices should
also be in a form which can be used in radial distribution system reliability
assessment. A source point in a distribution system can be represented by
a reliability model which permits station effects to be included in distribution
system evaluation. Station reliability indices can be considered in two groups.
One group contains outage probabilities and frequencies for each set of
external connections. These outage sets are mutually exclusive, i.e., the
remaining connections must be in service if one set of the connections is in
the outage state. This group of station reliability indices can be used as input
data in composite system reliability assessment. The second group contains
the outage probability, outage rate, and outage duration for every point.
These outage events are not mutually exclusive, i.e., the remaining points
can be in service or on outage when a point is in an outage state. This group
of indices contains the necessary reliability data to evaluate a particular
point and therefore can be used as input data in radial distribution system
reliability assessment. In order to discriminate between these two groups of
indices, the terms "connection set" and "point" are utilized.

6.4.2. Station Component Modeling

Station components are usually represented by Markov models in ana-
lytical station reliability evaluation. (8-12) This implies that the transition rates
between different states are constant and that the duration in each state is
exponentially distributed. The Markov model assumption may be invalid
for many repairable components. As discussed in Section 6.3, component
repair activities are generally not exponentially distributed. Log-normal(l5)
or Weibull distributions(24) have been suggested to represent repair or restor-
ation times. The component representation will be non-Markovian when
234 CHAPTER 6

Repair

U R

Failure

Figure 6.9. Two-state component model.

one or more of the transition rates are time-dependent. The model cannot
be easily solved using analytical methods in the case of a non-Markovian
process.(I)
A station contains three basic component typeS(2,12): bus sections, trans-
formers (or lines/cables), and breakers. Bus sections can be represented by
a two-state model as shown in Figure 6.9. Transformers (or line/cables) can
be represented by a three-state model as shown in Figure 6.10. This generally
indicates that after the fault, a transformer (or line/cable) is first isolated
by switching action and then repaired. A breaker can be represented by the
more complex three-state model shown in Figure 6.11. There are two failure
modes in this model, designated as passive and active failures, respectively.(2)
These are defined as:
1. Passive event: A component failure mode that does not cause opera-
tion of protection devices and therefore does not have an impact on
the remaining healthy components. Service is restored by repairing

Repair
U R

1 3

Failure

Switching
S

Figure 6.10. Three-state component model.

DISTRIBUTION SYSTEM AND STATION ADEQUACY ASSESSMENT 235

Repair

U R
Passive failure
1 3

Active
failure

Switching
S

Figure 6.11. State space diagram for active and passive failures.

or replacing the failed component. An open-circuit failure is an

example of a passive event.
2. Active event: A component failure mode that causes the operation
of the primary protection zone around the failed component and
can therefore cause the removal of other healthy components and
branches from service. The actively failed component is isolated and
the protective breakers are reclosed, which leads to service being
restored to some or all of the load points. It should be noted, how-
ever, that the failed component itself can be restored to service only
after repair or replacement.
It can be seen from the above general definitions that the passive and
active events can also apply to nonbreaker-type components if it is necessary
to model them using the more detailed representation. The four-state model
shown in Figure 6.12 can be used when maintenance activity is considered.
Two maintenance considerations should be recognized. One is to assume
that both the time to the maintenance state and the time back to the up-
state follow a given distribution. The parameter(s) of the distribution can
be obtained from historical records of maintenance activities. This approach
can be used to assess the general effects of maintenance on station reliability.
The second consideration is to recognize maintenance as a series of scheduled
events which occur at prespecified intervals. In this case, the scheduled out-
age is not a random event. Both considerations can be easily simulated using
the component state duration sampling method.
236 CHAPTER 6

Out

Repair Maintenance
outage

Repair
U R
Passive failure
1 3
Active
failure

Switching
S

Figure 6.12. Four-state component model.

6.4.3. Component State Duration Sampling

Procedure
As noted earlier, the component state duration sampling technique can
be used to simulate any probability distribution. In the following analysis,
for generality, repair activities are assumed to be Weibull distributed and
other activities are assumed to be exponentially distributed. The cumulative
probability distribution functions associated with failure, switching, and
repair times are given by equations (6.10) to (6.13). If maintenance activity
is expressed by a distribution, it can be included in a similar way:
Fa(t) = l-e-.1.a1 (6.10)
Fp(t) = 1 - e-ApI (6.11)
s(t) = 1 - e-/Jswl (6.12)
R(t) = 1- e-(lja)fI (6.13)
DISTRIBUTION SYSTEM AND STATION ADEQUACY ASSESSMENT 237

where A.. is the active failure rate, Ap the passive failure rate, J.lsw the switching
rate, and a and p are the scale and shape parameters, respectively, of the
Weibull distribution. The component state duration sampling procedure
involves the following four basic steps:

(a) Artificial Component Operating History. The operating

history of each component is created by random sampling. The sampling
values of operating time, switching time, and repair time are designated as
time to active failure (TTFa), time to passive failure (TTFp), time to switch
(TTS), and time to repair (TTR), respectively. TTFa , TTFp, TTS, and TTR
can be obtained using the inverse transform method given in Section 3.4.2,
i.e.,

1
A.. Uf a
TTFa =--In (6.14)

1
Ap Unp
TTFp=--In (6.15)

1
TTS=--ln Us (6.16)
J.lsw

TTR=a(-ln Ur)I/P (6.17)

where Ufa , Ufp , Us, and Ur are four independent, uniformly distributed
random numbers between [0, 1]. If the component resides in the switching
state, the sample switching time is given by equation (6.16). If the component
resides in the repair state, the sample repair time is given by equation (6.17).
If the component resides in the normal operating state, determination of the
sample operating time is dependent on the type of the component. If the
component is represented by a model without passive failure, such as a bus
or a transformer, the sample operating time TTF is given by equation (6.14).
If a model with passive failure is used, such as in the case of a breaker, the
sample TTF is obtained by comparing TTFa and TTFp given in equations
(6.14) and (6.15), respectively, i.e.,

TTF=min{TTFa, TTFp} (6.18)

The sample time value of the component residing in an up state is TTF,

i.e.,

Tup=TTF (6.19)
238 CHAPTER 6

The sample time value of the component residing in a down state is

dependent on how it transits from the up state. If the repair state is reached
by a passive failure, the time value is TIR, i.e.,
Tdn=TIR (6.20)
If the repair state is reached by an active failure, the time value is given by
(6.21)

(b) Modified Component Operating History. The artificial

component operating history obtained using the described method is a
random event sequence. This operating history can be modified to
include specified deterministic station operating factors. Two aspects are
involved:
1. Station operation is associated with both protection and switching
activities. When a component fails, protective breakers may open
and the effected breakers can be reclosed when a failed component
is switched out. Switching logic is prespecified in the station design
and operating guidelines. The effects of the protection and switching
activities can be incorporated by modifying the artificial operating
history.
2. A scheduled maintenance activity cannot be modeled by a probabil-
ity distribution as it is a prespecified event. The down time due to
the scheduled maintenance can be easily imposed on the artificial
component operating history.

(c) Connection Set or Point Operating History. As noted in

Section 6.4.1, there are two groups of station reliability indices. One group
contains the mutually exclusive connection set indices. The other contains
nonmutuallyexclusive (source or load) point indices. The complete station
operating history can be obtained by combining the individual modified
component operating histories. Connection set or point operating histories
can be obtained by examining the connectivity from source point(s) to load
point(s) for each station system state in the complete station operating
history.

(d) Calculation of Reliability Indices and Their

Distributions. The required reliability indices can be calculated from con-
nection set or point operating histories. The availability (A), unavailability
(U), average failure rate (A), repair time (r), and the outage frequency (I)
associated with a connection set or a point can be obtained using equations
DISTRIBUTION SYSTEM AND STATION ADEQUACY ASSESSMENT 239

(6.22) to (6.26):

(6.22)

(6.23)

N
A (6.24)
I:l Tupi
I:l Tdni
r (6.25)
N

(6.26)

where N is the number of outages associated with a connection set or a

point in the simulation span, Tupi is the ith operating duration associated
with a connection set or a point in the simulation span, and Tdni is the ith
outage duration associated with a connection set or a point in the simulation
span.
The state duration sampling approach can be used to provide informa-
tion on reliability index distributions. The probability that a specified num-
ber of outages will occur within one time unit (normally one year) is equal
to the corresponding subtotal time units divided by the total simulation time
units, i.e.,
N
Pr(i)= n ' (i=0, 1, . .. , n) (6.27)
Ij=o~
where Ni is the number of time units in which only i outages occur within
one time unit, n the maximum outages within one time unit, Pr(i) is the
probability that i outages occur within one time unit.
Similarly, the outage duration probability distribution associated with
a connection set or a point can be obtained using the following equation:
N
Pd(i)= m ' (i=0, 1, ... , m) (6.28)
Ij=oNj
where Ni is the number of outages in which outage duration length is between
Ti and Ti+ 1 , m the number of segments into which outages are divided in
terms of their duration length, while Pd(i) is the probability of outages
whose durations are between T, and Ti + 1.
240 CHAPTER 6

L
_-!..4_---1~ ~L
1 2

(b)

1
CD
5
2
r ~ L1 1
CD
5
2
r
x9
.L1

4
CD 18 ~ L2 4
CD 18 .L 2

(c) (d)

Figure 6.13. Five simple station configurations.

6.4.4. Numerical Example 1

Figure 6.13 shows five simple station configurations. (2,25) The station
component data are given in Table 6.11. A normally opened breaker was
assumed to have the same data as a normally closed breaker.
Case (a). Base case [Figure 6.13a].
Case (b). As in Case (a) but without the breakers [Figure 6.13b].
Case (c). The bus section is split and the load divided equally between
the bus sections [Figure 6. 13c].
Case (d). As in Case (c) but with a normally closed bus section breaker
between the two busbars [Figure 6.13d].
Case (e). As in Case (d) but with a normally opened bus section
breaker [Figure 6.13e].
All failure, switching, repair, and scheduled maintenance activities were
modeled using exponential distributions. The five station configurations were
analyzed using the Monte Carlo simulation approach(25.26) and the analytical
method.(2) The results are presented in Tables 6.12 to 6.14. It can be seen
DISTRIBUTION SYSTEM AND STATION ADEQUACY ASSESSMENT 241

Table 6.11. Component Reliability Data for the

Simple Station ConfigurationsB
A.. 4 r Tsw A..n rm
Component (f/yr) (f/yr) (hr/f) (hr/f) (f/yr) (hr/f)
Breaker 0.01 0.01 3.00 1.0 0.1 5.0
Bus section 0.024 2.00
Transformer 0.10 50.00 1.0 0.2 10.0
Line 0.09 7.33 1.0
DA... active failure rate; 4. passive failure rate; r. repair time; Tow.
switching time; A..n. transition rate to maintenance state; rm. mainten-
ance time.

Table 6.12. Load Point Outage Rates

Analytical Monte Carlo Relative
Case (failures/yr) (failures/yr) error (%)
(a) 0.0449686 0.0462504 2.85
(b) 0.4046820 0.4062153 0.38
(c) 0.4140000 0.4152556 0.30
(d) 0.2248393 0.2279579 1.39
(e) 0.2240000 0.2245554 0.25

Table 6.13. Load Point Unavailability

Analytical Monte Carlo Relative
Case (hr/yr) (hr/yr) error (%)

(a) 0.0760151 0.0773285 1.73

(b) 0.4346979 0.4342171 O.ll
(c) 7.8977000 7.9237038 0.33
(d) 0.2553120 0.2584898 1.24
(e) 0.2480000 0.2500889 0.84

Table 6.14. Load Point Outage Durations

Analytical Monte Carlo Relative
Case (hr/failure) (hr/failure) error (%)
(a) 1.6904025 1.6719678 1.09
(b) 1.0741716 1.0689864 0.48
(c) 19.0765700 19.0987848 0.12
(d) 1.1355310 1.1339697 0.14
(e) 1.1071429 1.1137387 0.60
242 CHAPTER 6

that the two methods provide basically the same load point reliability indices.
The effect of nonexponential state residence distributions is illustrated
by assuming that the repair times are Weibull distributed. All failure, switch-
ing, and scheduled maintenance activities are modeled using exponential
distributions. The mean values of the repair times were held constant in
order to compare the results. The shape parameter p was set at 0.5, 1.0,2.0,
and 4.0, respectively. The Weibull distribution is an exponential distribution
when the shape parameter is equal to 1.0. Load point reliability indices are
shown in Tables 6.15 to 6.17. It can be seen from these results that the mean
reliability indices are not influenced by the change in shape parameter p
of the Weibull distribution. The utilization of different distributions for

Table 6.15. Load Point Outage Rates Using Wei bull

Repair Time Distributions
Load point outage rates (failure/yr)
Case P=O.s P=1.0 P=2.0 P=4.0
(a) 0.0458337 0.0462504 0.0459448 0.0458337
(b) 0.4060688 0.4062153 0.4061176 0.4061663
(c) 0.4155035 0.4152556 0.4155748 0.4154948
(d) 0.2282507 0.2279579 0.2278603 0.2280067
(e) 0.2245162 0.2245554 0.2245554 0.2244770

Table 6.16. Load Point Unavailability Using Weibull

Repair Time Distributions
Load point unavailability (hr/yr)
Case p=0.5 P=1.0 P=2.0 P=4.0
(a) 0.0759160 0.0773285 0.0760292 0.0751924
(b) 0.4317163 0.4342171 0.4316121 0.4313477
(c) 8.0848246 7.9237038 7.9006220 7.9030106
(d) 0.2612150 0.2584898 0.2556588 0.2574964
(e) 0.2501599 0.2500889 0.2496769 0.2493530

Table 6.17. Load Point Average Outage Durations

Using Wei bull Repair Time Distributions
Load point outage durations (hr/failure)
Case p=0.5 P=1.0 P=2.0 P=4.0
(a) 1.6563483 1.6719678 1.6548078 1.6405622
(b) 1.0632128 1.0689864 1.0628287 1.0620500
(c) 19.4758735 19.0987848 19.0284731 19.0378941
(d) 1.1444557 1.1339697 1.1220307 1.1293701
(e) 1.1142493 1.1137387 1.1119038 1.1108492
DISTRIBUTION SYSTEM AND STATION ADEQUACY ASSESSMENT 243

Table 6.18. Load Point Outage Rate Probability

Distributions
Probability of outage rates
Case 0 2 3 4
(a) 0.954750 0.044250 0.001000 0.000000 0.000000
(b) 0.665415 0.271268 0.055805 0.006780 0.000683
(c) 0.645200 0.300160 0.049440 0.004960 0.000240
(d) 0.795750 0.182293 0.020537 0.001268 0.000195
(e) 0.798706 0.180078 0.019255 0.001882 0.000078

component repair durations does not affect the mean load point reliability
indices but only influences the distributions of these indices.
The load point outage rate probability distributions for the five station
configurations are illustrated in Table 6.18. Columns 2, 3, 4, 5, and 6. show
the probabilities that there are 0, 1, 2, 3, and 4 outages within one year,
respectively. The load point outage rate probability distribution for Case
(b) is shown in Figure 6.14 for illustrative purposes.
Table 6.19 shows the load point outage duration probability distribu-
tions. Columns 2, 3, 4, 5, and 6. show the probabilities of outage durations
between 0 to 2 hr/ failure, 2 to 4, 4 to 6, 6 to 8, and 8 to 10, respectively.
The probability of an outage duration longer than 10 hr is not shown in

1.0

0.8

Mean outage rate

~ 0.6 = 0.406215 (occ/yr)
:0
ctI
.D
e
0- 0.4

0.2

0.0
0 1 2 3 4
Outages per year

Figure 6 .14. Load point outage rate probability distribution for Case (b).
244 CHAPTER 6

Table 6.19. Load Point Outage Duration Probability Distributions

Probability of outage durations
Case (0-2) (2-4) (4-6) (6-8) (8-10)
(a) 0.711111 0.184384 0.069069 0.021622 0.009610
(b) 0.852768 0.121172 0.019935 0.003483 0.002402
(c) 0.069996 0.067104 0.047243 0.036830 0.505206
(d) 0.833298 0.131393 0.024395 0.007276 0.002996
(e) 0.840203 0.126266 0.025498 0.005414 0.001921

this table. The load point outage duration probability distribution for Case
(b) is given in Figure 6.15.

6.4.5. Numerical Example 2

Figure 6.16 shows a station configuration taken from a Canadian Elec-
trical Association (CEA) report.(27) This station is composed of eleven com-
ponents, numbered from 1 to 11, which include five breakers, three bus
sections, and three transformers. The configuration has six external connec-
tions (points) in total. This is a general configuration. As a voltage step-up
station, connections (1), (2), and (3) would be connected to transmission
lines, and (4), (5), and (6) to generating units. As a voltage step-down

1.0

Mean outage duration

= 1.068986 (hr/failure)

0-2 2-4 4-6 6-8 8-10

Outage duration (hr/failure)

Figure 6.15. Load point outage duration probability distribution for Case (b).
DISTRIBUTION SYSTEM AND STATION ADEQUACY ASSESSMENT 245

(1) (2) (3)

t t t
l~~-t~J:
~9 ~10 ~11
!
(4)
!
(5)
!
(6)
Figure 6.16. Station configuration with six connections.

station, connections (I), (2), and (3) would be connected to transmission

lines, and (4), (5), and (6) to distribution systems or direct customers. The
protection scheme associated with all of the breakers is assumed to be non-
directional. A breaker can therefore be tripped out due to faults of the
components which are directly or indirectly connected to it. For example,
Breaker I can be tripped not only by a fault of component 6 or 7, but also
possibly by faults of components 2, 4, 5,9, and 10, depending on the protec-
tion coordination between these components. The component reliability data
are given in Table 6.20. A breaker is represented using the three-state model
shown in Figure 6.11, a transformer by the three-state model shown in
Figure 6.10, and a bus section by the two-state model shown in Figure
6.9. Maintenance outages were not considered and all component state
durations including failure, repair, and switching activities were assumed to
be exponentially distributed. These are not necessary assumptions when
using the Monte Carlo approach. As seen earlier, maintenance activities and
any distribution can be incorporated with little or no extra complications.

Table 6.20. Component Reliability Data for

the Six-Connection Station
A.. A" r Tow
Component (f/yr) (f/yr) (hr/f) (hr/f)
Breaker 0.002 0.0001 126.0 1.0
Bus 0.025 13.0
Transformer 0.026 43.1 1.0
246 CHAPTER 6

Table 6.21. Connection Set Outage Probabilities

for the Station in Figure 6.16
Outage probability
Connection set Monte Carlo Analytical
(I) 0.303784E - 04 0.301867E-04
(2) 0.311544E-04 0.301876E-04
(3) 0.323512E-04 0.301867E-04
(4) 0.129359E-03 0.127842E-03
(5) 0.129863E-03 0.127846E-03
(6) O.l30515E-03 0.127842E - 03
(1)+ (4) 0.392641E-04 0.402876E - 04
(2)+(5) 0.40925IE-04 0.402828E - 04
(3)+ (6) 0.418704E-04 0.402876E - 04
(I) + (2) + (4) + (5) 0.186389E-06 0.2078I1E-06
(2) + (3) + (5) + (6) 0.237587E - 06 0.2078I1E-06

Both the state duration sampling approach and the analytical method
were used to evaluate the station configuration connection set reliability.
The outage probabilities obtained using the two methods are shown in Table
6.21. The table is not completely exhaustive as the outage probability of
high-order connection sets is very small compared with the low-order set
outage probabilities and therefore some high-order set outages are not listed.
It can be seen from Table 6.21 that the results obtained by the two methods
are basically the same.
Sensitivity analysis is an important aspect of quantitative reliability
assessment and can be used to determine how the reliability indices of each
connection set are affected by varying selected component parameters, such
as failure rates or repair times. The effects of variation in breaker failure
rates on the connection set outage probability and frequency are shown in
Tables 6.22 and 6.23. In this study, Case I is the base case and the breaker
failure rates are doubled and tripled in Case II and Case III, respectively.
Table 6.22 shows the effect on the outage probability of the connection
sets due to varying breaker failure rates (active and passive failure rates are
changed at the same percentage). It can be seen that the outage probabilities
of some connection sets are greatly affected by the breaker failure rates and
those of other sets are basically constant. Table 6.23 shows the effect on
the outage frequency of the connection sets and leads to the same general
conclusion. A general appreciation of the phenomena can be obtained by
considering the configuration. For example, outages of Connection (1) are
caused by failures of Breaker 5 only. Outages of Connection set (1) + (4)
are caused by some combination of failures of Breaker 5, Bus 6, and Trans-
former 9. It should be noted that all connection set outages are by definition
mutually exclusive.
DISTRIBUTION SYSTEM AND STATION ADEQUACY ASSESSMENT 247

Table 6.22. Outage Probability Variation as a Function of

Breaker Failure Rates
Outage probability
Connection set Case I Case II Case III
(I) 0.303784E-04 0.610557E-04 0.897773E-04
(2) 0.311544E-04 0.58980IE-04 0.842579E-04
(3) 0.323512E-04 0.559409E-04 0.913654E-04
(4) 0.129359E - 03 0.133463E - 03 0.129398E - 03
(5) 0.129863E-03 0.126985E-03 0.12726IE-03
(6) 0.130515E-03 0.124308E-03 0.123885E-03
(1)+(4) 0.39264IE-04 0.393171E-04 0.411952E-04
(2) +(5) 0.40925IE-04 0.401165E-04 0.40262IE-04
(3) +(6) 0.418704E-04 0.417183E-04 0.41266IE-04
(1)+(2)+(4)+(5) 0.186389E-06 0.46684IE-06 0.743696E-06
(2) + (3) + (5) + (6) 0.237587E-06 0.464643E-06 0.685578E-06

Table 6.23. Outage Frequency Variation as a Function of

Breaker Failure Rates
Outage frequency (failure/yr)
Connection set Case I Case II CasellI
(I) 0.203704E-02 0.39753IE-02 0.629630E - 02
(2) 0.217284E-02 0.437037E-02 0.585185E - 02
(3) 0.19753IE-02 0.404938E - 02 0.661728E-02
(4) 0.265802E - 01 0.263827E-OI 0.260864E-OI
(5) 0.267160E-OI 0.25432IE-OI 0.256173E-Ol
(6) 0.261358E - 01 0.258765E-OI 0.254691E-0l
(1)+(4) O.537284E-Ol O.548889E-Ol O.566914E-Ol
(2)+ (5) O.536667E-OI 0.541605E-OI 0.566667E-OI
(3)+ (6) 0.533086E - 01 0.555062E - 01 O.571728E-OI
(I) + (2) + (4) + (5) O.165432E-02 0.395062E-02 0.596296E - 02
(2) + (3) + (5) + (6) 0.202469E - 02 0.407407E-02 O.606273E - 02

Similar sensitivity studies can be performed in which other parameters

such as breaker repair times, bus failure and repair rates, and transformer
failure and repair rates are varied.
The connection set reliability indices are important input data when
incorporating the effects of station configurations in composite system reli-
ability assessment. In other cases, such as when an external point of the
station is a supply point of a radial distribution system, reliability indices
at each single point which are not mutually exclusive of those of other points
are required. Nonmutual exclusion means that other points may be in the
operating state or outage state if one point is in the outage state.
248 CHAPTER 6

The sensitivity studies associated with the point reliability indices are
shown in Figure 6.17. Figure 6.17a shows the effect on the point outage
probability (OP) of varying breaker failure rates (active and passive failure
rates are changed at the same percentage). This figure includes four curves.
Curve I is the simulated outage probability (SOP) at Point (1) and Curve
II is the analytical outage probability (AOP) at this point. Curves III and
IV are the simulated and analytical outage probabilities, respectively, of
Point (4). It can be seen that the outage probability of Point (I) is greatly
affected by the breaker failure rates and that the outage probability of Point
(4) is almost constant. The results obtained by the two calculation techniques
are basically the same. Figure 6.17b shows the effect on point outage rates
(OR). Curve I is the simulated outage rate (SOR) of Point (1) and Curve
II the simulated outage rate (SOR) of Point (4). The outage rates of both
points increase when the breaker failure rates increase. Figure 6.17c shows
the effect on average point outage durations (OD). Both curves are simulated
results (SOD). Curve I shows the value of Point (I) and Curve II the value
of Point (4). Curve I increases as the breaker failure rates increase, which
means that the average point outage duration will be longer when the breaker
failure rates increase. Conversely, the average outage duration of Point (4)
will decrease as the breaker failure rates increase.
The phenomena illustrated in Figure 6.17 can be explained qUalitatively
as follows. Because of the symmetrical locations of Points (1) and (4) and
the specified action logic of Breakers 1 and 5, the outage rates of Points (1)
and (4) are basically the same. It can be determined from the component
reliability data that the transformer has the largest outage probability. How-
ever, the outage of Transformer 9 creates different effects on Point (I) or
(4). Breakers I and 5 will be reclosed after failed Transformer 9 is isolated.
The outage duration of Point (1) caused by this faulted transformer is there-
fore only the transformer switching time. The outage duration of Point (4)
due to the transformer failure is the transformer switching time plus its
repair time. This leads to the fact that Point (4) has a longer outage duration
and therefore higher outage probability than Point (1). When the breaker
failure rates increase, contributions due to Breakers 1 and 5 increase and
therefore the outage rates of Points (1) and (4) also increase. Only one
breaker outage (Breaker 5) is required to cause the outage of Point (1). The
outage probability of this point therefore has a relatively large increase with
breaker failure rates. This leads to increase in the outage duration of Point
(I) with the breaker failure rates. Simultaneous outage of the two breakers
(both Breakers I and 5) is required to create an outage of Point (4). The
outage probability of this point is affected only slightly and therefore its
outage duration decreases with the breaker failure rates.
The outage rate probability distributions at the six points are illustrated
in Table 6.24. Columns 2, 3, 4, 5, and 6 show the probabilities of 0, 1,2, 3,
DISTRIBUTION SYSTEM AND STATION ADEQUACY ASSESSMENT 249

"10 20.0
><

15.0
.~
:sos
.c
e
11.

10.0
____ III. SOP to (4)

--<>- IV. AOP to (4)

5.0
0.000 0.005 0.010 0.015
Failure rate (flyr)
0.075

0.070
"C'
~
e 0.065
Q)

N
0'"
0.060
_ _ I. SORto(l)

-0- II. SOR to (4)

0.055
0.000 0.005 0.010 0.015
Failure rate (f/yr)

30

25
~
5-
c
0
! 20
'"
"0
Q)
C)

~
0 15
_ _ I. SOD to (1)

-0- II. 500 to (4)

10
0.000 0.005 0.010 0.015
Failure rate (f/yr)

Figure 6.17. Reliability indices at points (1) and (4) as a function of the breaker
failure rates.
250 CHAPTER 6

Table 6.24. Point Outage Rate Probability Distributions

Probability of outage rates
Point 0 2 3 4
(I) 0.946160 0.052185 0.001617 0.000037 0.000000
(2) 0.944074 0.054395 0.001519 0.000012 0.000000
(3) 0.946210 0.052259 0.001506 0.000025 0.000000
(4) 0.946235 0.052123 0.001605 0.000037 0.000000
(5) 0.944185 0.054296 0.001519 0.000000 0.000000
(6) 0.946259 0.052210 0.001506 0.000025 0.000000

and 4 outages within 1 yr, respectively. The outage rate probability distribu-
tion at Point (1) is shown in Figure 6.18 for illustrative purpose.
Table 6.25 shows the point outage duration probability distributions.

°
Columns 2, 3, 4, 5, and 6 show the probabilities of outage durations between
to 10 hr/ failure, 10 to 20, 20 to 30, 30 to 40, and 40 to 50, respectively.
The probability of an outage duration longer than 50 hr is not shown in
this table. The outage duration probability distribution at Point (1) is shown
in Figure 6.19.

1.0
Mean outage rate
=0.055581 (occ/yr)
O.B

~ 0.6
:c
«I
.0
0
~

Il. 0.4

0.2

0.0
0 1 2 3 4
Outages per year

Figure 6.18. Outage rate probability distribution at point (1).

DISTRIBUTION SYSTEM AND STATION ADEQUACY ASSESSMENT 251

Table 6 .25. Point Outage Duration Probability Distributions

Probability of outage durations
Point (0-10) (10-20) (20-30) (30-40) (40-50)
(1) 0.761672 0.116719 0.056247 0.021787 0.007337
(2) 0.763695 0.107197 0.052417 0.025779 0.015252
(3) 0.750613 0.111756 0.055989 0.030560 0.015838
(4) 0.410154 0.183923 0.120908 0.076375 0.046538
(5) 0.419681 0.190568 0.110680 0.074289 0.049957
(6) 0.403885 0.184193 0.119893 0.082831 0.048895

1.0

0.8
Mean outage duration
= 11.0199 (hr/failure)
~ 0.6
:0
cu
.0
e
a.. 0.4

0.2

0.0
0-10 10-20 20-30 30-40 40-50
Outage duration (hrlfailure)

Figure 6.19. Outage duration probability distribution at point (I).

6.5. CONCLUSIONS
This chapter discusses the application of the time sequential Monte
Carlo simulation method to distribution system and station reliability
assessment. In each case, component operating histories are randomly
simulated by the component state duration sampling technique. Practical
deterministic events in the operating process, such as a prespecified switch-
252 CHAPTER 6

ing logic or a scheduled maintenance, etc., can be imposed to obtain

modified component operating histories. The system operating history can
be created by combining the modified component operating histories.
Reliability indices at load points (or source points) can be obtained by
evaluating each system state. In the case of radial distribution systems,
the modified component operating histories can also be combined with
the conventional analytical method to obtain the load point reliability
indices. The basic criterion in system state evaluation is connectivity from
source point(s) to load point(s) and is associated with component failures,
switching logic of breakers and section switches, action procedure for
protection devices, and operating guidelines for backup supply sources,
etc. Station reliability assessment is generally more complex than radial
distribution system reliability assessment as it involves multiple source
points, multiple breakers, and usually more complicated protection
coordination.
Mean reliability indices can be readily calculated using the conventional
analytical method. (I) The advantages of the Monte Carlo simulation method
are that any component state duration probability distributions can be simu-
lated, reliability index probability distributions can be calculated, and more
complex operating procedures can be considered. The index probability dis-
tributions provide an added dimension to reliability evaluation. Maintenance
times can be modeled using probability distributions similar to those used
for component failure or repair times. Scheduled maintenance, however, is
not a random event and it may be represented by a prespecified interval.
Both cases can be readily simulated using the time sequential Monte Carlo
method.
Frequency and duration parameters are major indices in distribution
system and station reliability assessment. In distribution system reliability
evaluation, these indices can be divided into two fundamental groups. The
first group contains the three basic load point indices: failure rate, outage
duration, and annual outage time. The second group contains the system
performance indices, in which the most commonly used ones are SAIFI,
SAIDI, CAIDI, ASAI, and ASUI. The system performance indices are
weighted averages of the three basic load point indices. There are also two
basic groups of indices in station reliability assessment. One group contains
the outage probability and outage frequency of the external connection sets.
These sets are mutually exclusive and can be used as input data in composite
system adequacy assessment. The other group contains the outage proba-
bility, outage rate, and outage duration for each individual connection
and are called point indices. These indices are not mutually exclusive
and can be used as input data in radial distribution system reliability
assessment.
DISTRIBUTION SYSTEM AND STATION ADEQUACY ASSESSMENT 253

6.6. REFERENCES
1. R. Billinton and R. N. Allan, Reliability Evaluation of Engineering Systems: Concepts and
Techniques, Plenum, New York, 1983.
2. R. Billinton and R. N. Allan, Reliability Evaluation of Power System, Plenum, New York,
1984.
3. R. Billinton and R. N. Allan, Reliability Assessment of Large Electric Power Systems,
Kluwer, Boston, 1988.
4. R. Billinton and M. S. Grover, "Distribution System Reliability Assessment," CEA Trans.,
Vol. 13, Pt. III, 1974.
5. M. S. Grover and R. Billinton, "Quantitative Evaluation of Permanent Outages in Distribu-
tion Systems," IEEE Trans. on PAS, Vol. 94, 1975, pp. 733-741
6. R. Billinton, J. E. D. Northcote-Green, T. D. Vismor, and C. L. Brooks, "Integrated
Distribution System Reliability Evaluation: Part I-Current Practices," CEA Engineering
and Operating Division Meeting, March 1980.
7. R. Billinton and R. Goel, "An Analytical Approach to Evaluate Probability Distributions
Associated with the Reliability Indices of Electric Distribution System," IEEE Trans. on
PD, Vol. 1, No.3, 1986, pp. 245-251.
8. M. S. Grover and R. Billinton, "A Computerized Approach to Substation and Switching
Station Reliability Evaluation," IEEE Trans. on PAS, Vol. 93, 1974, pp. 1488-1497.
9. M. S. Grover and R. Billinton, "Substation and Switching Station Reliability Evaluation,"
CEA Trans., Vol. 13, Pt. III, 1974, Paper No. 74-SP-151.
10. R. Billinton and T. K. P. Medicheria, "Station Originated Multiple Outages in the Reli-
ability Analysis of a Composite Generation and Transmission System," IEEE Trans. on
PAS, Vol. 100, No.8, 1981, pp. 3870-3878.
11. R. Billinton, P. K. Vohra, and S. Kumar, "Effect of Station Originated Outages in a
Composite System Adequacy Evaluation of the IEEE Reliability Test System," IEEE
Trans. on PAS, Vol. 104, No. 10, 1985, pp.2649-2656.
12. R. Billinton and P. K. Vohra, "Station Initiated Outages in Composite System Adequacy
Evaluation," Proc. lEE, Vol. 134, Pt. C, 1987, pp. 10-16.
13. R. Billinton, "Distribution System Reliability Performance and Evaluation," Electrical
Power and Energy Systems, Vol. 10, No.3, 1988, pp. 190-200.
14. R. Billinton, E. Wojczynski, and M. Godfrey, "Practical Calculations of Distribution
System Reliability Indices and Their Probability Distributions," CEA Trans., Vol. 20, Pt. I,
1981.
15. R. Billinton and E. Wojczynski, "Distributional Variation of Distribution System
Reliability Indices," IEEE Trans. on PAS, Vol. 104, No. 11, 1985, pp. 3152-3160
16. D. O. Koval and R. Billinton, "Statistical and Analytical Evaluation of the Duration and
Cost of Consumer Interruptions," 1979 IEEE/PES Winter Meeting, Paper A79-057-1, Feb.
1979.
17. R. Billinton, E. Wojczynski, and V. Rodych, "Probability Distributions Associated with
Distribution System Reliability Indices," 1980 Reliability Conference for the Electric Power
Industry, 1980.
18. IEEE Committee Report, "Report on Reliability Survey of Industrial Plants, Part II: Cost
of Power Outages, Plant Restart Time, Critical Service Loss Duration Time, and Type
of Loads Lost Versus Time of Power Outages," IEEE Trans. on Industry Applications,
Vol. IA-lO, No.2, 1974, pp.236-241.
19. P. E. Gannon, Power Systems Reliability Subcommittee Report, "Cost of ElectricaI Inter-
ruptions in Commercial Buildings," IEEE 1975 I$CPS Conference, pp. 123-129, 1975.
254 CHAPTER 6

20. E. Wojczynski, R. Billinton, and G. Wacker, "Interruption Cost Methodology and

Results-A Canadian Commercial and Small Industrial Survey," IEEE Trans. on PAS,
Vol. 103, No.2, 1984, pp. 437-444.
21. A. D. Patton, "Probability Distribution of Transmission and Distribution Reliability
Performance Indices," 1979 Reliability Conference for the Electric Power Industry,
pp. 120-123.
22. US Department of Energy, "The National Electric Reliability Study: Executive Summary,"
DOE/EP-003 Dist. Category UC-97C, April, 1981.
23. R. BilIinton, L. Gan, L. Goel, G. Lian, and W. Li, "Applications of Monte Carlo Simula-
tion in Power System Adequacy Assessment," CEA Trans., Vol. 30, Pt. 3, 1991.
24. EPRI Final Report, "Predicting Transmission Outage for System Reliability Evaluation,"
Tech. Report, EPRI, EI-3880, Vol. 1, May 1985.
25. R. BilIinton and G. Lian, "Substation Reliability Worth Assessment Using A Monte Carlo
Approach," 1992 Inter-RAMQ Conference for the Power Industry, Philadelphia,
Aug. 1992.
26. R. BilIinton and G. Lian, "Station Reliability Evaluation Using A Monte Carlo
Approach," IEEE Trans. on PD, Vol. 8, No.3, 1993, pp. 1239-1245.
27. CEA Working Group, "Report on Basic Station Single Line Diagrams in Use by Canadian
Utilities and Consultants," CEA Trans., Vol. 19, Pt. 5, March, 1980.
 7
Reliability Cost/
Worth Assessment

7.1. INTRODUCTION
The basic function of a modern electric power system is to provide electric
power and energy to its customers at the lowest possible cost and at accept-
able levels of reliability. Reliability worth assessment provides the oppor-
tunity to incorporate cost analysis and quantitative reliability assessment
into a common structured framework.(I) Direct evaluation of reliability
worth or benefit is very difficult and perhaps impossible and a variety of
methods have evolved to provide monetary estimates. (2,3) Interruption costs
are most often used to provide an indirect measure of reliability worth. It
is important to realize that, while power system reliability assessment has
become a well established practice over the last few decades, reliability worth
assessment or unreliability cost evaluation is still relatively immature. The
major reason for this is that the quantification of interruption costs is a
complex and often subjective task.
Customer surveys provide a comprehensive approach to evaluate the
impacts of interruptions to customer service due to failures in electric energy
supply. Such surveys are normally undertaken for each user group, e.g.,
commercial, industrial, residential, etc., and can provide reasonably rela-
tively definitive results. (4-12) The University of Saskatchewan has conducted
several systematic customer surveys. The first series was done in 1980-1985
on behalf of the Canadian Electrical Association (CEA), and the second in
1990-1992, sponsored by the Natural Sciences and Engineering Research
Council (NSERC) together with seven participating Canadian electric power
utilities. The data compiled from these surveys have been used to formulate
Sector Customer Damage Functions (SCDF), which depict the sector inter-
ruption cost as a function of the interruption duration. The SCDF can be
255
256 CHAPTER 7

aggregated to create the Composite Customer Damage Functions (CCDF),

which measure the cost associated with power interruptions as a function
of the interruption duration for the customer mix at the bus, or in the area
of interest, or for the whole system. Section 7.2 briefly discusses basic
customer survey techniques and customer damage functions.
Interruption duration, frequency, and load curtailed are three funda-
mental quantities in power system reliability evaluation. Reliability worth
analysis provides a value-based assessment which reflects the integrated
effects of these three quantities. The CDF provides the cost/reliability link
in such an evaluation. Reliability worth assessment can be performed at the
different hierarchical levels and functional zones described in Chapter 2.
Sections 7.3,7.4, and 7.5 discuss generating capacity adequacy worth (HLl),
composite system adequacy worth (HL2), and distribution system adequacy
worth assessment, respectively. This is only one part of an overall economic
analysis. However, it provides extremely important input to the overall
decision-making process. In a power system planning context, an optimal
alternative reinforcement or expansion scheme should achieve minimum
total cost, which is the sum of the investment, operating, and customer
interruption costs. Section 7.6 describes an overall economic assessment
model at HL2.

7.2. CUSTOMER SURVEYS AND

CUSTOMER DAMAGE FUNCTIONS

7.2.1. Basic Customer Survey Methods

The various approaches that have been used to evaluate the impacts of
interruptions in service to electric customers can be grouped into the three
categories of indirect analytical evaluations, case studies of actual blackouts,
and customer surveys. Customer surveys require a large amount of effort
and subsequent statistical analysis. However, they provide the best approach
to interruption cost evaluation as customers are in the best position to
understand how outages impact on those of their activities that depend on
electricity. This approach can also include various complex factors which
cannot be easily considered using other techniques.(4,S) The three basic
approaches to conducting a customer survey are: (1) the contingent valua-
tion method, (2) the direct costing method, and (3) the indirect costing
method. (5) It is possible and desirable to use more than one method in a
given customer survey.
RELIABILITY COST/WORTH ASSESSMENT 257

(a) Contingent Valuation Method. The contingent valuation

method establishes a monetary value of outage cost for incremental changes
in levels of service. This is quantified either through the consumer's willing-
ness to pay (WTP) to avoid having an interruption, or the willingness to
accept (WTA) compensation for having had one. In theory, WTP values
should be nearly equal to WTA values, but actual customer valuations reveal
that WTP values are significantly less than WTA values. This is due to the
fact that customers normally do not have a choice of suppliers, and therefore
their responses may be governed largely by their concern for potential rate
changes. They may react against providing further money for a service they
consider already paid for. Valuations based on WTP and WTA are valuable
measures and may serve as upper and lower bounds, respectively, in cost-
of-outage assessment.

(b) Direct Costing Method. A direct costing method is the most

obvious approach for determining customer interruption costs for given
outage conditions. The respondent is given a "worksheet" and asked to
identify the impacts and evaluate the costs associated with particular outage
scenarios. This approach provides consistent results in those situations where
most losses tend to be tangible, directly identifiable, and quantifiable. Thus
it is most applicable for the industrial sector and for most large electrical
users. It can also be effective in the commercial/retail markets but must be
used with care.

(c) Indirect Costing Method. The indirect costing method is

based on the economic principle of substitution in which the valuation of a
replacement good is used as a measure of worth of the original good. It is
extremely useful when social considerations and inconvenience effects are
expected (or are known) to comprise a significant part of the overall interrup-
tion costs, such as in the residential sector. This approach attempts to pro-
vide a means to lessen the perceptions associated with rate-related
antagonism and customers' lack of experience in rating the worth of reli-
ability. The respondents are asked questions that relate to the context of
their experience. These could include: the cost of hypothetical insurance
policies to compensate for possible interruption effects, preparatory actions
the respondent might take in the event of recurring interruptions, or ranking
a set of reliability/rate alternatives. Evaluations of the financial burden that
customers would be willing to bear to alleviate the effects of the interruption
can be obtained through responses to these questions. The derived expendit-
ures can be considered to be the respondents' perception of the value of
avoiding the interruption consequences. As such, they represent an indirect
estimate of their perception of reliability worth.
258 CHAPTER 7

7.2.2. Questionnaire Content and Data

Treatment
(a) Questionnaire Content. It is obviously desirable to investigate
all possible factors which might affect interruption costs. The length of a
questionnaire, however, is limited by the amount of effort that respondents
are prepared to put in to fill it out. This limitation is particularly relevant
in the residential sector, where a significant portion of the outage costs are
related to somewhat intangible impacts. A questionnaire must therefore be
developed with considerable care. In general, the following factors should
be included:
Residential questionnaire: Interruption characteristics (duration, fre-
quency, and season, day of week, and time of day of the occurrence); user
characteristics (number and age of household members, sex and education
of the respondent); type and location of the dwelling (urban or rural);
electric heat usage; attitude toward utilities; undesirability as a function of
activity interruption; income earning business at home; and user experience
with interruption.
Commercial and industrial questionnaires: Customer uses of electric
energy; type of standby electrical supply equipment and its purpose; past
interruption frequency; cost estimates for a typical interruption time (e.g.,
on Friday at 10:00 am near the end of January) as a function of duration;
cost variations with the month of the year, day of the week, and time of
day; possibility and amount of cost saving if duration of interruption is
known or if advance warning is given; power rationing preference; and
demographic information on the nature and size of the company's operation
(e.g., number of employees, shifts, sales volume, etc.). The industrial ques-
tionnaire also requires start-up time for various interruption durations.

(b) Data Treatment. Customer surveys can generate a considerable

amount of data including some "bad" information, and it is necessary to
conduct appropriate statistical analyses before utilizing these raw data. This
includes calculations of mean values, standard deviations, and correlation
coefficients, and filtering the "bad" data. Such analyses should be conducted
for each customer category, for each service area, and for the regions within
servtce areas.

7.2.3. Customer Damage Functions

(a) Sector Customer Damage Function (SCDF). The SCDF
portrays the unit interruption cost as a function of the interruption duration
RELIABILITY COST/WORTH ASSESSMENT 259

Table 7.1. Sector Customer Damage Functions for the Seven Customer
Categories
Interruption cost of customer sectors (S/kW)
Duration Agri. Large user Resid. Gover. Indus. Commer. Office
1 min 0.060 1.005 0.001 0.044 1.625 0.381 4.778
20 min 0.343 1.508 0.093 0.369 3.868 2.969 9.878
Ihr 0.649 2.225 0.482 1.492 9.085 8.552 21.065
4hr 2.064 3.968 4.914 6.558 25.163 31.317 68.830
8hr 4.120 8.240 15.690 26.040 55.808 83.008 119.160

for an individual customer sector. The Standard Industrial Classification

(SIC) can be used to categorize customers. The seven main sectors are:
larger users, industrial, commercial, agriculture, residential, governmental,
and offices. In order to obtain unit interruption cost estimates, the actual
dollar values reported by customers are normalized using their annual peak
or average demands (in $/kW) and a weighted-mean unit interruption cost
estimate for each category is calculated. If the interruption duration is short,
the weighting factor is the percentage of the annual peak demand of each
respondent with respect to the total annual peak demand of all respondents
in the same category. If the interruption duration is relatively long, the
weighting factor is the percentage of the annual energy consumption of
each respondent with respect to the total annual energy consumption of all
respondents in the same category.
Table 7.1 shows Sector Customer Damage Functions (in $/kW) for the
seven customer categories based on Canadian customer surveys. Figure 7.1
provides a graphical portrayal of the SCDF.

(b) Composite Customer Damage Functions (CCDF). A

CCDF is the measure of the interruption cost as a function of the
interruption duration for a customer mix at a bus, in a service area, or
in a whole system. The customer mix in terms of energy consumption
or peak demand percentages must be known so that the interruption
costs for each of the various customer categories can be proportionally
weighted. In the case of durations shorter than 0.5 hr, the weighting
factor is the percentage of the annual peak load. In the case of durations
longer than 0.5 hr, the weighting factor is the percentage of the annual
energy consumption. The weighted costs are summed for each interruption
duration to give the total cost for the customer mix for that duration.
Table 7.2 shows a representative load mix in terms of both energy
consumption and peak demand. Table 7.3 presents the Composite Cus-
tomer Damage Function based on the Sector Customer Damage Functions
260 CHAPTER 7

200
100

10

~
-~
i ii
0
0
c: .1

a....
0

:::l • large users

Q; .01 .- • Industrial

:E .-/",/'
o commercial
• agricultural
A residential
.001 & govemment &
Institution
o office buildings

.0001
1 minute 20 min 1 hour 4 hr 8 hr

Duration

Figure 7.1. Sector customer damage functions.

Table 7.2. Distribution of Customer Mix

by Energy Consumption and Peak Demand
Energy Peak demand
Customer category (%) (%)
Agriculture 2.5 4.0
Large user 31.0 30.0
Residential 31.0 34.0
Government 5.5 6.0
Industrial 19.0 14.0
Commercial 9.0 10.0
Office 2.0 2.0

in Table 7.1 and the customer mix in Table 7.2. Figure 7.2 provides a
graphical portrayal of the CCDF.
Interruption cost estimates can be used as a surrogate for electric power
system reliability worth. In applying CDF, three important points should
be appreciated. First, as with other reliability data, a sufficient number of
RELIABILITY COST/WORTH ASSESSMENT 261

Table 7.3. Composite Customer Damage

Function for the Customer Mix
Interruption duration Interruption cost ($/ kW)
1 min 0.67
20 min 1.56
1 hr 3.85
4hr 12.14
8 hr 29.41

100
- -
-
.-

~
--
::s:. 10
-
/
V
-
fh
"-"
( J)
.
0

V
()
c:
0
li - ..
......
--=
::J
Q) ~
~ ~ I'

-
.1
1 minute 20 min 1 hour 4 hr 8 hr
Duration

Figure 7 .2. Composite customer damage function.

customer samples is required to ensure reasonable accuracy. Second,

customer surveys should be continuously updated. Third, interruption costs
are socioeconomic/ demographic/ geographic specific. In other words, cus-
tomer damage functions are generally different for different areas, regions,
and countries.
262 CHAPTER 7

7.3. GENERATING SYSTEM RELIABILITY

WORTH ASSESSMENT
7.3.1. Assessment Techniques
The most popular HLl adequacy index at the present time is the loss
of load expectation (LOLE). This index has been used by some utilities for
a long time and is being considered and introduced by others. This index
cannot, however, be easily linked with customer outage costs to assess HLI
adequacy worth. The most convenient index for this purpose is the loss of
energy expectation (LOE5) or expected energy not supplied (EENS).
Billinton et al. (13) show the development of a monetary value for
unserved energy using the system CCDF. Two techniques are presented.
The first is an analytical approach based on the classical frequency and
duration technique. (14) The second approach uses sequential Monte Carlo
simulation to calculate the unserved energy cost. The results in the case
studied are relatively close. The monetary value associated with unserved
energy has been designated as the Interrupted Energy Assessment Rate
(IEAR).(13) This monetary value can be used in a wide range of studies and
developed for application at all three hierarchical levels. Equations (7.1),
(7.2), and (7.3) are the basic equations used in the analytical application of
this concept. Given that it is possible to calculate the frequency and duration
associated with a load loss event, the EENS in MWh/yr is given by equation
(7.1):
N
EENS = I CiFiDi (7.1)
i=1

where Ci is the load curtailment ofload loss event i in MW, Fi the frequency
ofload loss event i in occ./yr, Di the duration of load loss event i in hours,
and N is the total number of load loss events.
The total Expected Interruption Cost (EIC) in k$/yr is given by
N
EIC= I CjFjW(Dj) (7.2)
i= 1

where W(D j ) is the customer damage function, i.e., the unit interruption
cost for the duration D j of load loss event i.
The Interrupted Energy Assessment Rate (lEAR) in $/kWh is defined
as
~~I C;F;W(Dj)
lEAR =_L..._,--=-_ __ (7.3)
I::I C;F;D j
RELIABILITY COST/WORTH ASSESSMENT 263

The lEAR is an important concept in generating system reliability assess-

ment and is quite stable for a given generating system, i.e., it generally does
not vary significantly when the system load level and other factors change.
Consequently, EIC for different load levels can be obtained by multiplying
the lEAR by the EENS, which can be calculated using a wide variety of
methods.
Equations (7.1)-(7.3) apply not only to the analytical approach but
also to the system state sampling technique. This is due to the fact that the
system state sampling technique creates load curtailments, frequencies, and
durations of individual load loss events. It is important to appreciate that
the analytical approach uses average duration and frequency values for D;
and F;in equations (7.1), (7.2), and (7.3). The cost components are therefore
based on average values rather than on specific random events and the load
loss event duration and frequency distributions. This has little effect on the
EENS index but can lead to errors in the EIC and in tum to the lEAR,
particularly in the case of customer damage functions with a relatively high
degree of nonlinearity. Specific random system states can be simulated if the
system state sampling technique is used. This is illustrated in Section 7.4 for
composite system reliability worth assessment.
In a sequential Monte Carlo simulation, individual load loss events are
encountered sequentially and therefore not only the specific random system
states but also the transition process between system states can be simulated.
In other words, the effects of the load loss event duration and frequency
distributions can be considered in the simulation. The state duration sam-
pling technique for generating system reliability evaluation is discussed in
detail in Chapter 4. The EIC and lEAR using this technique are obtained
from equations (7.4) and (7.5):

L~l W(D;)E;/D;
EIC=-'------ (7.4)
M

L~l W(D;)E;/D;
lEAR = -'-------;.,---- (7.5)
L~lE;

where W(D;) is the customer damage function in $jkW, D; the duration of

interruption i in hours, E; the energy not supplied of interruption i in MWh,
N the total number of interruptions experienced in simulated years, and M
is the number of simulated years.
It is noted that interruption i in a sequential system capacity margin
profile can be a single load loss event or a sequence of two or more successive
load loss events.
264 CHAPTER 7

7.3.2. Numerical Example

Consider the lEEE RTS for which the basic data are given in Appendix
A.I. The additional data associated with the generating unit derated states,
peaking units, and the nonexponential distribution state durations are given
in Section 4.2.4. The composite customer damage function shown in Table
7.3 was used in this study. The following results(15) were obtained using the
component state duration sampling technique:
Case 1: Base Case-The EIC and lEAR indices in the base case are:
EIC=6255.7 k$/yr and lEAR = 5.310 $/kWh.
Case 2: Peak Load Variations-The results for different peak loads are
shown in Table 7.4. It can be seen that although the value of EIC increases
as the system peak load increases, the estimated lEAR does not change
significantly with peak load. This is an important observation.

Case 3: Generating Unit Derated States-This case is the same as Case

2, except that the 350-MW and the 400-MW units were assumed to have
one derated state. The results are shown in Table 7.5. Recognition of generat-
ing unit derated states has minimal effects on the lEAR.
Case 4: Peaking Units-This is the same as Case I for the annual
peak load of 2850 MW, except that additional 25-MW gas turbine units are
assumed to be peaking units. The results are given in Table 7.6.
Case 5: Nonexponential Distributions-This is the same as Case I for
the annual peak load of 2850 MW, except that unit operating or repair
durations are assumed to be nonexponentially distributed. The four distribu-

Table 7.4. EIC and lEAR Indices for Different Peak loads
Annual system peak load (MW)
Index 2650 2750 2850 2950 3050
EIC (k$/yr) 1343.3 2935.4 6255.7 12508.0 23909.1
lEAR ($/kWh) 5.293 5.296 5.310 5.317 5.328

Table 7.5. EIC and lEAR Indices for the Derating Case
Annual system peak load (MW)
Index 2650 2750 2850 2950 3050
EIC (k$/yr) 857.1 1857.9 3989.8 8401.7 16421.5
lEAR ($/kWh) 5.270 5.281 5.264 5.290 5.306
RELIABILITY COST/WORTH ASSESSMENT 265

Table 7.6. EIC and lEAR Indices for the Peaking Unit
Addition Case
Number of peaking units added
Index Zero units One unit Two units
EIC (k$/yr) 6255.7 4957.0 3322.2
lEAR ($/kWh) 5.310 5.310 5.282

Table 7.7. EIC and lEAR Indices for the Nonexponential Distribution
Case
Nonexponential distribution conditions
Index Cond. 1 Cond.2 Condo 3 Cond.4
EIC (k$/yr) 6221.8 6155.7 6055.8 6257.5
lEAR ($/kWh) 5.302 5.310 5.303 5.301

tion conditions are identical to those given in Section 4.2.4.(d). The results
are shown in Table 7.7. The effects of the different distributions on the EIC
and lEAR values are minimal in this case. It should be noted that this is
possibly not a general conclusion for the EIC index.

7.3.3. Application Of Reliability Worth

Assessment in Generation Planning
HLI adequacy depends upon many factors, such as the installed capac-
ity, generating unit availability and size, load profile, maintenance require-
ments, reserve margin, etc. Additional generating capacity above the system
peak load is generally maintained to protect against excessive unforeseen
outages and loads in excess of forecasts and to provide opportunities for
generating unit preventive maintenance. System adequacy can be signifi-
cantly increased by maintaining a higher reserve level, which results in
increased installation costs and reduced supply interruption costs. The selec-
tion of an optimal adequacy level, therefore, is an important engineering
decision which should consider the costs of providing the specified reliability
level and the corresponding benefits accruing to society due to these levels.
Basic criteria, such as loss of load expectation (LOLE), cannot be used to
relate system adequacy to economic parameters such as customer costs due
to supply interruptions. Indices such as EENS which recognize the severity
associated with generation capacity deficiencies must be used. There is a
growing interest in generation system expansion planning using economic
theory approaches which simultaneously optimize system costs and customer
266 CHAPTER 7

Table 7.S. Generating Unit Operating Costs

Rated Fixed Variable
capacity cost cost
(MW) Type (S/kW/yr) (S/MWh)
12 Oil 10.0 63.30
20 Combus. 3.0 103.60
Turbine
50 Hydro 2.5 0.50
76 Coal 10.0 15.30
100 Oil 8.5 52.80
155 Coal 7.0 12.44
197 Oil 5.0 50.62
350 Coal 4.5 12.10
400 Nuclear 5.0 6.30

failure costs. Power system planners are faced with the difficult task of
maintaining reasonable system reliability levels while minimizing the invest-
ment and operating costs. Judicious decisions can be made by estimating
the costs and benefits associated with various levels of expansion and includ-
ing the inherent uncertainties associated with system parameters.
The application of reliability worth assessment in generation planning
is illustrated using the IEEE R TS. The basic data of the IEEE R TS are
given in Appendix A.I. The initial analysis was done by assuming that the
total system peak load is increased from 2850 MW to 3000 MW while the
annual load curve shape remains unchanged, i.e., the 8736 hourly loads
are increased proportionally. The total installed capacity provided by 32
generating units is 3405 MW. The associated reserve margin is 13.5% for
this system peak load. Under these conditions, the expected customer inter-
ruption cost is 16,860 k$/yr. The total fixed and variable operating cost is
126,740 k$/yr. The system operating cost was obtained using the generating
unit operating costs given in Table 7.8(16) and an economic generation load-
ing approach. The purpose of the study is to determine the least cost reserve
margin and five additional20-MW units were considered sequentially. These
units were assumed to have the same failure data as the 20-MW unit given
in Appendix A.l and a capital cost of $17 million per unit. The annual
investment cost was calculated using the present value method (see Section
7.6.2). Under the assumption of a 30-yr economic life and a 10010 discount
rate, the annual investment cost is 1802 k$/yr. Table 7.9 presents the
expected interruption, investment, and operating costs for the base case and
with the sequential addition of five 20-MW units.
It can be seen that the expected interruption costs decrease rapidly as
additional capacity is added to the system while the operating costs increase
slightly. There are great differences between absolute values of the interrup-
RELIABILITY COST/WORTH ASSESSMENT 267

Table 7.9. Interruption, Investment, and Operating Costs with

Sequential Capacity Additions of 20-MW units
Reserve Investment Operating
Number of margin EIC cost cost
units added (%) (kS/yr) (kS/yr) (kS/yr)
0 13.50 16860 0 126740
1 14.l6 14606 1802 126930
2 14.83 12651 3604 127010
3 15.50 11170 5406 127060
4 16.l7 9960 7208 127090
5 16.83 8680 9010 127120

tion, investment, and operating costs. An incremental comparison approach

can be used to highlight the effect of the added units, since the investment
is associated only with the added units while the interruption and operating
costs are for the whole system. The increment costs are shown in Table 7.10.
The values in the second column are the interruption cost reduction caused
by unit addition(s) compared to the base case, and those in the fourth
column are the increase in the operating cost. The values in the last column
are the annual net gains, which are presented graphically in Figure 7.3. The
maximum benefit or the least cost reserve margin occurs with the addition
of the two 20-MW units. The reserve margin and the LOEE at this point
are 14.83% and 2386.94 MWh/yr respectively. The optimum reserve margin
is obviously dependent on the data used in the system evaluation, including
the perceived customer interruption costs. The optimum reserve margin is
also very dependent on the size and type of the units used in the proposed
expansion and will vary with different proposed configurations. It is not,
however, a fixed predetermined value which can be used under all conditions
and expansion scenarios.

Table 7.10. Incremental Costs and Annual Net Gains with Sequential
Capacity Additions of 20-MW Units
EIC Investment Operat. cost Annual
reduction cost increase net gain
Number of (kS/yr) (kS/yr) (kS/yr) (kS/yr)
units added (1) (2) (3) (1) - [(2) + (3)]
0 0 0 0 0
1 2254 1802 190 +262
2 4209 3604 270 +335
3 5690 5406 320 -36
4 6900 7208 350 -658
5 8180 9010 380 -1210
 268 CHAPTER 7

400

200
Number of units added
0
-c- 1 2 4 5
I'd
Q) -200

-
~
~
~ -400
c:::
.Cij

-01
-600
Q)
z -800

-1000

-1200

-1400
Figure 7.3. Annual net gains due to 20-MW unit additions for the IEEE RTS generating
system.

In the explicit cost approach, the reserve margin is an outcome of the

analysis, not a fixed criterion used to drive the unit addition process. It is
noted that the analysis presented here is only a simple example and factors
such as effects of load growth, unit size and type, and different operating
conditions, etc., are not considered. These can be easily included using the
procedure presented.

7.4. COMPOSITE SYSTEM RELIABILITY

WORTH ASSESSMENT

7.4.1. Assessment Method

HL2 adequacy assessment involves the consideration of transmission
network constraints, generation rescheduling, overload alleviation, bus load
curtailment philosophies, etc., in addition to the recognition of equipment
failure and repair phenomena. The basic techniques for composite system
adequacy assessment are presented in Chapter 5 and can be extended to
composite system reliability worth assessment by introducing certain addi-
tional considerations. The fundamental procedures, however, remain the
same.
RELIABILITY COST/WORTH ASSESSMENT 269

(a) Minimum System Interruption Cost Model. The optimiza-

tion model given in Section 5.2 minimizes the total system load curtailment
while satisfying the power balance, linearized load flow equations, line flow
limits, and generation output limits. In order to conduct reliability worth
assessment, it is necessary to introduce customer damage functions into the
model. The following linear programming minimization model can be used
for a particular drawn system state to calculate the possible load curtailments
at different buses and the interruption costs for the system and for each load
bus:
NC IC.
min TD= I I Wy(Dk)Cij (7.6)
i=1 j=1

subject to

(i= 1, ... , L) (7.7)

NO NC IC. NC IC.
I PGi + I I Cij= I I PDij (7.8)
i=1 i=1 j=1 i=1 j=1

(i= 1, ... ,NG) (7.9)

(i= 1, ... , NC;j= 1, ... ,ICi ) (7.10)

(n= 1, ... , L) (7.11)

where PG1 is the generation variable at Bus i; PDij and Cij are the load and
the load curtailment variables, respectively, ofthejth sector customer at Bus
i; Tn is the line flow on Line n; Ani is an element of the relation matrix
between line flows and power injections; PG7'in, PG7'ax, and Tr;:ax are the
limits, respectively, of PGi and Tn; Wy(Dk) is the customer damage function
(in $/kW) of the jth sector customer at Bus i; Dk is the duration of the
system state k; NS, NG, NC, and L are the numbers of system buses,
generator buses, load buses, and system lines, respectively; ICi is the number
of customer sectors at Bus i.
The objective of this model is to minimize the total system interruption
cost while satisfying the power balance, the linearized load flow relationships,
line flow limits, and generation output limits. Customer damage functions
for different customer sectors at load buses and the effect of system state
durations on interruption costs have been incorporated in the model. The
constraints in this model are essentially the same as those in the model given
in Section 5.2. The difference between them is associated with the objec-
tive function. The objective of this model is to minimize the total system
270 CHAPTER 7

interruption cost while that of the model in Section 5.2 is to minimize the
total load curtailment. The two load curtailment philosophies described in
Section 5.2 are performed, respectively, by introducing weighting factors Wi
and pj in equation (5.20). The load curtailment philosophy in this model is
that the load with the lowest interruption cost is curtailed first, then that
with the next lowest interruption cost, and at the last that with the highest
interruption cost. This philosophy is followed automatically in the resolution
of the model. This model also calculates all the adequacy indices for compos-
ite system evaluation defined in Section 5.2.4. The system adequacy indices
obtained by the two models should be basically the same. The bus adequacy
indices, however, can be different due to the different load curtailment philos-
ophies used.

(b) Determination of System State Duration. In solving the

above minimization model, the system state duration Dk must be known. It
has been pointed out in Section 3.6.3 that if the state duration of each
component follows an exponential distribution with parameter Ai, where Ai
is the transition rate of component i at the present state, the duration of
the system state also follows an exponential distribution with parameter
A= I Ai. This means that the duration of the system state has the probability
density function

(7.12)

where m is the number of system components.

Two methods can be used to determine the system state duration Dk in
equation (7.6). The first method is to use the mean value of the system state
duration, i.e.,

(7.13)

Note that the transition rates Ai of components are dependent on the system
state. If the component i is in the up state for a drawn system state k, Ai is
its failure rate. If it is in the down state, Ai is its repair rate. Utilization of
the mean value of the system state duration can lead to errors, particularly
when the relevant customer damage functions have a high degree of non-
linearity. However, calculations indicate that in some cases, this error is not
significant and therefore can be accepted. The second method is to use the
sample value of the system state duration. In terms of the inverse transform
method, the sample value Dk of the system state duration having the
RELIABILITY COST/WORTH ASSESSMENT 271

probability density function shown in equation (7.12) can be obtained by

In U
Dk=--m-- (7.14)
LHAi
where U is a uniformly distributed random number between [0, 1].
Both methods given above are based on the assumption that the compo-
nent state durations follow exponential distributions. When the component
repair times follow nonexponential distributions, the distribution of the sys-
tem state duration for composite systems is complex and there is no simple
analytical expression for the distribution density function. If it is assumed
that the system state duration follows a known distribution such as a Weibull
or gamma distribution, the sample value of the system state duration can
be obtained using the Monte Carlo sampling methods given in Section 3.4.

(c) Index Calculations. Load curtailments for each customer sector

at each bus Cij can be obtained by solving the minimization model. Conse-
quently, both bus and system indices can be calculated. Note that Cij in the
model corresponds to a particular system state k whose duration is Dk • The
subscript k in the variables, except Dk in the above model, has been omitted
for simplicity. The annual bus and system indices are given by the following
equations:
N IC,
EENSi = L L CijkFkDk (7.15)
k=lj=1
N IC,
EIC;= L L CijkFkWy(Dk) (7.16)
k=1 j=1
NC
EENS = L EENS; (7.17)
;=1
NC
EIC= L EIC; (7.18)
;=1

where Cijk, Dk, W;j, NC, and IC; are as defined earlier; EENS; and EENS
are the Expected Energy Not Supplied at Bus i and for the system, respec-
tively (MWhjyr); EIC; and EIC are Expected Interruption Costs at Bus i
and for the system (k$jyr); N is the number of system states sampled; Fk
is the frequency of system state k (occ. jyr). The value of Fk can be obtained
from the following general relationship between Fk, Dk, and Pk :
Pk=FkDk (7.19)
where Pk is the probability of system state k.
272 CHAPTER 7

As in generation reliability worth evaluation, IEAR indices at HL2 can

also be calculated. The IEAR indices for the system and buses are given by
equations (7.20) and (7.21), respectively:

IEAR= EIC (7.20)

EENS

IEAR.= EICj (7.21)

I EENSj

Analyses indicate that the lEAR at HL2 does not vary significantly with
the system peak load. However, it may have a relatively large variation when
the system topological configuration changes. This is due to the fact that
changes in the system configuration lead to a change of network constraints
which, in turn, may change the order of bus load curtailments when minimiz-
ing the total interruption cost. This case can happen particularly when a
line addition creates a new loop from an original radial branch. Therefore,
the HL2 IEAR should be utilized with caution when the system con-
figuration changes significantly. In this case, it is advisable to evaluate EIC
directly.

7.4.2. Modified IEEE Reliability Test System

and Additional Data
The original IEEE RTS version(17) has a relatively strong transmission
network. Berloldi et al. (18) suggested several modifications based on an EPRI
version of the IEEE RTS(19) for the purpose of conducting transmission
planning studies. The Modified IEEE Reliability Test System (MRTS) was
used to conduct the case studies described in this section and in Section 7.6.
The basic data of the IEEE RTS are given in Appendix A.1. The modifi-
cations and additional data are as follows.

(a) Load and Installed Capacity. The system peak load was
increased to 125% of the annual peak value given in the original version,
assuming that the distribution among the load buses and the percentage
values of the 8736 hourly loads remain unchanged. Eight generators are
added at the selected buses shown in Table 7.11. The failure data of these
eight generators are identical to those of the generators having the same
capacity in the original system.
RELIABILITY COST/WORTH ASSESSMENT 273

Table 7.11. Added Generating

Units Based on the Original
IEEE RTS
Bus No. Unit size
76MWx 1
2 76MWx 1
13 197MWx 1
22 50MWx4
23 350MWx 1

(b) Transmission Network. Six lines and one transformer branch

were removed from the original version and are shown by the dashed lines
in Figure 7.4.

(c) Additional Data. The additional data required in order to con-

duct composite system reliability worth assessment are the sector customer
damage functions and the customer sector allocations at load buses. The
seven sector customer damage functions are the same as those given in Table
7.1. The customer sector allocations are given in Table 7.12.

7.4.3. Numerical Results

(a) Base Case. Reliability worth evaluation for the IEEE MRTS
was conducted using the described method. A 15-step load model of the
annual load duration curve was used to evaluate the annual bus and system
EENS, EIC, and lEAR indices. The results for the base case are given in
Table 7.13. There are no load curtailments and interruption costs at Buses
1, 2, 4, 5, and 8. This can be explained as follows. In addition to effects
associated with the system topology and load flow constraints, customer
sector allocations and specific SCDFs are the important aspects causing
differences between load curtailments and interruption cost indices at the
various buses. It can be seen from Table 7.1 that agriculture and large
industrial users have the lowest unit interruption costs. When a minimum
interruption cost principle is utilized in the resolution of the linear program-
ming model, agriculture and large industrial users are automatically curtailed
first. There are, however, no agriculture and large industrial customers at
Buses 1,2,4,5, and 8. In addition, these buses are either the generator buses
(l and 2) or close to the generator buses (4, 5, and 8).

(b) Effect of Load Level Variations. Annual system EENS and

EIC indices and the corresponding lEAR for various load levels are shown
274 CHAPTER 7

230kV

~...+-BUS12

1L..-_
I _ _ _ _ _-J
"',
"
' ..
' .., BUSS
138kV ' ....
"
'"
".
BUS 1

Figure 7.4. IEEE MRTS network configuration.

in Table 7.14. The EENS and EIC indices increase with load levels while
the system lEAR remains virtually constant.

(c) Effect of Composite Customer Damage Functions.

Replacing the sector customer damage functions by composite customer
damage functions may lead to different conclusions. The bus CCDFs for the
IEEE MRTS were calculated using the customer sector allocations shown in
Table 7.12 and the SDCFs given in Table 7.1. These CCDFs are presented
in Table 7.15.
RELIABILITY COST/WORTH ASSESSMENT 275

Table 7.12. Customer Sector Allocations at Load Buses

Load percentage of customer sector (in %)
Peak
Bus load Large
No. (MW) Agri. user Resid. Gover. Indus. Commer. Office
I 135.00 0.0 0.0 34.03 15.83 36.94 13.20 0.0
2 121.25 0.0 0.0 50.05 35.26 0.0 14.69 0.0
3 225.00 6.33 0.0 52.50 0.0 33.25 7.92 0.0
4 92.50 0.0 0.0 34.52 46.22 0.0 19.26 0.0
5 88.75 0.0 0.0 51.38 0.0 28.10 20.07 0.0
6 170.00 8.38 0.0 49.70 0.0 29.34 10.48 2.10
7 156.25 18.24 0.0 38.44 0.0 31.92 11.40 0.0
8 213.75 0.0 0.0 55.00 15.00 11.67 16.66 1.67
9 218.75 19.54 48.86 23.46 0.0 0.0 40.87 3.27
10 243.75 8.77 21.92 41.54 0.0 20.46 7.31 0.0
13 331.25 6.45 16.13 30.09 9.69 22.59 10.75 4.30
14 242.50 0.0 44.07 32.42 0.0 20.57 2.94 0.0
15 396.25 0.0 67.43 17.29 0.0 0.0 10.78 4.50
16 125.00 0.0 42.75 25.90 17.10 0.0 14.25 0.0
18 416.25 0.0 56.49 18.69 0.0 11.98 6.85 5.99
19 226.25 0.0 61.41 30.72 0.0 0.0 7.87 0.0
20 160.00 0.0 33.40 42.11 13.36 0.0 11.13 0.0

Table 7.13. Bus and System EENS, EIC, and lEAR for
the IEEE MRTS for the Base Case
Bus No. EENS EIC lEAR
or system (MWh/yr) (k$/yr) ($/kWh)
0.00 0.00
2 0.00 0.00
3 259.96 160.98 0.619
4 0.00 0.00
5 0.00 0.00
6 41.97 21.62 0.515
7 32.42 16.69 0.515
8 0.00 0.00
9 61.16 32.32 0.528
10 78.87 41.47 0.525
13 32.06 16.51 0.515
14 1.13 1.15 1.018
15 2.66 2.72 1.023
16 66.10 69.57 1.052
18 76.24 78.78 1.033
19 664.96 691.84 1.040
20 574.43 597.92 1.041
System 1891.96 1731.57 0.915
276 CHAPTER 7

Table 7.14. System EENS, EIC, and lEAR in the IEEE MRTS at Different
Load Levels
Load Peak load EENS EIe IEAR
increment (%) (MW) (MWh/yr) (k$/yr) ($/kWh)
0.0 3562.5 1891.96 1731.57 0.915
4.0 3705.0 4331.45 3968.66 0.916
6.0 3776.3 6254.57 5733.87 0.917
8.0 3747.5 8855.44 8103.47 0.915
10.0 3918.8 12436.87 11319.84 0.910

Table 7.15. Bus Composite Customer Damage Functions for

the IEEE MRTS ($/kW)
Bus No. 1 min 20 min 1 hr 4hr 8hr
0.6579 1.9108 4.8851 16.1394 41.0340
2 0.0720 0.6128 2.0236 9.3723 29.2284
3 0.5748 1.5918 3.9922 13.5575 33.6284
4 0.0941 0.7745 2.5031 10.7591 33.4392
5 0.5336 1.7310 4.5191 15.9031 40.4739
6 0.6226 1.7284 4.2981 14.7255 35.7186
7 0.5735 1.6714 4.1785 13.8676 34.0596
8 0.3401 1.2175 3.3257 12.9898 34.8674
9 0.6778 1.2933 2.4323 7.2708 16.4510
10 0.5863 1.4077 3.2288 10.5297 26.1714
13 0.7840 1.9468 4.5678 14.8979 35.4935
14 0.7887 1.5777 3.2570 9.4386 22.6382
15 0.9339 1.7975 3.4535 9.9986 22.5795
16 0.4917 1.1549 2.5498 8.5531 23.8678
18 1.0749 2.1277 4.2839 12.4426 27.0968
19 0.6475 1.1893 2.1875 6.4110 16.4129
20 0.3844 0.9226 2.0973 7.7563 22.0770

The results obtained using the bus CCDFs with other conditions being
the same as those in the base case are given in Table 7.16. The buses at
which load curtailments and interruption costs are zero are not listed. It can
be seen that load curtailments occur only at Buses 3, 9, 10, 14, 15, 16, 19,
and 20 and mainly at Buses 9 and 19. This can be explained as follows.
Most system state durations are longer than one hour. It can be seen from
Table 7.15 that these buses, particularly Buses 9 and 19, have relatively low
unit interruption costs beyond I-hr outage duration. In addition, Buses 9
and 19 are relatively far away from generation sources. When SCDFs are
used, customer sectors with lowest unit interruption costs will be curtailed
first in the minimum interruption cost model. When bus CCDFs are used,
load curtailments will be performed in terms of the relative magnitudes of
RELIABILITY COST/WORTH ASSESSMENT 2n

Table 7.16. Bus and System EENS. EIC. and lEAR for
the IEEE MRTS Using Bus CCDF
Bus No. EENS EIe lEAR
or system (MWh/yr) (k$/yr) ($/kWh)
3 55.23 244.38 4.425
9 169.31 355.30 2.099
10 0.35 1.07 3.057
14 4.99 13.39 2.683
15 4.34 12.37 2.850
16 0.69 1.85 2.681
19 1519.32 3296.56 2.170
20 96.01 281.20 2.905
System 1850.24 4206.12 2.273

bus CCDFs. Comparing the system indices obtained using the two
approaches, it can be seen that the system EENS is basically the same in
the two cases while the systems EIC and lEAR are quite different. The
utilization of SCDFs or bus CCDFs reflects two completely different bus
load curtailment philosophies. Utilization of CCDFs may create the much
higher expected system interruption costs and lEAR values compared to
using SCDFs.

7.4.4. Application of ReliabilityWorth Assess-

ment in Composite System Planning
Composite system expansion planning is basically concerned with the
addition of new generation and transmission facilities at a specified time in
the future and at appropriate location(s). The objective is to select the most
economical and reliable expansion plan(s) in order to meet the expected
future load growth at minimum cost and optimum reliability subject to
economic and technical constraints. Comparison analysis between the
annual Expected Interruption Cost and the annual investment of different
alternatives is often utilized in composite system planning. This section
examines the impacts of additional generating and transmission facilities
using the IEEE MRTS.
The basic data for the IEEE MRTS is given in Section 7.4.2. A IS-step
load model for the annual load duration curve was used to evaluate the
annual expected interruption costs of three alternatives. The first alternative
(AI) is the addition of a circuit between Buses 16 and 17 (18 miles). The
second alternative (A2) is the addition of a circuit between Buses 20 and 23
(15 miles). The third alternative (A3) is the addition of a generating unit at
 278 CHAPTER 7

Table 7.17. EIC Indices of the Base System and Three

Alternatives for the IEEE MRTS (k$/yr)
Load
Alternative
increment Base
(%) system Al A2 A3
0.0 1731.57 1726.31 569.26 603.24
2.0 2636.98 2637.51 907.64 940.18
4.0 3968.66 3967.54 1427.01 1441.66
6.0 5733.87 5727.55 2178.04 2136.45
8.0 8103.47 8082.22 3251.35 3090.21
10.0 11319.84 11287.14 4813.06 4461.00

Bus 16 (155 MW). The annual expected interruption costs (EIC) of the base
system and three alternatives for load increments from 0.0010 to 10.00/0 are
shown in Table 7.17. Figure 7.5 is a graphical representation of the results.
It can be seen that Alternative I has basically the same EIC indices as the
base case. This means that the addition of the circuit between Buses 16 and
17 does not improve the reliability of the system. This is an example of the
noncoherence concept (the loosened definition) described in Section 5.10.
A2 is the addition of a shorter circuit between Buses 20 and 23 which leads
to much lower EIC than AI. This indicates that the different line addition

12.0
-....-

--
11.0 Basic System
10.0 -----0---- Alternative 1
Alternative 2
9.0
'i:"
-----60---- Alternative 3
as 8.0
Q)
7.0
~
-
:IE
0
jjj
6.0
5.0
4.0
3.0
2.0
1.0 ----------_. ---
0.0
0 2 4 6 8 10
Load increment (%)
Figure 7.5. Variations in EIC index for the IEEE MRTS with load increments.
RELIABILITY COST/WORTH ASSESSMENT 279

locations have completely different impacts on composite system reliability.

It is interesting to note that A2 and A3 have basically the same EIC indices
for load increases of 0.0% to 10.0%. A 155-MW generating unit requires
much higher investment cost than one 15-mile 230-kV circuit. The addition
of the 155-MW unit at Bus 16 should therefore be avoided or at least
deferred.
Further analyses can be conducted to evaluate the total cost, which is
the sum of the annual expected interruption cost and the annual investment
for the alternatives. The economic life of the power system facilities was
assumed to be 30 yr and the discount rate 10%. The unit capital cost of a
230-kV line is 90 k$/mile and the capital cost of a 155-MW generating unit
100 M$. The annual investment for the 230-kV circuit of 15 miles (A2) and
the 155-MW generating unit (A3) can be calculated using the present value
method (see Section 7.6.2) and the given economic data. The calculated
annual investment costs of A2 and A3 are 143.21 k$/yr and 10,607.9 k$/yr,
respectively. At the present load level (0.0% increment), Alternative 2 can
reduce the expected interruption cost by 1162.29 k$/yr (1731.55-569.26).
This reduction is much larger than its annual investment of 143.21 k$/yr
and therefore Alternative 2 is a beneficial option even for the present load
level. On the other hand, the reduction of the EIC due to Alternative 3
at 10.0% load increment level is 6858.84 k$/yr (11,319.84-4461.00). This
reduction is still smaller than the annual investment of A3 (10,607.9 k$/yr).
This indicates that the addition of the 155-MW unit is not a cost-effective
option even when the load has 10.0% growth.
The above analysis is based on composite system reliability worth
assessment. The method considers both investment and reliability worth on
a unified basis. The system operating cost is not included in the analysis.
This is discussed in Section 7.6.3 where it is shown that, although Alternative
1 does not improve system reliability, it improves the operating economy in
the system and is a good option from an overall minimum cost viewpoint.
It is therefore necessary to incorporate the system operating costs in the
model to provide a more complete analysis. This is discussed in more detail
in Section 7.6.

7.5. DISTRIBUTION SYSTEM RELIABILITY

WORTH ASSESSMENT
7.5.1. Assessment Techniques
As previously noted, both analytical and Monte Carlo techniques can
be used to provide average load point reliability indices. In order to include
280 CHAPTER 7

the effects of load point outage duration distributions on interruption costs,

however, it is usually necessary to apply a simulation approach. The
system operating cycle technique given in Section 6.3.2 can be used quite
effectively.
Three basic models are required for distribution system reliability
worth assessment. These are the cost model, the load model, and the
system model. The cost model is associated with the customer type at
the load point and described by an appropriate customer damage function
(CDF). The load model can be in the form of either average loads or
actual time-dependent loads in a chronological load curve. The system
model consists of the system configuration, relevant reliability parameters
of all components such as the main feeders and lateral branches serving
the individual customers, and the operating logic. The system model also
involves additional factors such as inclusion or not of disconnects on the
main feeders, fuses in lateral branches, alternate backup supply, replacing
a failed low voltage transformer with a spare instead of repairing it,
main and/or lateral branches consisting of underground cables or over-
head lines, etc. The individual failure events are determined by simulation
and the outage events evaluated in terms of the failure rate, average
outage duration, annual outage time, and load curtailments at each load
point. The basic system analysis method and evaluation technique are
discussed in Chapter 6. The following formulas can be used in conjunction
with the state duration sampling technique to evaluate the load point
EENS, EIe, and lEAR:

(7.22)

(7.23)

(7.24)

where Wk(D i) is the customer damage function at load point k, Di the

sampling value of duration for interruption i, PUc the demand at load point
k (it can be either the average load or the actual load during interruption
I), M the number of simulated years, and Nk is the number of interruptions
contributing to load point k in the span of the simulated years.
RELIABILITY COST/WORTH ASSESSMENT 281

J
I

1----r-_1_ _ ~r--_ _ _ _-r-3___ ............-4------1~ ____

5 6 7 8

LP1 LP2 LP3 LP4

Figure 7.6. Example system feeder configuration.

7.5.2. Numerical Example

(a) Example System. Consider the distribution system shown in
Figure 7.6.(20) The reliability data, i.e., failure rates and mean time to repair
for all components are given in Table 7.18. All switching actions require
1 hr. It has been assumed that the alternate supply is always available and
that the fuses in the laterals are fully reliable. The customer data at each of
the four load points are given in Table 7.19. The sector customer damage

Table 7.18. Example Distribution System Component

Reliability Data
Failure rate Mean time to repair
Component (failures/yr) (hr/failure)
1 0.0520 5.00
2 0.0488 5.00
3 0.0488 5.00
4 0.0390 5.00
5 0.0638 8.53
6 0.0540 9.17
7 0.0670 8.36
8 0.0638 8.53
Transfonner 0.0150 20.00

Table 7.19. Example Distribution System Customer

Data
Average Peak
Load load load Customer
point (kW) (kW) type
1 454.0 750.0 Commercial
2 450.0 729.1 Residential
3 566.0 916.7 Governmental
4 566.0 916.7 Governmental
282 CHAPTER 7

functions for the commercial, residential, and governmental customer

categories in the example system are the same as those given in Table 7.1.
The cost in $/kW of durations greater than 8 hr is evaluated using the
assumption that the slope is the same as that between the 4 and 8 hr duration.
Average loads at the load points are used in the study.

(b) Case Studies. The sequential Monte Carlo simulation approach

was used with 10,000 replication years. In order to examine the effects of
various distributions, the component state durations were assumed to be
described by either a normal, log-normal (LN), gamma, or exponential
(EXP) distribution. The annual average outage times and the annual
expected interruption costs of all load points for the different distribution
assumptions are shown in Table 7.20.
It can be seen that the different distribution assumptions provide basi-
cally the same annual average outage time indices. This is confirmed by the
analysis and results given in Chapter 6. Unlike the three basic load point
reliability indices, however, the interruption costs generally depend on the
outage duration distributions. Different distribution assumptions, in princi-
ple, lead to different interruption cost indices. In the given example, however,
Cases (c), (d), (e), and (f) provide basically the same interruption costs,
which are different from those obtained assuming the repair activities to be
exponentially distributed in Cases (a) and (b).
The load point interruption costs for the example distribution system
using an analytical approach are shown in Table 7.21. It can be seen that
the analytical load point interruption costs are quite close to those obtained
in Cases (c), (d), (e), and (f) of the simulation studies. This suggests that
although the average outage duration is used in the analytical method, in
some cases, such as when the component restoration mode distributions are
assumed to be other than exponential with a small standard deviation, the
analytical method provides results which are comparable to those from simu-
lation studies.

7.6. MINIMUM COST ASSESSMENT IN

COMPOSITE SYSTEM
EXPANSION PLANNING
7.6.1. Basic Concepts
The fundamental objective of an electric power utility is to plan and oper-
ate the system to satisfy the load and energy requirements of its customers
RELIABILITY COST/WORTH ASSESSMENT 283

Table 7.20. Load Point Annual Average Outage Times and

Expected Interruption Costs
Load Annual Outage Expected interruption
Case Description point time (hr/yr) costs (k$/yr)
(a) EXPrep.; 1 0.951 5.130
all other 2 0.827 0.895
times are 3 0.956 2.783
EXP 4 0.912 2.752
Total 11.560
(b) EXP rep.; 1 0.920 4.918
all other 2 0.863 0.976
times are 3 0.940 2.676
LN (sd= m/6) 4 0.901 2.680
Total 11.250
(c) all times 0.961 4.299
areLN 2 0.848 0.646
(sd=m/6) 3 0.976 1.571
4 0.913 1.497
Total 8.013
(d) Normal rep. 0.962 4.303
(sd=m/6); 2 0.848 0.646
other times 3 0.976 1.573
LN (sd=m/6) 4 0.913 1.499
Total 8.021
(e) Gamma rep. 1 0.959 4.290
(sd=m/6); 2 0.832 0.643
all other 3 0.983 1.601
times EXP 4 0.906 1.500
Total 8.034
(f) LN rep. 1 0.962 4.277
(sd=m/l0); 2 0.844 0.656
other times 3 0.976 1.561
EXP 4 0.911 1.480
Total 7.974

Table 7.21. Load Point Interruption

Costs Using the Analytical Method
Expected interruption
Load point costs (k$/yr)
1 4.171
2 0.666
3 1.477
4 1.419
284 CHAPTER 7

at as low a cost as possible and with an acceptable level of continuity and

quality of supply. This involves the two aspects of economy and reliability.
System reliability worth can be indirectly incorporated by interruption cost
evaluation and, consequently, the system adequacy problem is converted
into a cost problem. This makes it possible to assess system economy and
reliability on a unified basis. System operating costs are important aspects
of overall system economy. Simulation of the system operating cost therefore
should be incorporated in composite system expansion planning so that an
optimum decision can be made in terms of overall minimum cost.
A potential optimal generation loading schedule (pOGLS) can be easily
obtained when transmission network constraints and component forced out-
ages are not considered. In practical composite system operation, however,
the power allocation between generators will deviate from the POGLS due
to transmission network constraints and system component forced outages.
This leads to an increase in the total operating cost. The addition of genera-
tion or/and transmission facilities can move a practical operating schedule
closer to the POGLS and therefore improve not only reliability but also the
operating economy of a composite system.
An optimal expansion decision achieves the minimum total cost Q:

min Q=I+O+D (7.25)

where I is the investment cost, 0 the operating cost, and D the customer
damage (interruption) cost associated with system unreliability.
The investment cost is a fixed value for a particular addition decision
and can be easily calculated using the present value method. Both operating
and customer damage costs depend on system states and therefore it is
necessary to simulate a considerable number of system contingency states
involving various load levels in order to obtain the annual expected values
of these costs. The minimization linear program model described in Section
7.4.1 can be extended to include system generating cost simulation.

7.6.2. Minimum Cost Assessment Method

(a) Annual Investment Cost. The total cost of an expansion deci-
sion is the sum of investment, operating, and damage costs. This total cost
can be expressed on the basis of one year. The annual investment cost (I)
can be calculated using the present value method:
'(1 + .)n
I=A I I (7.26)
(l +ir-l
RELIABILITY COST/WORTH ASSESSMENT 285

where A is the capital cost of the additional system facility in a particular

expansion decision at the starting year, i is the discount rate which generally
equals the interest rate plus the inflation rate, and n is the economic life of
the added system facility.

(b) Minimization Model for Operating and Damage

Costs. It is necessary to simulate a considerable number of system states
involving various load levels in order to obtain the annual expected operating
and damage costs. The system states can be selected using the Monte Carlo
sampling technique. The following linear programming minimization model
can be used to calculate the power allocation between generators, possible
load curtailments at different buses, operating costs at each generator and
of the system and damage (interruption) costs at each load bus and of the
system, for a particular drawn system state:
NG IG, NC IC,
min TOD = I I DkbijPGij+ I I Wif{Dk)Cif (7.27)
i=i j=i

subject to

(i= 1, ... ,L) (7.28)

NG IG, NC IC, NC IC,

I I PGif+ I I Cif= I I PDif (7.29)
i=i j=i

PGijin~PGij~PGijax (i= 1, ... , NG;j= 1, ... ,IGi ) (7.30)

O~ Cif~PDif (i= 1, ... , NC;j= 1, ... ,ICi ) (7.31)

(n= 1, ... ,L) (7.32)

where PGif is the generation variable of the jth generator at Bus i; PGijin,
PGijax are the lower and upper bounds of PGif ; IGi is the number of genera-
tors at Bus i; bif is the unit generating cost (in $/kWh) of the jth generator
at Bus i, all other notations are the same as those used in the minimization
model given in Section 7.4.1.
This model is an extension of that in Section 7.4.1, in which simulation
of the system operating cost has been included. The objective of the model
is to minimize the sum of the operating and damage costs while satisfying
the power balance, the linearized load flow relationships, the line power
flow limits and output limits of each generating unit. The customer damage
functions for different customer sectors and effects of system state durations
on the interrupted damage cost are incorporated in the model. In order to
286 CHAPTER 7

reduce the scale of the LP problem, generators which are connected to the
same bus and have the same unit generating cost (bij) can be represented by
an equivalent generation variable in the model.
The values of Cij and PGij obtained by solving this model correspond
to a particular system state k whose duration is Dk • As in the previous
model, the subscript k in the variables except Dk in the model has been
omitted for simplicity. The system state duration Dk can be calculated using
the sampling method given in Section 7.4.1. The model provides the annual
Expected Energy Produced by Generators (EEPG) and Expected Genera-
tion Cost (EGC) of the generation buses and the system in addition to
EENS and EIC indices of the load buses and the system. The equations for
calcuating bus and system EENS and EIC indices are the same as those
given in Section 7.4.1, i.e., equations (7.15)-(7.18). The annual bus and
system EEPG and EGC indices are given by the following equations:
N IG,
EEPG;= L L PGijkFkDk (7.33)
k~lj~1

N IG,
EGC;= L L PGijkFkDkbij (7.34)
k~1 j~1

NG
EEPG = L EEPG; (7.35)

NG
EGC= L EGC; (7.36)
;~I

where PGijb Db Fk , bij, and IG; are as defined earlier; EEPG; and EEPG
are the expected energy produced by generators at Bus i and for the system,
respectively (MWh/yr); EGC; and EGC are the expected generation cost at
Bus i and for the system (k$/yr); N is the number of system states. The
system state frequency Fk (occ./yr) is obtained using equation (7.19).

(c) Basic Minimum Cost Assessment Procedure. A computer

program named MICOAS for minimum cost assessment in composite system
expansion planning has been developed at the University of Saskatchewan
based on the above minimization model and Monte Carlo simulation tech-
niques. A brief description of the MICOAS program follows. The flowchart
is shown in Figure 7.7.(21)
1. A multistep annual load model is created which eliminates the chron-
ology and aggregates the load states using hourly load records. The number
of load level steps depends on the sensitivities of the composite system
adequacy indices to load variation (see Section 5.3.2). All load level steps
RELIABILITY COST/WORTH ASSESSMENT 287

Select the ith step load level with startin i=1

Select a system state and determine its duration

by Monte-Carlo techniques

Solve the minimization model

Coefficient of variation is small enough?

yes
All load levels are considered?

Figure 7.7. Flowchart of the minimum cost assessment method.

are considered successively and the results for each load level are weighted
by their probabilities to obtain the annual indices.
2. The system states at a particular load level are selected using the
simulation techniques discussed in Chapter 5. Generating unit and transmis-
sion line states are modeled using two-state random variables. If necessary,
other system considerations, such as generating unit derated states, (22) effects
of regional weather on transmission line outages,(23) and bus load uncertainty
and correlation, (24) etc., can also be incorporated using the methods
presented in Chapter 5.
3. System state durations are modeled using the sampling method given
in Section 7.4.1. System state frequencies are calculated using equation
(7.19).
4. The operating and damage cost minimization model presented earlier
is solved to obtain the generation allocations, generation costs, load curtail-
ments, and interruption costs.
5. Steps (2)-(4) are repeated until convergence for each particular load
level. Annual bus and system EENS, EIC, EEPG, and EGC indices are
288 CHAPTER 7

calculated using equations (7.15) to (7.18) and (7.33) to (7.36) by consider-

ing all load level steps.
6. The annual investment cost is calculated using equation (7.26). The
total cost shown in equation (7.25) is obtained.
In addition to the load bus indices, the generator bus indices, and
the system indices noted above, the MICOAS program also provides two
transmission line indices. These are: (i) probability of the line power at
its limit in no-load-curtailment system states; and (ii) proportion of load
curtailments associated with the line power at its limit in total load curtail-
ments. The line indices together with load bus and generator bus indices
provide initial information regarding what system facilities to add and where
to locate them. System indices provide quantitative evaluation of overall
system economy and reliability and can be used to select the optimal expan-
sion plan.

7.6.3. Case Studies

In order to demonstrate the application of the minimum cost assessment
method in composite system expansion planning, case studies were con-
ducted using the IEEE MRTS given in Section 7.4.2. The basic system data,
system single line diagram, and the customer data (the customer sector
allocation at load buses) are given in Section 7.4.2. The sector customer
damage functions are given in Table 7.1. The additional data required to
perform minimum cost assessment are generating unit operating costs given
in Table 7.8(16) and investment cost related data given in Table 7.22.

(a) Selecting the Optimal Expansion Plan at the Present

Load Level. The IEEE MRTS was considered to be the original system
at the present load level. The line, load bus, and generator bus indices
calculated using the MICOAS program indicate that the lines from Bus 16
to Bus 17 and from Bus IS to Bus 21 have the largest and the next-largest
probabilities of the line powers being at their limits in no-load-curtailment
states, respectively [line index (i)]. The line from Bus 20 to Bus 23 has the
largest proportion of load curtailments associated with the line power at its

Table 7.22. Investment Cost Related Data

Life of facilities 30yr
Discount rate 10%
Capital cost of 230-kV line 90k$/mile
Capital cost of ISS-MW generating unit 100 M$/unit
RELIABILITY COST/WORTH ASSESSMENT 289

limit in total load curtailments [line index (ii)]. The expected generation
costs at generator buses 18 and 21 deviate most significantly from their
bus generation costs in the ponential optimal generation loading schedule
(POGLS). The largest and the next-largest expected energy not supplied
(EENSi ) are at Bus 19 and Bus 20, respectively. The following possible
expansion plans are therefore initially selected:
Ll-l adding a circuit from Bus IS to Bus 21
Ll-2 adding a circuit from Bus 16 to Bus 17
Ll-3 adding a circuit from Bus 17 to Bus 18
Ll-4 adding a circuit from Bus 20 to Bus 23
GI adding a 155-MW generating unit at Bus 16.
The cost and adequacy of each of these five alternate plans were evalua-
ted using the minimum cost assessment method. The basic system indices
are shown in Table 7.23. The annual operating costs are much higher than

Table 7.23. Basic System Indices of the Original System and the
Alternate Plans at the Present Load Level
System index
Alternate
plans AGCP AEGC EGC EENS EIC EOIC AEIC AIOIC
Base system 185.53 189.56 229.55 1902.54 1.73 231.28 0.00 231.28
Lt-I
Index 185.53 185.62 226.37 1901.52 1.73 228.10 0.33 228.42
Diff. 0.00 3.94 3.18 1.02 0.00 3.18 -0.33 2.86
Lt-2
Index 185.53 185.63 226.38 1901.54 1.73 228.11 0.17 228.28
Diff. 0.00 3.93 3.17 1.00 0.00 3.17 -0.17 3.00
Lt-3
Index 185.53 190.56 23Q.48 1903.57 1.73 232.21 0.10 232.31
Diff. 0.00 -1.00 -0.93 -1.03 0.00 -0.93 -0.10 -1.03
Lt-4
Index 185.53 189.53 229.12 694.39 0.56 229.68 0.14 229.82
Diff. 0.00 0.03 0.43 1208.15 1.17 1.60 -0.14 1.46
G1
Index 183.97 188.06 224.81 685.63 0.60 225.41 10.61 236.02
Diff. 1.56 1.50 4.74 1216.91 1.13 5.87 -10.61 -4.74
L2
Index 185.53 185.63 225.96 693.04 0.56 226.52 0.32 226.84
Diff. 0.00 3.93 3.59 1209.50 1.17 4.76 -0.32 4.44
L2G1
Index 183.97 184.45 221.87 386.56 0.31 222.18 10.92 233.10
Diff. 1.56 5.11 7.68 1515.98 1.42 9.10 -10.92 -1.82
290 CHAPTER 7

the annual interruption costs or the annual investment costs due to the
system facility additions and therefore it is necessary to show the differences
between the indices for the original system and those of the alternate plans.
These differences are designated by "Diff." in Table 7.23. The interpretation
of the basic system indices is as follows:
AGCP Annual Generating Cost in the Potential optimal genera-
tion loading schedule (M$/yr). This is the system generat-
ing cost when transmission network constraints and
component outages are not considered and generators are
loaded according to their increasing $/MWh (fixed opera-
tion costs are not included).
AEGC Annual Economic Generating Cost in the normal system
state (M$/yr). This is the system economic generating cost
when transmission network constraints are incorporated
but generator and line outages are not considered. This cost
corresponds to economic dispatch of the composite system
in the normal state (the fixed operation costs are not
included).
EGC Expected Generating Cost in normal and outage states
(M$/yr). This is the annual expected system economic
operating costs when transmission network constraints and
system component outages are considered. This cost corre-
sponds to the first term of the minimization model for oper-
ating and damage costs (the fixed operation costs are
included).
EENS Expected Energy Not Supplied for the composite system
(MWh/yr).
EIC Expected Interruption Cost (M$/yr). This is the annual
expected minimum damage costs when transmission net-
work constraints and system component outages are con-
sidered. This cost corresponds to the second term of the
minimization model for operating and damage costs.
EOIC Expected Operating and Interruption Cost (M$/yr). It
equals the sum of EGC and EIC.
AEIC Annual Expansion Investment Cost (M$/yr).
AIOIC Annual Investment, Operating and Interruption Cost (M$/
yr). It equals the sum of EOIC and AEIC.
It can be seen from Table 7.23 that although the Ll-l and Ll-2 alternate
plans do not improve the system reliability when compared to the original
system (the EENS indices and the expected annual interruption costs remain
unchanged), these two plans greatly improve system operating economy
RELIABILITY COST/WORTH ASSESSMENT 291

(the annual expected generating cost decreases by 3.18 M$ and 3.17 M$,
respectively). This indicates that the addition of a system facility depends
on not only the improvement in system reliability but also on the improve-
ment in system operating economy and, in some cases, system operating
economy can be the dominant factor. It is interesting to note that the differ-
ent additions in the Ll-l and Ll-2 plans have the same effects on system
reliability (no improvement) and basically make the same improvements in
system operating economy. The addition of a circuit from Bus 15 to Bus 21
requires line reinforcement of 34 miles and an annual investment cost of
0.33 M$. The total annual investment, operating, and interruption cost of
the Ll-l plan decreases by 2.86 M$ compared to the original system. The
addition of a circuit from Bus 16 to Bus 17 requires only line reinforcement
of 18 miles and the annual investment cost of 0.17 M$. The total annual
investment, operating, and interruption cost of the plan Ll-2 decrease by
3.00 M$. Plan Ll-2 therefore is superior to Plan Ll-1. The Ll-3 plan does
not improve system reliability (the EENS index and the expected annual
interruption cost remain unchanged) nor system operating economy (the
annual expected generating cost increases by 0.93 M$). This means that the
addition of the 10-mile line from Bus 17 to Bus 18 does nothing other than
increase the investment cost. The Ll-4 plan improves system reliability quite
considerably (the EENS index decreases from 1902 MWh/yr to 694 MWh/
yr and the expected annual interruption cost decreases by 1.17 M$) but has
only a small impact on system operating economy (the expected annual
generating cost decreases only by 0.43 M$). By adding the annual investment
cost of the circuit from Bus 20 to Bus 23 (0.14 M$), the total annual invest-
ment, operating, and interruption cost of the LI-4 plan decreases by 1.46 M$.
The 01 plan improves both system reliability (the EENS index decreases to
685 MWh/yr and the expected annual interruption cost decreases by
1.13 M$) and system operating economy (the expected annual generating
cost decreases by 4.74 M$). The annual investment cost, however, increases
by 10.61 M$ due to the addition of a 155-MW generating unit at Bus 16.
The total annual investment, operating, and interruption cost of the 01 plan
increases 4.74 M$ compared to the original system. In this case, the annual
investment cost is the dominant factor.
The results for the initial five alternate plans suggest that the following
two expansion options should be examined further:
L2 adding a circuit from Bus 16 to Bus 17 and a circuit from Bus
20 to Bus 23
L2Gl adding a 155-MW generating unit at Bus 16 to the L2 plan
The basic system indices for these two new alternatives are also shown
in Table 7.23. The L2 plan considerably improves system reliability and
292 CHAPTER 7

operating economy (the EENS index decreases to 693 MWh/yr and the
expected annual operating and interruption cost decreases by 4.76 M$) for
only a small increase in the annual investment cost (0.32 M$). The total
annual investment, operating, and interruption cost decreases by 4.44 M$.
Although the L2G 1 plan leads to a greater decrease in the expected annual
operating and interruption cost (9.10 M$) than the L2 plan, it requires much
higher annual investment cost (10.92 M$). The total annual investment,
operating, and interruption cost of the L2G 1 plan is higher that of the
original system.
Using the minimum cost assessment method and comparing the
decreases in the total annual costs for the alternate plans, the L2 plan is the
optimal expansion plan for the base system at the present load level.

(b) Selecting an Optimal Expansion Plan to Meet Load

Growth. The objective of power system expansion analysis is the selection
of the optimal (the most economical and reliable) expansion plan for the
system not only at the present load level but also recognizing future load
growth. The results in Section 7.4.4 indicate that when only reliability worth
and investment cost are considered, the addition of a 155-MW generating
unit at Bus 16 is not cost-effective even for 10.0010 load growth. It can be
seen from the previous expansion analysis that the addition of this unit can
improve system operating economy considerably, but that the Gl and L2Gl
plans in which the 155-MW generating unit is added are not economical at
the present load level. It may be reasonable, however, to add such a generat-
ing unit as the demand grows if the total investment, operating, and interrup-
tion cost is considered. The L2, Gl, and L2Gl plans were therefore re-
examined using the minimum cost assessment method assuming load growth
of 4%, 6%, 8%, and 10%, respectively. The basic system indices are shown
in Tables 7.24 to 7.27. For comparison, the calculation results for the original
system have been also listed. The values in the "Diff." rows are the decreased
values of the system indices for the three alternatives compared to those of
the original system.
The results in Tables 7.24 to 7.27 indicate that in the 4% load growth
case, the G 1 plan involves higher total annual cost than for the original
system. The addition of a 155-MW generating unit at Bus 16 is therefore
uneconomical at this load level. When load growth is equal to or greater
than 6%, however, the Gl plan becomes more economical than the original
system. The L2G 1 plan always has lower total annual cost at the various
load growth levels than the original system and the G 1 plan. This is due to
the fact that additions of two circuits from Bus 16 to Bus 17 and from Bus
20 to Bus 23 decrease the operating and interruption costs considerably but
only slightly increase the annual investment cost. At load levels smaller than
RELIABILITY COST/WORTH ASSESSMENT 293

Table 7.24. Basic System Indices of the Original System and the
Alternate Plans for 4% Load Growth
System index
Alternate
plans AGCP AEGC EGC EENS EIC EOIC AEIC AIOIC
Base system 198.39 201.83 245.96 4358.47 3.97 249.93 0.00 249.93
L2
Index 198.39 198.43 242.57 1749.28 1.43 244.00 0.32 244.32
Diff. 0.00 3.40 3.39 2609.19 2.54 5.93 -0.32 5.61
G1
Index 195.37 198.91 238.99 1614.27 1.44 240.43 10.61 251.04
Diff. 3.02 2.92 6.97 2744.20 2.53 9.50 -10.61 -1.11
L2G1
Index 195.37 195.84 236.32 860.74 0.71 237.03 10.92 247.95
Diff. 3.02 5.99 9.64 3497.73 3.26 12.90 -10.92 1.98

Table 7.25. Basic System Indices of the Original System and the
Alternate Plans for 6% Load Growth
System index
Alternate
plans AGCP AEGC EGC EENS EIC EOIC AEIC AIOIC
Base
system 205.40 208.65 254.85 6285.00 5.7.3 260.58 0.00 260.58
L2
Index 205.40 205.49 251.50 2680.03 2.18 253.68 0.32 254.00
Diff. 0.00 3.16 3.35 3604.97 3.55 6.90 -0.32 6.58
G1
Index 201.46 204.74 246.62 2375.62 2.13 248.75 10.61 259.36
Diff. 3.94 3.91 8.23 3909.38 3.60 11.83 -10.61 1.22
L2Gl
Index 201.46 201.94 244.07 1259.18 1.04 245.11 10.92 256.03
Diff. 3.94 6.71 10.78 5025.82 4.69 15.47 -10.92 4.55

8% load growth, the L2 plan decreases the total annual cost more than the
L2G 1 plan. In the case of 8% load growth, the L2 and L2G 1 plans result
in basically the same decrease in the total annual cost. Considering that the
L2G 1 plan leads to much smaller expected energy not supplied (EENS) and
much lower interruption cost (EIC) than the L2 plan, it is necessary for 8%
load growth to add a 155-MW generating unit at Bus 16. When the load
growth is larger than 8%, the L2G 1 plan results in a larger decrease in the
total annual investment, operating, and interruption cost than the L2 plan.
294 CHAPTER 7

Table 7.26. Basic System Indices of the Original System and the
Alternate Plans for 8% Load Growth
System index
Alternate
plans AGCP AEGC EGC EENS EIC EOIC AEIC AIOIC
Base system 213.19 216.18 264.45 8891.05 8.10 272.55 0.00 272.55
L2
Index 213.19 213.24 261.04 4014.22 3.25 264.29 0.32 264.61
Diff. 0.00 2.94 3.41 4876.83 4.85 8.26 -0.32 7.94
Gl
Index 207.85 210.83 254.75 3446.02 3.09 257.84 10.61 268.45
Diff. 5.34 4.35 9.70 5445.03 5.01 14.71 -10.61 4.10
L2Gl
Index 207.85 208.27 252.29 1819.48 1.50 253.79 10.92 264.71
Diff. 5.34 7.91 12.16 7071.57 6.60 18.76 -10.92 7.84

Table 7.27. Basic System Indices of the Original System and the
Alternate Plans for 10% Load Growth
System index
Alternate
plans AGCP AEGC EGC EENS EIC EOIC AEIC AIOIC
Base system 222.08 224.82 274.91 12480.87 11.32 286.23 0.00 286.23
L2
Index 222.08 222.13 271.39 5963.51 4.81 276.20 0.32 276.52
Diff. 0.00 2.69 3.52 6517.36 6.51 10.03 -0.32 9.71
Gl
Index 214.58 217.37 263.34 4976.70 4.46 267.80 10.61 278.41
Diff. 7.50 7.45 11.57 7504.17 6.86 18.43 -10.61 7.82
L2Gl
Index 214.58 215.04 260.98 2644.54 2.17 263.15 10.92 274.07
Diff. 7.50 9.78 13.93 9836.33 9.15 23.08 -10.92 12.16

Based on these results, the final expansion decision should be that before
load growth reaches 8%, two circuits from Bus 16 to Bus 17 and from Bus
20 to Bus 23 are required and, when load growth reaches 8%, a 155-MW
generating unit should be added at Bus 16.

7.7. CONCLUSIONS
This chapter deals with reliability cost/worth assessment using Monte
Carlo simulation. Direct evaluation of reliability worth is very difficult and
perhaps impossible. The customer interruption costs due to unreliability can
RELIABILITY COST/WORTH ASSESSMENT 295

be used as an indirect measure of reliability worth. The procedure illustrated

to assess customer interruption costs at HLl, HL2, and in the distribution
functional zone utilizes Customer Damage Functions (CDF). Interruption
cost data obtained from customer surveys are specific to area, region, and
country. In general, the data for a particular area, or region, or country
cannot be automatically utilized for other areas, regions, and countries. As
with other reliability data, such as component failure rates and repair times,
interruption cost data also should be continuously monitored and updated.
Analytical techniques utilize average outage state durations and basi-
cally provide average or expected values for load points and system indices.
In many reliability worth evaluations, the use of analytical techniques will
provide acceptable results. In the case of customer damage functions
exhibiting a high degree of nonlinearity, however, analytical methods may
create relatively large errors. Monte Carlo methods which sample system
outage state durations are theoretically accurate, and the effects on interrup-
tion costs of outage duration distributions are automatically evaluated.
The basic index for reliability worth assessment is the annual Expected
Interruption Cost (EIC). This index is a combination of the unit interruption
cost, the demand not supplied, the outage duration, and the outage fre-
quency and therefore it contains more information than other reliability
indices. The EIC is a monetary representation of reliability worth. This
makes it possible to compare reliability worth with other economic indices
such as investment and operating costs on a common base. In generation
planning, for example, if the investment cost of a generating unit is smaller
than the decrease in the system EIC, the addition of the unit is beneficial.
This principle also generally applies to composite system and distribution
system planning. In the case of composite systems, however, the addition of
a generation or transmission facility may not only improve system reliability
but also operating economy. Minimum cost assessment associated with the
total investment, operating, and interruption costs is therefore a more realis-
tic method for composite system expansion planning.
Generating system reliability worth assessment involves the calculation
of the annual system EIC. The Interrupted Energy Assessment Rate (lEAR)
is an important factor and concept in generating system reliability worth
assessment. lEAR basically does not vary with the peak load level and
therefore the system EIC can be obtained by multiplying the EENS by the
lEAR. The lEAR can also be calculated at HL2 and in the distribution
functional zone. In the case of composite systems, however, lEAR should
be used with caution. This is due to the fact that lEAR at HL2 may not
be the same for different load curtailment philosophies and for different
configuration changes. Distribution system reliability assessment involves
the calculation of the load point EIC indices. Composite system reliability
296 CHAPTER 7

worth assessment involves both system and bus EIC indices. Complications
in composite system evaluation lie not only in system analysis associated with
load flow calculations, contingency studies, overload alleviation, generation
rescheduling, and load curtailments, but they also result from incorporation
of sector customer damage functions (SCDF) at the load buses. Utilization
of bus composite customer damage functions (CCDF) can create different
results from those obtained using the SCDF in composite system reliability
worth assessment.

7.S. REFERENCES
1. G. Wacker, E. Wojczynski, and R. Billinton, "Cost/Benefit Considerations in Providing an
Adequate Electric Energy Supply," Third International Symposium on Large Engineering
Systems, July 10-11, 1980, St. John's, Newfoundland, pp. 3-8.
2. R. Billinton, G. Wacker, and E. Wojczynski, "Comprehensive Bibliography on Electrical
Service Interruption Costs," IEEE Trans. on PAS, Vol. 102, 1983, pp. 1831-1837.
3. G. Tollefson, R. Billinton, and G. Wacker, "Comprehensive Bibliography on Reliability
Worth and Electrical Service Consumer Interruption Costs: 1980-1990," IEEE Trans. on
Power Systems, Vol. 6, No.4, 1991, pp. 1508-1514.
4. L. V. Skof, "Ontario Hydro Surveys on Power System Reliability: Summary of Customer
Viewpoints," Ontario Hydro Report, R&MR 80-12, EPRI Seminar, October 11-13, 1983.
5. R. Billinton, G. Wacker, and E. Wojczynski, "Customer Damage Resulting from Electric
Service Interruptions," Canadian Electrical Association, R&D Project 907, U131 Report,
1982.
6. E. Wojczynski, R. Billinton, and G. Wacker, "Interruption Cost Methodology and
Results--A Canadian Commercia1 and Small Industrial Survey," IEEE Trans. on PAS,
Vol. 103, 1984, pp.437-444.
7. G. Wacker, E. Wojczynski, and R. Billinton, "Interruption Cost Methodology and
Results--A Canadian Residential Survey," IEEE Trans. on PAS, Vol. 102, 1983, pp. 3385-
3392.
8. Ontario Hydro Survey on Power System Reliability: Viewpoint of Small Industrial Users
(Under 5000 kW), OH Report No. R&U 78-3, April, 1978.
9. R. Billinton, G. Wacker, and R. K. Subramaniam, "Factors Affecting the Development
of a Residential Customer Damage Function," IEEE Trans. on PWRS, Vol. 2, No.1,
1987, pp. 204--209.
10. R. Billinton, G. Wacker, and R. K. Subramaniam, "Factors Affecting the Development
of a Commercial Customer Damage Function," IEEE Trans. on PWRS, Vol. I, No.4,
1986, pp. 28-33.
11. R. K. Subramaniam, R. Billinton, and G. Wacker, "Factors Affecting the Development
of an Industrial Customer Damage Function," IEEE Trans. on PAS, Vol. 104, No. 11,
1985,pp.3209-3215.
12. G. Wacker, P. Kos, and R. Billinton, "Farm Losses Resulting Electric Service Interrup-
tions," CEA Meeting, Toronto, March 1989.
13. R. Billinton, J. Oteng-Adjei, and R. Ghajar, "Comparison of Two Alternate Methods to
Establish an Interrupted Energy Assessment Rate," IEEE Trans. on PWRS, Vol. 2, No.3,
1987, pp. 751-757.
RELIABILITY COST/WORTH ASSESSMENT 297

14. R. Billinton and R. N. Allan, Reliability Evaluation of Power System, Plenum, New York,
1984.
15. L. Gan, "Multi-Area Generation System Adequacy Assessment by Monte Carlo Simula-
tion," M.Sc. Thesis, University of Saskatchewan, 1991.
16. J. Oteng-Adjei, "Cost/Benefit Assessment In Electric Power Systems," Ph.D. Thesis,
University of Saskatchewan, Canada, January 1990.
17. IEEE Committee Report, "IEEE Reliability Test System," IEEE Trans. on PAS, Vol. 98,
1979, pp. 2047-2054.
18. O. Berloldi, L. Salvaderi, and S. Scalcino, "Monte Carlo Approach in Planning Studies-
An Application to IEEE RTS," IEEE Trans. on PS, Vol. 3, No.3, 1988, pp. 1146-1154.
19. EPRI Report, "Transmission System Reliability Methods," EL-2526, Vol. I, July 1982,
Appendix Q.
20. L. Goel and R. Billinton, "Evaluation ofInterrupted Energy Assessment Rates in Distribu-
tion Systems," IEEE Trans. on PD, Vol. 6, No.4, 1991, pp. 1876-1882.
21. W. Li and R. Billinton, "A Minimum Cost Assessment Method for Composite Generation
and Transmission System Expansion Planning," IEEE Trans. on PS, Vol. 8, No.2, 1993,
pp. 628-635.
22. R. Billinton and W. Li, "Consideration of Multi-State Generating Unit Models in Com-
posite System Adequacy Assessment Using Monte Carlo Simulation," Can. Journal of
Electrical and Computer Engineering, Vol. 17, No.1, 1992, pp. 24-28.
23. R. Billinton and W. Li, "A Novel Method for Incorporating Weather Effects in Composite
System Adequacy Evaluation," IEEE Trans. on PS, Vol. 6, No.3, 1991, pp. 1154-1160.
24. W. Li and R. Billinton, "Effects of Bus Load Uncertainty and Correlation in Composite
System Adequacy Evaluation," IEEE Trans. on PS, Vol. 6, No.3, 1991, pp. 1522-1529.
 APPENDIX A
Reliability Test
Systems

A.1. IEEE RELIABILITY TEST SYSTEM

(IEEE RTS)
The IEEE Reliability Test System (RTS) was developed by the Subcom-
mittee on the Application of Probability Methods in the IEEE Power Engin-
eering Society to provide a common test system which could be used for
comparing the results obtained by different methods. The details of the R TS
can be found in Reference 1. Since the IEEE R TS was created, there have
been many new developments in power system reliability evaluation. Addi-
tional data or modifications are therefore required to conduct extended
studies. (2-8) This appendix presents the basic data and the additional data
for the IEEE RTS.

A.1.1. Load Model

The basic annual peak load for the test system is 2850 MW. Table A.I
gives data on weekly peak loads in percentage of the annual peak load. If
week I is taken as January, Table A.1 describes a winter peaking system. If
week 1 is taken as a summer month, a summer peaking system can be
described. Table A.2 gives a daily peak load cycle, in percentage of the
weekly peak. The same weekly peak load cycle is assumed to apply for all
seasons. The data in Tables A.l and A.2 together with the annual peak load
define a daily peak load model of 52 x 7 = 364. days, with Monday as the
first day of the year. Table A.3 gives weekday and weekend hourly load
models for each of three seasons. Combination of Tables A.l, A.2, and
A.3 with the annual peak load defines an hourly load model of 364 x 24 =
8736 hr.
299
300 APPENDIX A

Table A.1. Weekly Peak Load in Percentage of Annual Peak

Week Peak Week Peak Week Peak Week Peak
86.2 14 75.0 27 75.5 40 72.4
2 90.0 15 72.1 28 81.6 41 74.3
3 87.8 16 80.0 29 80.1 42 74.4
4 83.4 17 75.4 30 88.0 43 80.0
5 88.0 18 83.7 31 72.2 44 88.1
6 84.1 19 87.0 32 77.6 45 88.5
7 83.2 20 88.0 33 80.0 46 90.9
8 80.6 21 85.6 34 72.9 47 94.0
9 74.0 22 81.1 35 72.6 48 89.0
10 73.7 23 90.0 36 70.5 49 94.2
11 71.5 24 88.7 37 78.0 50 97.0
12 72.7 25 89.6 38 69.5 51 100.0
13 70.4 26 86.1 39 72.4 52 95.2

Table A.2. Daily Peak Load in

Percent of Weekly Peak
Day Peak load
Monday 93
Tuesday 100
Wednesday 98
Thursday 96
Friday 94
Saturday 77
Sunday 75

A.1.2. Generating System

Table AA shows the generating unit ratings and reliability data. The
SO-MW units are assumed to be hydro units and their capacity and energy
limitations are given in Table A.S. Table A.6 gives operating cost data for
the generating units. Power production data are given in terms of heat rates
at selected output levels, since fuel costs are subject to considerable variation
due to geographical location and other factors. The following fuel costs were
suggested for general use (1979 base):

#6 oil $2.30/MBtu
#2 oil $3.00/MBtu
Coal $1.20/MBtu
Nuclear $O.60/MBtu
RELIABILITY TEST SYSTEMS 301

Table A.3. Hourly Peak Load in Percentage of Daily Peak

Winter week Summer weeks Spring/fall weeks
1-8 & 44--52 18-30 9-17 & 31-43
Hour Wkdy" Wknd Q
Wkdy Wknd Wkdy Wknd
12-1 am 67 78 64 74 63 75
1- 2 63 72 60 70 62 73
2- 3 60 68 58 66 60 69
3- 4 59 66 56 65 58 66
4-5 59 64 56 64 59 65
5- 6 60 65 58 62 65 65
6-7 74 66 64 62 72 68
7- 8 86 70 76 66 85 74
8- 9 95 80 87 81 95 83
9-10 96 88 95 86 99 89
10-11 96 90 99 91 100 92
11-Noon 95 91 100 93 99 94
Noon-l pm 95 90 99 93 93 91
1- 2 95 88 100 92 92 90
2- 3 93 87 100 91 90 90
3- 4 94 87 97 91 88 86
4-5 99 91 96 92 90 85
5- 6 100 100 96 94 92 88
6-7 100 99 93 95 96 92
7- 8 96 97 92 95 98 100
8- 9 91 94 92 100 96 97
9-10 83 92 93 93 90 95
10-11 73 87 87 88 80 90
11-12 63 81 72 80 70 85
~dy=weekday. Wknd=weekend.

Table A.4. Generating Unit Reliability Data

Number Forced Scheduled
Unit size of outage MITF MITR maintenance
(MW) units rate (hr) (hr) (weeks/yr)
12 5 0.02 2940 60 2
20 4 0.10 450 50 2
50 6 0.Ql 1980 20 2
76 4 0.02 1960 40 3
100 3 0.04 1200 50 3
155 4 0.04 960 40 4
197 3 0.05 950 50 4
350 1 0.08 1150 100 5
400 2 0.12 1100 150 6
302 APPENDIX A

Table A.5. Hydro Capacity and Energy

Capacity Energy
availableD distributionb
Quarter (%) (%)
1 100 35
2 100 35
3 90 10
4 90 20
Dl()O% capacity = 50 MW,
blOO% energy =200 GWh.

A.1 .3. Transmission System

The transmission network consists of 24 bus locations connected by 38
lines and transformers, as shown in Figure A.I. The transmission lines are
at two voltages, 138 kV and 230 kV. The 230-kV system is the top part of
Figure A.I, with 230/138 kV tie stations at Buses 11, 12, and 24. The loca-
tions of the generating units are shown in Table A. 7. Table A.8 gives data
on generating unit MVAr capacities. The system has voltage corrective
devices at Bus 14 (synchronous condenser) and Bus 6 (reactor). Table A.9
gives the MVAr capacities of these devices. Bus load data at the time of
system peak are shown in Table A. 10. Transmission line forced outage data
are given in Table A.II. Impedance and rating data for lines and trans-
formers are given in Table A.12. The "B" values in the impedance data are
the total, not the values in one leg of an equivalent circuit.
Outages on station components which are not switched as a part of a
line are not included in the outage data in Table A.II. The following data
are provided for bus sections:

138 kV 230kV
Faults per bus section/yr 0.027 0.021
Percentage of permanent faults 42 43
Outage duration for permanent faults (hr) 19 13

The following statistics are provided for circuit breakers:

Physical failures/breaker/yr 0.0066

Breaker operational failure per breaker/yr 0.0031
Outage duration (hr) 72
RELIABILITY TEST SYSTEMS 303

Table A.6. Generating Unit Operating Cost Data

O&MCost
Unit Heat
size Output rate Fixed Variable
(MW) Type Fuel (%) (BTu/kWh) ($/kW/yr) ($/MWh)
12 Fossil #6 oil 20 15600 10.0 0.90
steam 50 12900
80 11900
100 12000
20 Combus. #2 oil 80 15000 0.3 5.00
turbine 100 14500
50 Hydro
76 Fossil Coal 20 13000 10.5 0.90
steam 50 12900
80 11900
100 12000
100 Fossil #6 oil 25 13000 8.5 0.80
steam 55 10600
80 10100
100 10000
155 Fossil Coal 35 11200 7.0 0.80
steam 60 10100
80 9800
100 9700
197 Fossil #6 oil 35 10750 5.0 0.70
steam 60 9850
80 9840
100 9600
350 Fossil Coal 40 10200 4.5 0.70
steam 65 9600
80 9500
100 9500
400 Nuclear LWR 25 12550 5.0 0.30
steam 50 10825
80 10170
100 10000

A physical failure is a mandatory unscheduled removal from service for

repair or replacement. An operational failure is a failure to clear a fault
within the breaker's normal protection zone.
There are several lines which are assumed to be on a common right-of-
way or common tower for at least a part of their length. These line pairs
are indicated in Figure A.i by circles around the line pair, and an associated
letter identification. Table A.13 gives the actual length of common right-of-
way or common tower facility.
304 APPENDIX A

Table A.7. Generating Unit Locations

Unit 1 Unit 2 Unit 3 Unit 4 Unit 5 Unit 6
Bus (MW) (MW) (MW) (MW) (MW) (MW)
20 20 76 76
2 20 20 76 76
7 100 100 100
l3 197 197 197
15 12 12 12 12 12 155
16 155
18 400
21 400
22 50 50 50 50 50 50
23 155 155 350

Table A.S. Generating Unit

MVAr Capacities
MVAr
Size
(MW) Minimum Maximum
12 o 6
20 o 10
50 -lO 16
76 -25 30
100 o 60
155 -50 80
197 o 80
350 -25 150
400 -50 200

Table A.9. Voltage Correction

Devices
Device Bus MVAr capacity
Synchronous 14 50 Reactive
Condenser 200 Capacitive
Reactor 6 100 Reactive
RELIABILITY TEST SYSTEMS 305

230 kV

BUSS
138 kV

Figure A.1. IEEE Reliability Test System.

A.1.4. Additional Data

The extended studies using the IEEE RTS require additional data. The
following additional data are given in the text.
1. Table 4.2 gives derated state data for the 400-MW and 350-MW
generating units.
2. Table 4.4 gives data on additional 25-MW gas turbine units.
3. Data associated with nonexponential distribution repair times of
generating units are shown in Section 4.2.4(d).
4. Data associated with bus load uncertainty and correlation are shown
in Section 5.7.3.
306 APPENDIX A

Table A.10. Bus Load Data

Load
Bus MW MVAr
1 108 22
2 97 20
3 180 37
4 74 15
5 71 14
6 136 28
7 125 25
8 171 35
9 175 36
10 195 40
13 265 54
14 194 39
15 317 64
16 100 20
18 333 68
19 181 37
20 128 26
Total 2850 580

5. Table 7.1 gives sector customer damage functions for seven customer
categories. Table 7.12 gives customer sector allocations (in %) at the
load buses. The bus peak load MW values in Table 7.12 are for the
Modified IEEE R TS. These values are 125% of the bus peak loads
in the original R TS. The percentage loads for the customer sectors
remain unchanged.
6. Table 7.8 gives modified generating unit operating cost data. Table
7.22 gives investment-cost-related data.
7. The IEEE R TS has an oversized transmission network. Several
modifications have been suggested for the purpose of transmission
planning studies. The Modified IEEE R TS is described in Section
7.4.2.

A.2. ROY BILLINTON TEST SYSTEM (RBTS)

The RBTS is a six-bus composite system developed at the University
of Saskatechewan for educational purpose. It is sufficiently small to permit
the conduct of a large number of reliability studies with reasonable solution
time but sufficiently detailed to reflect the actual complexities involved in
practical reliability analysis and can be used to examine a newly developed
technique or method. The details of the RBTS are given in Billinton et al. (9)
RELIABILITY TEST SYSTEMS 307

Table A.11. Transmission Line Length and Forced Outage Data

Permanent
Transient
From To Length Outage rate Outage duration outage rate
bus bus (miles) (occjyr) (hr) (occjyr)
2 3 0.24 16 0.0
3 55 0.51 10 2.9
5 22 0.33 10 1.2
2 4 33 0.39 10 1.7
2 6 50 0.48 10 2.6
3 9 31 0.38 10 1.6
3 24 0 0.02 768 0.0
4 9 27 0.36 10 1.4
5 10 23 0.34 10 1.2
6 10 16 0.33 35 0.0
7 8 16 0.30 10 0.8
8 9 43 0.44 10 2.3
8 10 43 0.44 10 2.3
9 11 0 0.02 768 0.0
9 12 0 0.02 768 0.0
10 11 0 0.02 768 0.0
10 12 0 0.02 768 0.0
II 13 33 0.40 11 0.8
II 14 29 0.39 II 0.7
12 13 33 0.40 II 0.8
12 23 67 0.52 11 1.6
13 23 60 0.49 II L5
14 16 27 0.38 II 0.7
15 16 12 0.33 11 0.3
15 21 34 0.41 11 0.8
15 21 34 0.41 II 0.8
15 24 36 0.41 11 0.9
16 17 18 0.35 11 0.4
16 19 16 0.34 11 0.4
17 18 10 0.32 11 0.2
17 22 73 0.54 11 1.8
18 21 18 0.35 11 0.4
18 21 18 0.35 11 0.4
19 20 27.5 0.38 II 0.7
19 20 27.5 0.38 11 0.7
20 23 15 0.34 11 0.4
20 23 15 0.34 II 0.4
21 22 47 0.45 11 1.2
308 APPENDIX A

Table A.12. Impedance and Rating Data

Impedance Rating (MVA)
p.u. (100 MVA base)
From To Short Long
bus bus R X B Normal term term Equipment
2 0.0026 0.0139 0.4611 175 200 193 138-kV cable
3 0.0546 0.2112 0.0572 175 220 208 138-kV line
5 0.0218 0.0845 0.0229 175 220 208 138-kV line
2 4 0.0328 0.1267 0.0343 175 220 208 138-kV line
2 6 0.0497 0.1920 0.0520 175 220 208 138-kV line
3 9 0.0308 0.1190 0.0322 175 220 208 138-kV line
3 24 0.0023 0.0839 400 600 510 Transformer
4 9 0.0268 0.1037 0.0281 175 220 208 138-kV line
5 10 0.0228 0.0883 0.0239 175 220 208 138-kV line
6 10 0.0139 0.0605 2.4590 175 200 193 138-kV cable
7 8 0.0159 0.0614 0.0166 175 220 208 138-kV line
8 9 0.0427 0.1651 0.0447 175 220 208 138-kV line
8 10 0.0427 0.1651 0.0447 175 220 208 138-kV line
9 11 0.0023 0.0839 400 600 510 Transformer
9 12 0.0023 0.0839 400 600 510 Transformer
10 11 0.0023 0.0839 400 600 510 Transformer
10 12 0.0023 0.0839 400 600 510 Transformer
11 13 0.0061 0.0476 0.0999 500 625 600 230-kV line
11 14 0.0054 0.0418 0.0879 500 625 600 230-kV line
12 13 0.0061 0.0476 0.0999 500 525 600 230-kV line
12 23 0.0124 0.0966 0.2030 500 525 600 230-kV line
13 23 0.01ll 0.0865 0.1818 500 525 600 230-kV line
14 26 0.0050 0.0389 0.0818 500 625 600 230-kV line
15 16 0.0022 0.0173 0.0364 500 625 600 230-kV line
15 21 0.0063 0.0490 0.1030 500 525 600 230-kV line
15 21 0.0063 0.0490 0.1030 500 525 600 230-kV line
15 24 0.0067 0.0519 0.1091 500 525 600 230-kV line
16 17 0.0033 0.0259 0.0545 500 525 600 230-kV line
16 19 0.0030 0.0231 0.0485 500 525 600 230-kV line
17 18 0.0018 0.0144 0.0303 500 525 600 230-kV line
17 22 0.0135 0.1053 0.2212 500 625 600 230-kV line
18 21 0.0033 0.0259 0.0545 500 625 600 230-kV line
18 21 0.0033 0.0259 0.0545 500 625 600 230-kV line
19 20 0.0051 0.0396 0.0833 SOO 525 600 230-kV line
19 20 0.0051 0.0396 0.0833 500 525 600 230-kV line
20 23 0.0028 0.0216 0.0455 500 525 600 230-kV line
20 23 0.0028 0.0216 0.0455 500 525 600 230-kV line
21 22 0.0087 0.0678 0.1424 500 525 600 230-kV line
RELIABILITY TEST SYSTEMS 309

Table A.13. Circuits on Common Right-of-Way or Common Structure

Right-or-way Common ROW Common structure
identification From bus To bus (miles) (miles)
A 22 21 45.0
22 17 45.0
B 23 20 15.0
23 20 15.0
C 21 18 18.0
21 18 18.0
0 15 21 34.0
15 21 34.0
E 13 11 33.0
13 12 33.0
F 8 10 43.0
8 9 43.0
G 20 19 27.5
20 19 27.5

A.2.1. Brief Description of the RBTS

The single line diagram of the test system is shown in Figure A.2. The
system has two generator buses, four load buses, nine transmission lines,
and eleven generating units. The system voltage level is 230 kV and the
voltage limits for system buses are assumed to be between 1.05 p.u. and
0.97 p.u. The system peak load is 185 MWand the total installed generating
capacity is 240 MW. The transmission network shown in Figure A.2 has
been drawn to give a geographical representation. The line lengths are shown
in proportion to their actual lengths. Customer sectors at the load buses are
given in Figure A.2.

A.2.2. Load Model

The annual peak load for the system is 185 MW. The data on weekly
peak loads in percentage of the annual peak load, daily peak load in percent-
age of the weekly peak, and hourly peak load in percentage of the daily
peak are the same as those given in Tables A.I, A.2, and A.3.

A.2.3. Generating System

The generating unit ratings and reliability data for the RBTS are shown
in Table A.14. The two 4O-MW thermal units have been given an optional
three-state representation. The derated model is shown in Figure A.3. It
310 APPENDIX A

All load points have residential

and commercial customers

2x40MW
1x20MW G
1x10MW

BUS1 ....~-

L1 L6
L4
BUS4
Industrial
85MW 40MW
L5 La

20MW
L9

Figure A.2. Single line diagram of the RBTS.

has been assumed that there are no transitions between the derated state
and the down state. The state probabilities and transition rates of the
derated model are such that the derating-adjusted two-state model data are
identical to those given in Table A.l4. The two-state model is also shown
in Figure A.3.
The generating unit cost data are shown in Table A.15. The operating
costs include materials, supplies, manpower, etc. The fuel costs are those
directly associated with energy production. The fuel cost for a hydro unit
includes water rental charges. The fixed costs include the annual charges
which continue as long as capital is tied up in the enterprise and whether
or not the equipment is operating. The capital cost is the total cost to install
RELIABILITY TEST SYSTEMS 311

(40MW)
UP
~p=0.97

.6 J1 .. 194

Derating-Adjusted (OMW)
(OMW) Forced Outage Rate
DOWN (OAFOR) • 0.030
DOWN
PON '" 0.03 POE -0.02 PON =0.02

FOR-O.03

Figure A.3. Two- and three-state models for a 4O-MW thermal generating unit.

Table A.14. Generating Unit Rating and Reliability Data

Unit No. Forced Failure Repair Scheduled
size of outage rate rate maintenance
(MW) Type units rate (I/yr) (I/yr) (weeks/yr)
5 Hydro 2 0.010 2.0 198.0 2
10 Thermal I 0.020 4.0 196.0 2
20 Hydro 4 0.015 2.4 157.6 2
20 Thermal 0.025 5.0 195.0 2
40 Hydro 1 0.020 3.0 147.0 2
40 Thermal 2 0.030 6.0 194.0 2

Table A.15. Generating Unit Cost Data

Loading Variable cost
Unit No. Fixed Capital
order ($/MWh)
size of cost cost
(MW) Type units 1st 2nd Fuel Operat. Total (k$/yr) (M$)
40 (Hydro) 1 1 0.45 0.05 0.50 100.0 160.0
20 (Hydro) 2 2-3 2-3 0.45 0.05 0.50 50.0 80.0
40 (Thermal) 2 8-9 4--5 9.50 2.50 12.00 790.0 80.0
20 (Thermal) 10 6 9.75 2.50 12.25 680.0 60.0
10 (Thermal) 1 11 7 10.00 2.50 12.50 600.0 40.0
20 (Hydro) 2 4--5 8-9 0.45 0.05 0.50 50.0 80.0
5 (Hydro) 2 6-7 10-11 0.45 0.05 0.50 12.5 40.0
312 APPENDIX A

Table A.16. Generation, Outage, and Cost Data for Additional Gas
Turbines
Fuel Operating Fixed Capital
Capacity MlTF MlTR cost cost cost cost
(MW) FOR (br) (hr) ($/MWh) ($/MWh) (k$/yr) (M$)
10 0.12 550 75 52.0 4.5 40.0 5.0

a generating unit. Two loading orders are given in Table A.l5. The first
loading order is on a purely economic basis. The second loading order
allocates some hydro units as peaking units which could reflect limited energy
considerations. Either of the loading orders can be selected depending upon
the operating philosophy in conducting reliability studies. Additional gas
turbine units can be added to the RBTS. The generation, outage, and cost
data for these units are given in Table A16.

A.2.4. Transmission System

The transmission network consists of six buses and nine transmission
lines. The locations of the generating units are shown in Table A.l7. Table
Al8 gives data on generating unit MVAr capacity. Bus load data at the
time of system peak in MW and in percentage of the total system load are
shown in Table A19. At 0.98 power factor, the reactive load MVAr

Table A.17. Generating Unit locations

Bus Unit I Unit 2 Unit 3 Unit 4 Unit 5 Unit 6 Unit 7
No. (MW) (MW) (MW) (MW) (MW) (MW) (MW)
(Thermal plant)
40 40 10 20
2 (Hydro plant)
5 5 40 20 20 20 20

Table A.18. Generating Unit MVAr

Capacities
MVAr
Unit size
(MW) Minimum Maximum
5 o 5
10 o 7
20 -7 12
40 -15 17
RELIABILITY TEST SYSTEMS 313

Table A.19. Bus Load Data

Bus load in %
Bus Load (MW) of system load
2 20.0 10.81
3 85.0 45.95
4 40.0 21.62
5 20.0 10.81
6 20.0 10.81
Total 185.0 100.00

requirements at each bus is 20% of the corresponding MW load. The load

forecast uncertainty can be assumed to follow a normal distribution having
a standard deviation from 2.5% to 10010.
Table A.20 shows the transmission line lengths and outage data. The
permanent outage rate of a given line is obtained using a value of
0.02 outages/yr/km. Line transient outage rates are calculated using a value
of 0.05 outages/yr/km. The outage duration of a transient outage is assumed
to be less than I min and is, therefore, not included in Table A.20. Outages
of substation components which are not switched as a part of a line are not
included in the outage data given in Table A.20. Two pairs of transmission
lines are assumed to be on a common right-of-way or common tower for
their entire length. The common mode data for these two line pairs are given
in Table A.21. The line impedance and rating data are given in Tables A.22.

A.2.S. Station Data

The station configurations for the load and generator buses are given
in the extended single line diagram shown in Figure A.4.

Table A.20. Transmission Line Length and Outage Data

Permanent
Transient
Length Outage rate Duration outage rate
Line From To (kM) (occjyr) (hr) (occjyr)
1 3 75 1.5 10.0 3.75
2 2 4 250 5.0 10.0 12.50
3 1 2 200 4.0 10.0 10.00
4 3 4 50 1.0 10.0 2.50
5 3 5 50 1.0 10.0 2.50
6 1 3 75 1.5 10.0 3.75
7 2 4 250 5.0 10.0 12.50
8 4 5 50 1.0 10.0 2.50
9 5 6 50 1.0 10.0 2.50
314 APPENDIX A

Table A.21. Common Mode Data for the Circuits on

a Common Right-of-Way or a Common Tower
Common Outage Outage
Buses
Line Line length rate duration
pair From To No. (kM) (occ/yr) (hr)
A 1 3 1&6 75 0.15 16.0
B 2 4 2&7 250 0.50 16.0

Table A.22. Line Impedance and Rating Data

(100-MVA Base and 230-kV Base)
Buses Impedance (p.u.)
Current
Line From To R X B/2 rating (p.u.)
1,6 1 3 0.0342 0.180 0.0106 0.85
2, 7 2 4 0.1140 0.600 0.0352 0.71
3 1 2 0.0912 0.480 0.0282 0.71
4 3 4 0.0228 0.120 0.0071 0.71
5 3 5 0.0028 0.120 0.0071 0.71
8 4 5 0.0228 0.120 0.0071 0.71
9 5 6 0.0028 0.120 0.0071 0.71

The station equipment data are as follows:

Circuit Breaker
Active failure rate = 0.0066 failure/yr
Passive failure rate = 0.0005 failure/yr
Average outage duration = 72 hr
Maintenance outage rate = 0.2 outages/yr
Maintenance time = 108 hr
Switching time = 1 hr

Bus Section
Failure rate = 0.22 failures/yr
Outage duration = 10 hr

Station Transformer
Failure rate = 0.02 failures/yr
Outage duration = 768 hr
Maintenance outage rate = 0.2 outages/yr
Maintenance time = 72 hr
Switching time = 1 hr
RELIABILITY TEST SYSTEMS 315

G1(40MW) G5 (5MW) G6 (5MW)

Generating (G) (G)

r:a
Station 1 , . L . ,.L. _ @ Generating
(Thermal) "1"" J ) 1 .- Station 2
,'-o-'l. '-0-'...,." )-'-0-'1:'-0-'1 (Hydro)

~}@
~
G3 ~~
~n<;;~~~
E'-G>'- 20MW
{~>@
(10MW)
:r:-'-o-'~ (20MW) 1 {G8 (20MW)
G9 (20MW) [S>@ G10(20MW)
G11(20MW)

'r , '-o-'i
6@ ~
G2(40MW) 1 2 @ 7

r-----11--4-- G7 (40MW)

Load
~'<>':r,<>,1~
~-r'-O-"
Load
BUS 4

f
BUS 3

~-r~
~ 40MW
I

'<Yr'
20MW
Load
BUS 5

r-.L""'l. Load
~T~ BUS 6

t
20MW

Figure A.4. Extended single line diagram of the RBTS.

A.2.6. Reliability Worth Assessment Data

The sector customer damage function data are the same as those given
in Table 7.1. Customer sector allocations at buses are shown in Table A.23.
The system customer mix by energy consumption and peak demand and
system composite customer damage function for the customer mix are the
same as those given in Tables 7.2 and 7.3, respectively. Bus composite cus-
tomer damage functions can be calculated using the data given in Tables
7.1 and A.23.
316 APPENDIX A

Table A.23. Customer Sector Allocations at Load Buses

Load percentage of customer sector (in %)
Peak
Bus load Large
No. (MW) AgrL user Resid. Gover. Indus. Commer. Office
2 20 0.0 0.0 50.95 22.20 12.95 13.90 0.0
3 85 0.0 65.29 23.16 0.0 4.58 4.35 2.62
4 40 0.0 0.0 37.12 0.0 42.08 20.80 0.0
5 20 0.0 0.0 50.05 33.30 0.0 9.25 7.4
6 20 37.0 0.0 40.8 0.0 12.95 9.25 0.0

A.3. REFERENCES
l. IEEE Committee Report, "IEEE Reliability Test System," IEEE Trans. on PAS, Vol. 98,
1979, pp. 2047-2054.
2. R. N. Allan, R. Billinton, and N. M. Abdel-Gaead, "The IEEE Reliability Test System-
Extension to and Evaluation of the Generating System," IEEE Trans. on PWR, Vol. I,
No.4, 1986, pp. 1-7.
3. R. Billinton, P. K. Vohra, and S. Kumar, "Effects of Station Originated Outages in a
Composite System Adequacy Evaluation of the IEEE Reliability Test System," IEEE Trans.
on PAS, Vol. 104, No. 10, 1985, pp. 2649-2656.
4. W. Li and R. Billinton, "Effects of Bus Load Uncertainty and Correlation in Composite
System Adequacy Evaluation," IEEE Trans. on PS, Vol. 6, No.3, 1991, pp. 1522-1529.
5. J. Oteng-Adjei, "Cost/Benefit Assessment in Electric Power Systems," Ph.D. Thesis, Univer-
sity of Saskatchewan, Canada, January 1990.
6. W. Li and R. Billinton, "A Minimum Cost Assessment Method for Composite Generation
and Transmission System Expansion Planning," IEEE Trans. on PS, Vol. 8, No.2, 1993,
pp. 628-635.
7. O. Bertloldi, L. Salvaderi, and S. Scalcino, "Monte Carlo Approach in Planning Studies---
An Application to IEEE RTS," IEEE Trans. on PS, Vol. 3, No.3, 1988, pp. 1146-1154.
8. EPRI Report, "Transmission System Reliability Methods," EL-2526, Vol. I, July 1982,
Appendix Q.
9. R. Billinton, S. Kumar, N. Chowdhury, K. Chu, K. Debnath, L. Goel, E. Khan, P. Kos,
G. Nourbakhsh, and J. Oteng-Adjei, "A Reliability Test System for Educational Purposes:
Basic Data," IEEE Trans. on PS, Vol. 4, No.3, 1989, pp. 1238-1244.
 APPENDIX B
Elements of
Probability
and Statistics

B.1. PROBABILITY CONCEPT AND

CALCULATION RULES

B.1.1. Probability Concept

A random event can be considered as a phenomenon which mayor may
not occur in a given trial, time, or space. Assume that an experiment is
repeated N times under the same conditions and that the number of occur-
rences of event A is M. The ratio M / N approaches a determined value when
N becomes very large. This value is defined as the probability of event A
occurring, i.e.,

P(A) = lim M (B.t)

N-+oo N

Probability is a numerical measure of likelihood of a random event occurring

and its value lies between 0.0 and 1.0.

B.1.2. Probability Calculation Rules

Some important and basic rules of probability theory are as follows:
P(A u B) = P(A) + P(B) - P(A (\ B) (B.2)
P(A (\ B) = P(A)P(B/A) (B.3)
317
318 APPENDIX B

where P(A u B) is the probability of the occurrence of either A or B or

both, P(A n B) is the probability of the simultaneous occurrence of events
A and B, and P(B/ A) is the conditional probability of B occurring given
that A has occurred. If events A and B are mutually exclusive, P(A n B) =
O. This can be generalized to N events. If N events are mutually exclusive,
the following probability equation applies:
P(A\ u A2 U ... U AN) = P(A\) + P(A2 ) + ... + P(AN) (B.4)
If events A and B are independent, P(A n B) = P(A)P(B/A) = P(A)P(B).
This can also be generalized to N events. If N events are independent of
each other, the following probability equation holds:
P(A\ n A2 n ... nAN) = P(A\)P(A2 ) ••• P(AN) (B.5)
If the events {B\, B2 , ••• , BN} represent a full and mutually exclusive
set, i.e., P(B\) + P(B 2) + ... + P(BN) = 1.0 and P(Bjn Bj ) =0.0 (i=/::j; i,j=
1,2, ... , N), then for any event A,
N
P(A) = I P(Bj)P(A/Bj) (B.6)
j=\

B.2. PROBABILITY DISTRIBUTIONS OF

RANDOM VARIABLES

B.2.1. Probability Distribution Function and

Density -Function
A random event or phenomenon can be represented by a random vari-
able. Given a continuous random variable X, the probability of X being not
larger than a real number x is a function of x. This function is defined as
the cumulative distribution function F(x) of random variable X, i.e.,
F(x)=P(X5.x) (-oo<x<oo) (B.7)
The cumulative distribution function indicates the probabilities of all pos-
sible values of X. Function F(x) can be expressed in the following form:

F(x) = fX f(x) dx (B.8)

-00

where f(x) is the probability density function, and

f(x)=dF(x) (B.9)
dx
ELEMENTS OF PROBABILITY AND STATISTICS 319

The probability of X lying between a and b can be calculated by

P(a~X~b)= fbf(X)dX (B.IO)

a

When a = b, the integration value given by equation (B.IO) is zero. This

indicates that the probability of a continuous random variable being equal
to a single point value is zero.
Assume that the random variable Y is a function of a random variable
X, i.e., y= y(x). The following relationship exists between the probability
density functions of Y and X:

f(y)=f(x) 1:1 (B.ll)

The concepts and the relationship shown in equation (B.II) can be

generalized to the case of multiple random variables (random vector). Con-
sider a two-dimensional random vector and assume that the random vector
~ = (X, Y) has a density functionf(x, y) and the random vector 11 = (Zl' Z2)
is a function of~, i.e., ZI=ZI(X,y) and Z2=Z2(X,y). The inverse functions
can be expressed as X=X(ZI' Z2) and y= Y(ZI' Z2). The density function of
the random vector 11 is given by
f(ZI, Z2) = f[x(z\, Z2), y(z\, z2)]IJI (B.l2)
where J is the Jacobian determinant of the inverse functions:

ax ax
J= OZ\ OZ2 (B.13)
oy oy
OZ\ OZ2

When a random variable is discrete, its probability density function can

be expressed as
(k= I, 2, ... ) (B.14)
and its cumulative probability distribution function as
F(Xk) = P(X ~ Xk) (B.lS)
The relationship between the density function and the cumulative distribu-
tion function of a discrete random variable is described by
F(Xk)=Ipi (B.16)

and
(B.17)
320 APPENDIX B

8.2.2. Important Distributions in Reliability

Evaluation
The following five distributions are used in the text:

I. Exponential Distribution

The density function:

f(t) =A exp(-At) (t~O) (B. IS)

The cumulative distribution function:

F(t) = I-exp(-At) (B.19)

The mean and variance of the exponential distribution are 1/A and
1/A2, respectively.

2. Normal Distribution

The density function:

f(t) = _1_ exp[- (t - II )2J (-00 ~t~ (0) (B.20)

cr.J2ii 2cr 2

where II and u 2 are the mean and variance of the normal distribution.

3. Log-Normal Distribution

The density function:

[(Int - ---.:....-
f( t) = -I- exp - -'--------::' II )2J (t>O) (B.21)
tuJ2ii 2u2
It should be noted that II and u 2 in equation (B.21) are not the mean
and variance of the log-normal distribution. The mean and variance of the
log-normal distribution are given, respectively, by

E(t) = exp(1I + ~2) (B.22)

and

V(t) = exp(211 + ~)[exp( ( 2) -1] (B.23)

ELEMENTS OF PROBABILITY AND STATISTICS 321

If the mean and variance of the log-normal distribution (Le., E and V) are
specified, the parameters}J. and a 2 in equation (B.21) are given by

}J. = InL v+E:2) 1/2] (B.24)

and

(B.25)

4. Gamma Distribution
The density function:

f(t)= rP- 1
p exp[-~
]
(t~O, P>O, a>O) (B.26)
a r(p) a
where a and p are the scale and shape parameters of the gamma distribution.
There are no explicit analytical expressions for the cumulative distribu-
tion functions for normal, log-normal, and gamma distributions.
5. Weibull Distribution
The density function:

( 00 > t ~ 0, P> 0, a > 0) (B.27)

The cumulative distribution function:

F(t)=l-exp[ -(;rJ (B.28)

where a and p are the scale and shape parameters of the Weibull
distribution.

B.3. NUMERICAL CHARACTERISTICS OF

RANDOM VARIABLES
Random variables can be generally described by one or more param-
eters rather than as a specific distribution. These parameters are known
as numerical characteristics. The most useful numerical characteristics are
mathematical expectation (mean), variance, covariance, and correlation
functions.
322 APPENDIX B

B.3.1. Expectation and Variance

If a random variable X has a probability density functionf(x) and the
random variable Y is a function of X, i.e., y =y(x), then

E(Y) = f<Xl y(x)f(x) dx (B.29)

-<Xl
where E( Y) is called the expected or mean value of the random variable Y.
The expected value of the random variable X is

E(X) = f<Xl xf(x) dx (B.30)

-<Xl

t:
where E(X) indicates the mean of all possible values of X.
The variance of X is the expected value of the function [x - E(x)]2, i.e.,

E([X - E(X)]2) = [x- E(X)]2f(x) dx (B.31)

The variance indicates the degree of dispersion of the possible values of X

from its mean. The variance is often expressed by the notation V(X) or
u 2 (X). The square root of the variance, u(X), is known as the standard
deviation.
In the case of a discrete random variable, equations (B.29), (B.30), and
(B.31) become equations (B.32), (B.33), and (B.34):
n
E( Y) = L y(Xi)Pi (B.32)
i=1
n
E(X) = L XiPi (B.33)
i=1
n
V(X) = L [Xi- E(X)]2pi (B.34)
i=1

B.3.2. Covariance and Correlation Function

Given an N-dimension random vector 11 = (Xh X 2, ... ,XN ), thecovari-
ance between any two elements Xi and Xj is defined as
Cij=E{[Xi- E(Xi)][Xj - E(Xj )]}
(B.35)
ELEMENTS OF PROBABILITY AND STATISTICS 323

The covariance is often expressed by the notation cov(Xi , Xj)' The covari-
ance between an element and itself is its variance, i.e.,
cov(Xi , Xi) = V(Xi ) (B.36)
The covariances of all elements of random vector 1) form the covariance
matrix:
CII Cl2 C!N
C2! C22 C2N
(Cij) = (B.37)

CN! CN2 CNN

This is a symmetrical matrix whose diagonal components are variances of

each element of 1).
The correlation function (coefficient) of Xi and Xj is defined as
cov(Xi , Xj)
(B.38)
Pij= .jV(Xi ) .jV(Xj )

The correlation functions of all elements of random vector 1) form the corre-
lation matrix:

PII Pl2 PIN

P2! P22 P2N
(Pij) = (B.39)

PNl PN2 PNN

The absolute value of Pij is smaller or equal to 1.0. When Pij= 0, Xi and Xj
are not correlated; when pij>O, Xi and Xj are positively correlated; when
Pij < 0, Xi and Xj are negatively correlated.

B.4. LIMIT THEOREMS

The limit theorems are the theoretical bases of Monte Carlo simulation.

B.4.1. Law of Large Numbers

The following is one of several different representations of the law of
large numbers:
324 APPENDIX B

If N independent random variables X), X 2 , ••• ,XN follow the same

distribution and E(Xi ) = p, then for a sufficiently small positive number e

lim
N--+oo
p[l~N I xi- p l<e]=1.o
i=1
(B.40)

The law of large numbers indicates that when N is very large, the arithmetic
mean of a group of random variables approaches its expectation with a very
large probability.

B.4.2. Central Limit Theorem

If N independent random variables XI, X 2 , ••• ,XN follow the same
distribution and E(Xi ) = p, V(Xi ) = 0"2, then

·
11m p[II/NI~1 X i - pi <x]_
- -I- fX e-t2/2 dt (B.4I)
N--+oo 0" / IN Jfi -00

This theorem indicates that when N is sufficiently large, the arithmetic mean
approximately follows a normal distribution. If the effect of each variable
Xi is "uniformly small," the central limit theorem still holds even if these
random variables do not follow the same distribution.

B.S. PARAMETER ESTIMATION

Mathematically, the calculation of a reliability index using the Monte
Carlo method is a parameter estimation problem.

B.S.1. Basic Definitions

1. Population (X): set of all possible observed outcomes.
2. Samples (X I ,X2 , ••• ,XN ): a subset of the population.
3. Sample size (N): the number of samples in the subset.
4. Statistic: a function of the sample not containing any unknown
parameter.
5. Sampling distribution: distribution of a statistic.
ELEMENTS OF PROBABILITY AND STATISTICS 325

6. Estimate: Assume that a population has density function

f(x, 0" ... , OM) where 0" ... , OM are unknown parameters, and
that the statistic of the samples (}i= (}i(X" ... ,XN ) can be used to
estimate the parameter (} i of the population; (} i is called the estimate
of the parameter (}i.

8.5.2. Sample Mean and Sample Variance

Quantities X" X 2 , ••• ,XN are samples of a population X. The sample
mean is defined as
_ I N
X=- I Xi (B.42)
Ni~'

where X is an unbiased estimate of the population mean. The sample mean

is a random variable and its variance is liN of the population variance, i.e.,

- I
V(X)=- V(X) (B.43)
N

If the population follows a distribution with mean J.l and variance (j2, then
the sample mean approximately follows a normal distribution with mean J.l
and variance (j2I N when the sample size is sufficiently large.
The sample variance is defined as

I
S
2
=- IN (Xi-X)
- 2
(B.44)
N-I i~'

where ; is an unbiased estimate of the population variance. The sample

variance is also a random variable and its variance is given by

(B.4S)

where V4 is the fourth-order central moment of the population X.

In Monte Carlo simulation, the sample mean and the sample variance
are calculated repeatedly as the number of samples increases. The recursive
equations for the sample mean and the sample variance are as follows:

(B.46)
326 APPENDIX B

and
~ 1 ~ -2 -2 2
SN=-- [(N-2)SN-I +(N-l)X N- 1 - NX N+X N] (B.47)
N-l

B.6. REFERENCES
1. A. Papoulis, Probability, Random Variables and Stochastic Processes, McGraw-Hill, New
York,1965.
2. P. L. Meyer, Introductory Probability and Statistical Applications, Addison-Wesley, Reading,
Massachusetts, 1972.
3. A. W. Drake, Fundamentals of Applied Probability Theory, McGraw-Hili, New York, 1967.
4. R. Billinton and R. N. Allan, Reliability Evaluation of Engineering Systems: Concepts and
Techniques, Plenum, New York, 1983.
5. R. Y. Rubinstein, Simulation and the Monte Carlo Method, Wiley, New York, 1981.
6. Z. Fang, Computer Simulation and Monte Carlo Method, Publishing House of Beijing
Industrial Institute, 1988.
 APPENDIX c
Power System
Analysis Techniques

C.1. AC LOAD FLOW MODELS

C.1.1. Load Flow Equations

The basic load flow equations in polar coordinate form are as follows:

n
Pi=Vi I ~(Gijcos8ij+Bijsin8ij) (i= 1, ... , n) (C.l)
j=!

n
Qi= Vi I ~(Gijsin 8irBijcos 8ij) (i= 1, ... ,n) (C.2)
j=!

where Pi and Qi are bus real and reactive power injections at Bus i; V; and
8 i are the magnitude and angle of the voltage at Bus i; 8ij= 8 i - 8j ; Gij and
Bij are the real and imaginary parts of the element of the bus admittance
matrix; n is the number of system buses.
Each bus has four variables (V;, 8i , Pi, and Qi) and therefore n buses
have 4n variables in total. (C.l) and (C.2) consists of2n equations. In order
to solve the load flow equations, two of the four variables for each bus have
to be prespecified. In general, Pi and Qi of the load buses are known and
they are called PQ buses; Pi and V; of generator buses are specified and they
are called PV buses; Vi and 8i of one bus in the system must be specified to
adjust the power balance of whole system, and this bus is called the swing
bus.
327
328 APPENDIX C

C.1.2. Newton-Raphson Model

The load flow equations can be solved by the successive linearization
method. This is the well-known Newton-Raphson model. Equations (C.l)
and (C.2) are linearized to yield the following matrix equation:

(C.3)

The Jacobian coefficient matrix is a (n+m-l)-dimensional square matrix

where nand m are the numbers of all buses and load buses, respectively;
AVIV implies that its elements are AV;jV;. The elements of the Jacobian
matrix are calculated by

Hij= :~:= V; Jj{Gij sin oij- Bijcos Oij)

J
(C.4)
oP; 2
.. 00; -Q.-
H··=-= , B··V·
.. ,

Nij= :~ f}= V; Jj{Gij cos oij+Bijsin oij)

J
(C.S)
apt 2
N,.;=-
, oV;' v'·=p·+G··V·
, .. ,

(C.6)
oQ; 2
J··=-=P1-G··V·
.. 00; .. ,

oQ;
D·=-v'·=H··
lJ of} J lJ

(C.7)
oQ; 2
D·=- V,.= Q.-
.. oV;' , B··V·
.. ,

C.1.3. Fast Decoupled Model

Branch reactances are normally much larger than branch resistances
and angle differences between two buses are very small in power systems.
This results in strong dominance of the diagonal matrix block, i.e., the values
POWER SYSTEM ANALYSIS TECHNIQUES 329

of matrix blocks Nand) are much smaller than those of Hand L. Equation
(C.3) can be decoupled by assuming N=O and )=0. Considering
IGijsin8ijl«IBijcos8ijl and IQil«IBiiV~I, the decoupled equation can be
further simplified to
[AP / V] = [B'][ VA8]
(C.S)
[AQ/V] = [B"][AV]

where [AP/V] and [AQ/V] are vectors whose elements are AP;/Vi and
AQ;/Vi, respectively; [VA8] is a vector whose elements are Vi A8 i . The
elements of the constant matrices [B'] and [B"] are obtained using
-1
Bij=-
xij (C.9)
Bii=- I Bij
jeR,

B~.=---.!L
y 2 2
rij+xij
(C.1O)
B1;=-2b iO - I Bij
jeR,

where rij and xij are the branch resistances and reactances, respectively; biO
is the branch susceptance between Bus i and the ground; R; is the set of
buses directly connected to Bus i.

C.2. DC LOAD FLOW MODELS

C.2.1. Basic Equations

DC load flow equations relate the real power to bus voltage angles.
DC load flow based models are widely used in composite system adequacy
evaluation because: (a) Many of the important reliability indices are associ-
ated with real power load curtailments and calculating these indices only
requires real power related information. (b) Load flow calculations in practi-
cal application indicate that in many systems there are relatively small
(3%-10%) differences between AC and DC load flows. These are small
compared to possible errors due to uncertainties in basic reliability data
such as component failure rates and outage times. (c) A large number of
system states have to be evaluated to guarantee accuracy of probability
330 APPENDIX C

indices. DC load flow based models including optimal power flow type can
be rapidly calculated.
It should be clearly appreciated that if voltage and reactive power con-
siderations are important requirements in a particular system study, then
DC load flow is not an acceptable approach.
DC load flow is based on the following four assumptions:
1. Branch resistances are much smaller than branch reactances. Branch
susceptances can be approximated by
1
bij~-- (C.lI)
xij
2. Voltage angle difference between two buses of a line is small and
therefore
sin Dij~D;-Dj
(C.I2)
cos Dij~ 1.0

3. Susceptances between the buses and the ground can be neglected,

i.e.,
hiO=hjO=O (C.13)
4. All bus voltage magnitudes are assumed to be 1.0 p.u.
Based on the above assumptions, the real line flow in a branch can be
calculated by

P .. =D;-Dj (C.14)
!J
xij
and therefore bus real power injections are
P;= L Pij=Bi;D;+ L BijDj (i= 1, ... ,n) (C.IS)
jeR, jeR,
where

'
Bij=--
1 and Bi;=- L Bij (C.16)
xij jeR,
Equation (C.IS) can be expressed in a matrix form:
P=[8][D] (C.17)
If Bus n is selected as the swing bus and we let Dn = 0, [8] is a (n - 1)
dimensional square matrix. It is exactly the same as [8] in equation (C.8).
POWER SYSTEM ANALYSIS TECHNIQUES 331

C.2.2. Relationship between Power

Injections and Line Flows
Based on the DC load flow equation (C.I7), a linear relationship
between power injections and line flows can be obtained. Combining equa-
tions (C.I4) and (C.17) yields
[Tp] = [AUP] (C.I8)
where Tp is the line flow vector and its elements are line flows {Pij}. Matrix
[A] is the relationship matrix between power injections and line flows and
its dimension is L x (n - 1), where L is the number of lines and n the number
of buses; [A] can be calculated directly from [U]. Assume that two buses
of line k are i and j. For k= 1, ... ,L, the kth row of matrix [A] is the
solution of the following linear equations:
[UUX] = [b] (C.19)
where

b= [0, ... ,0, ~, 0, ... ,0, _~, 0, ... ,o]T (C.20)

xf xf
itheiement jtheiement

C.3. OPTIMAL POWER FLOW

Optimal power flow (OPF) can be defined as finding a solution of the
power system operating state including control variables and state variables
to optimize a given objective and satisfy load flow equations (equality con-
straints) and all given feasibility and security requirements (inequality
constraints).
There are a number of possible types or forms of OPF depending on
different objectives and constraints. Generally, OPF is a nonlinear optimiza-
tion problem. Many linearized OPF models, however, are used for different
purposes, particularly in composite system reliability evaluation. The mini-
mization models given in Chapters 5 and 7 are essentially linearized OPF.
This section provides a general representation of OPF.
OPF is an optimization problem and can be mathematically described
as follows:
min F= f(P, Q, V, 8) (C.2I)
332 APPENDIX C

subject to
Pi(V, (j)=PD i (ieND) (C.22)
Qi(V, (j)=QD i (ieND) (C.23)
PG}"in:::;pi(V, (j):::;PG?",x (ieNG) (C.24)
QG;mn:::;Q;(V, (j):::;QG}"ax (ieNG) (C.25)
ITk(V, (j)I:::; Tk' (keL) (C.26)
(ieN) (C.27)
where P and Q are bus real and reactive power injection vectors, respectively;
Pi and Qi are their elements; V and (j are bus voltage magnitude and angle
vectors; Vi is an element of V; PD i and QDi are real and reactive loads at
Bus i; PG}"in, PGF, QG}"in, and QGF are lower and upper limits of real
and reactive generations at Bus i, respectively; Tk is the line power flow on
line k and T7(ax is its capacity limit; V}"in and V}"ax are lower and upper
limits of bus voltage magnitude at Bus i; ND, NG, N, and L are sets of
load buses, generator buses, all buses, and all lines in the system, respectively.
The P, Q, V, and (j in the objective function F may be different subsets of
these variables, depending on the selection of control and state variables.
In composite system reliability evaluation, it is necessary to introduce
bus load curtailment variables into the OPF model. This has been done in
the linearized OPF models given in Chapters 5 and 7. It is possible to
consider more constraints in an OPF model which can reflect system security,
such as operating limits on voltage and transient stability.

C.4. CONTINGENCY ANALYSIS

The purpose of contingency analysis in composite system adequacy
evaluation is to calculate line flows following one or more line outages and
judge if there are overloads in other lines. The most accurate method is to
rebuild the bus admittance matrix and resolve the load flow equations for
each line outage state. This is not practical when tens of thousands or even
hundreds of thousands of outage states need to be evaluated in a composite
system adequacy assessment. Outage state line flows can be approximately
obtained from information from the normal state using contingency analysis
techniques without resolving the load flow equations. A linearized load-
flow-based contingency analysis method is described in Section 5.2.2. This
section presents a supplementary AC load-flow-based sensitivity method for
contingency analysis.
POWER SYSTEM ANALYSIS TECHNIQUES 333

Figures C.I a and C.l b show the the preoutage and the postoutage states
for the line i-j outage. Figure C.lc shows the case in which additional power
APi+jAQi and APj+jAQj are injected at Buses i andj, respectively, in the
preoutage state. If the additional power injections can produce power flow
increments so that power flows on the rest of the system are the same as
those in the postoutage state, the effect of the additional power injections is
equivalent to the outage of line i-j. In the postoutage state,
PI +jQi= Pia +jQla
(C.28)
Pj +jQj = Pip +jQ}p
In the equivalent power injection case,
(Pi + APi) + j(QI+ AQi) = (Pia + Pij+ APij) + j(Qla + Qij+ AQij)
(C.29)
(Pj+ APj ) + j(Qj+ AQj) = (P}p + Pji + APji) + j(Q}p+ Qji+ AQji)
where Pij+ jQij and Pji +jQji are the power flows on line i-j in the preoutage
state and APij+ jAQij and APji +jAQji are power increments on line i-j due
to additional power injections to the preoutage state.
Subtracting equation (C.28) from equation (C.29) yields
APi +jAQ. = (Pij+ APij) +j(Qij+ AQij)
(C.30)
APj+jAQj= (Pji + APji) +j(Qji+ AQji)
Equation (C.30) can be rewritten in matrix form:

(C.31)

By introducing the sensitivity relationship between the power injection

increments and power flow increments on line i--j, equation (C.32) can be
obtained from equation (C.31):
oPij OPij oPij oPij
OPi OQi oPj oQj
0 0 oQij oQij oQij oQij

[P']
[~ ~} [M]
I 0 OPi OQi oPj oQj
Qij =
Pji 0 I oPji oPji oPji oPji
~;; (C.32)

Qji 0 0 OPi OQi oPj oQj AQj

OQji OQji OQji OQji
OPi OQi oPj oQj
334 APPENDIX C

Seta Setp

(a)

Seta Setp

(b)

(Pj + jOj) + (APj + jAOj)

Seta Setp

(~j+ jOlj) + (~i+ jOjl) +

(APij+ jAO lj ) (A~i+ jaO jl )

(e)

Figure C.1. Equivalent power injections for a line outage.

POWER SYSTEM ANALYSIS TECHNIQUES 335

The sensitivity matrix S can be calculated by decomposition into two factors:

oPIj oPIj oPIj oPIj 00; 00; 00; 00;
-
00; OOj oV; of} oP; oQ; oPj oQj
oQIj oQIj oQIj oQIj OOj OOj OOj OOj
00; OOj oV; of} oPi oQ; oPj o~
S= (C.33)
oPj; oPj; oPj; oPj; oJ'; oV; oV; oV;
- -
00; OOj oV; of} oP; oQ; oPj oQj
oQj; oQj; OQji o~; of} of} of} of}
00; OOj oV; of} oP; OQi oPj oQj
The first matrx is obtained from the explicit expression of the line power
as a function of the two terminal bus voltages. The elements of the second
matrix are included in the inverse matrix of the Newton-Raphson load flow
Jacobian matrix and can be obtained by solving the linear equations. For
example, the elements of the first row in the second matrix are included in
the solution of the following linear equation:
[J][X]=[b] (C.34)
where
[J] = Jacobian matrix prior to the outage
[b] = [0, ... ,0, 1,0, ... ,O]T
i
(corresponding to 0 i)
Having obtained the equivalent power injection increments from solving
equation (C.32), the increments of bus voltage magnitudes and angles due
to the line outage can be obtained by resolution of the equation

Ao J= [AI]
J[AVIV (C.35)

where
[AI] = [0, ... 0, AP;, 0, ... 0, APj> 0, ... 0, AQ;, 0, ... 0, AQj> 0, ... of
The line power flows following the line outage can be calculated using bus
voltages.
The concept and the procedure given above can also be applied to
multiple line outages. The procedure is similar for the fast decoupled load
flow model. The equivalent real and reactive power injections can be calcula-
ted separately in this case.
336 APPENDIX C

C.5. REFERENCES
1. w. F. Tinney and C. E. Hart, "Power Flow Solution by Newton's Method," IEEE Trans.
on PAS, Vol. 86, No. 11, 1967.
2. B. Stott and O. Alsac, "Fast Decoupled Load Flow," IEEE Trans. on PAS, Vol. 93, No.3,
1974.
3. W. Li, "Optimal Power Flow Solution by Combination of Penalty Method and BFS Tech-
nique," Journal ojChongqing University, No. I, 1982, pp. 95-107.
4. K. R. C. Mamandur and G. J. Berg, "Efficient Simulation of Line and Transformer Outages
in Power Systems," IEEE Trans. on PAS, Vol. 101, No. 10, 1982.
5. A. J. Wood and B. F. Wollenberg, Power Generation, Operation and Control, Wiley, New
York,1984.
6. W. Li, Secure and Economic Operation oj Power Systems-Models and Methods, Chongqing
University Publishing House, 1989.
 APPENDIX D
Optimization
Techniques

D.1. LINEAR PROGRAMMING

The following is a brief summary of the optimization techniques used in this
book.

D.1 .1. Basic Concepts

The basic linear programming problem is to minimize (or maximize) a
linear function while satisfying a set of linear equality and inequality con-
straints and has the following standard form:

min ex (D.I)
subject to
Ax=b (D.2)

x~O (D.3)

where e is an n-dimensional row vector, x an n-dimensional column vector,

b an m-dimensional column vector, b~O, and A is an m x n dimensional
matrix.
A linear programming problem in nonstandard form can be converted
into the standard form by equivalent transformation. In the following two
LP problems, for example, the left one can be converted into the right one
337
338 APPENDIX D

in the standard form by adding relaxation variable vector y:

min ex min ex
subject to subject to
Ax:::;;; 0 Ax+y=O
x~O x~O
y~O

Basic definitions associated with linear programming are as follows:

1. Feasible solution: A solution x satisfying (D.2) and (D.3). The set

of all feasible solutions is called a feasible zone.
2. Optimal and feasible solution: A feasible solution satisfying (D.1).
3. Base: If the rank of Matrix A is m, a nonsingular m x m dimensional
matrix block B in Matrix A is known as a base. The columns of B
are known as basic vectors. The elements of x corresponding to the
basic vectors are basic variables and the remaining elements of x are
nonbasic variables.
4. Basic solution: A solution of (0.2) with all nonbasic variables being
zero.
S. Basic feasible solution: A basic solution satisfying (0.2) and (D.3).
The base corresponding to the basic feasible solution is called the
feasible base.
6. Optimal basic solution: A basic feasible solution satisfying (D.1).
The base corresponding to the optimal basic solution is called the
optimal base.

0.1.2. Generalized Simplex Method

When variable x has both lower and upper bounds, the constraints of
x can be converted into a form O:::;;;x:::;;;h. A more general LP problem there-
fore is
min ex

subject to

Ax=b

O:::;;;x:::;;;h

The approach to solving such a general LP problem is called the generalized

simplex method. Its basic steps are as follows.
OPTIMIZATION TECHNIQUES 339

Step 1: Determine an initial basic feasible solution using the artificial

variable technique and create an initial simplex tableau:
b

o YI,m+1 YI,n YIO

o

Xi 0 0 ... 1 ... o Yi,m+1 Yi,n Y.o

Xm 0 0 1 Ym,m+1 Ym,n YmO

o 0

In the tableau, Yij and Y.o are coefficients corresponding to Matrix A and
Vector b, respectively, following each Gaussian elimination step; rj= Cj- Zj
U=m+ 1, ... ,n), where Cj are the coefficients in the objective function of
the original problem which are known as direct cost coefficients;
Zj=I CiYij are known as composite cost coefficients and rj are known as
relative cost coefficients; Zo = I CiY.o is the value of the objective function at
the present step; ej= + or - U= 1, ... ,n) which is called the sign row. For
the initial basic feasible solution
(if Xi is a basic variable)
X.= {Y.o
I 0 (if Xi is a nonbasic variable)
the ej take + signs.
Step 2: Select rk=min{rj<O,j=m+ 1, ... , n}. The kth column is
called the pivotal column. If there is no negative rj, the present solution is
already an optimal and feasible solution and the simplex process ends. The
values of the variables are determined according to the signs of ej. If ej = +,
Xj=Xj, and if ej= -, xj=hj - Xj.
Step 3: Calculate the following three values for the selected pivotal
column:
• hk
• 0 1 = min {Y.o/Yik} for all Yik>O (if there is no positive Yik, 0 1 = 00)
• 02=min{(Y.o-hi)/Yik} for all Yik<O (if there is no negative Yib
O2 = 00)
340 APPENDIX D

Step 4: Modify the simplex tableau according to the magnitude of the

three values in Step 3:
• If hk is minimum, the last column is subtracted by a column which
is obtained from the kth column mUltiplying hk and then the kth
column is multiplied by -1 (including the change of the sign for ek).
The base remains unchanged.
• If OJ is minimum and OJ appears in the qth row, then Yqk is selected
as a pivot.
• If (}2 is minimum and (}2 appears in the qth row, then YqO(new) =
YqO(oJd) - hq, Yqq is multiplied by -1, and the sign of eq is changed; Yqk
is selected as a pivot.
Step 5: With the selected pivot element Yqk. conduct Gaussian elimina-
tion in the simplex tableau so that the pivot becomes 1; the other elements
in the pivotal column become O. An updated simplex tableau is obtained
and go to Step 2.

0.1.3. Duality Principle

For each minimum-value linear programming problem, there is a corre-
sponding dual maximum-value linear programming problem. The
coefficients of the objective functions and the right-side terms of the con-
straints in the two LP problems exchange and they have an equal optimal
objective function value. This is known as the duality oflinear programming.
Two pairs of dual LP problems are given in the following. The first
pair is called the symmetrical form of the dual LP and the second pair the
nonsymmetrical form.
Primary problem Dual problem
Symmetrical form
mincx min }.h
subject to subject to
Ax~b AA S;;C
x~O A~O

Nonsymmetrical form
mmcx min }.h
subject to subject to
Ax=b AA S;;C
x~O

where x and A are variable vectors for the primary and dual problems,
respectively.
OPTIMIZATION TECHNIQUES 341

The optimal basic solution of the dual problem can be obtained from
the final simplex tableau of the primary problem. This solution is
(0.4)

where B is the optimal base and CB a subvector of C coefficient row in the

final simplex tableau, which corresponds to the basic variables. At optimal
solution, the nonbasic variables are zero and therefore the optimal value of
the objective function is
(0.5)

where XB is the basic variable vector. Differentiating equation (0.5) yields

AF= cBB- 1M (0.6)
Combining equations (0.4) and (0.6) yields
M=AAb (D.7)

This equation indicates that the optimal dual variable Ai is a partial differen-
tial of the objective function with respect to the right-side value bi of the
constraints. This is the fundamental to calculate sensitivity indices in reli-
ability assessment.

0.1.4. Dual Simplex Method

The simplex method given in Section D.l.2 is known as the primal
simplex algorithm. Starting from an initial feasible solution, an optimal
solution is gradually obtained while retaining feasibility in the primal algo-
rithm. The dual simplex method starts with an initial basic solution satisfying
optimality of the objective function but not satisfying feasibility. Feasibility
is gradually obtained under the condition that optimality is retained. Which
method is used depends upon the features of the problems to be solved. If
an initial feasible solution can be easily obtained, the primal algorithm is
used. If an initial optimal but nonfeasible solution can be obtained, the dual
algorithm is used. The primal simplex method is used in the LP models for
multi-area reliability evaluation in Chapter 4 and the increment-type LP
model for voltage adjustment in Section 5.9.2, while all other LP models in
Chapters 5 and 7 utilize the dual simplex method. It should be noted that
the dual algorithm is used to perform dual treatment on the simplex tableau
of the primal problem but not to solve a dual problem of the primal problem.
The basic steps of the dual simplex method can be summarized as follows:
342 APPENDIX D

Step 1: Create the initial simplex tableau of the primal problem.

Step 2: Find a dual basic feasible solution X B , i.e., in the simplex tab-
leau corresponding to this solution, rj;;:::O for j=m+ I, ... , n. If XB;;:::O, i.e.,
there is no negative element in the column b of the simplex tableau, then an
optimal and feasible solution is already obtained. If there are any negative
elements in the column b, go to Step 3.

Step 3: Select the smallest value in the negative elements of XB, i.e.,

min {(XB)tI(XB)i<O} =Xq

i

The Xq is the leaving base variable. This means that the qth row is the pivotal
row.

Step 4: Check all elements of the pivotal row Yqj U= I, ... ,n). If all
Yqj;;:::O, there is no feasible solution. If there are negative elements in the
pivotal row, then

(} = min {(Zj- cj)/YqjIYqj~O} = (Zk-Ck)/Yqk

j

where Zjo Cjo and Yqj are as defined in Section 0.1.2; Xk is the entering base
variable, which means that the kth column is the pivotal column.

Step 5: With the pivot element Yqk, conduct Gaussian elimination to

update the simplex tableau. An updated base is obtained and then a new
dual basic feasible solution is calculated: xB=B-1b. Go back to Step 2.

0.1.5. Linear Programming Relaxation

Technique
In composite system adequacy assessment, the number of constraints in
linear programming OPF models are very large, particularly when preventive
outage constraints are considered. Active constraints at an optimal solution
or in the process of resolution, however, are relatively few. If the active
constraints are satisfied, all constraints will be satisfied. The basic character-
istic of the linear programming relaxation technique is that a large-scale LP
problem is decomposed into a sequence of small-scale linear programming
problems and only a few active constraints are considered in each small LP
problem.
OPTIMIZATION TECHNIQUES 343

Consider the linear programming problem:

min ex
subject to
LPI { Alx=b
J:::; A 2x::;y
-!::; x::; i
where Al is a Jx S dimensional matrix; A2 is an Mx S dimensional matrix;
x is an S dimensional column vector; b is a J dimensional column vector;
e is an S dimensional row vector.
Let
and h=[O,e]

Z=[:J ~=[~J z=[~J

the linear programming problem LPI can be rewritten as LP2:
minhz

Az=d
~::;z::;z

where

and d=[~J
Let K eM, i.e., K is a subset of M. The following linear programming
problem is constructed:
minhHzh
subject to
LP2(K) { AH
GZH-
-dG
ZH::;ZH::;ZH

where hH denotes a subvector of h and the subscript set of its elements is

H. The definitions of ZH and dG are similar; Ag denotes a submatrix of A.
The subscript set of its columns is H and the subscript set of its rows is G;
G=KuJ and H=KuS.
The LP2(K) is a small-scale linear programming problem which con-
tains the partial inequalities and all equalities in the original problem LPI.
344 APPENDIX D

Subset K can be quite small and even an empty set initially. Consequently,
LP2(K) can be a very small problem despite the scale ofthe original problem.
A solution of LP2(K) satisfies optimality of the original problem but does
not necessarily satisfy its feasibility (not all inequality constraints are neces-
sarily satisfied). The LP2(K) is solved using the dual simplex method.
Let

zIA.K) = [Y.rl K )]
x(K)
denote the optimal basic solution of LP2(K) and /(K) is the sUbscript set
of the basic variable components of the solution. Assume that

R=M-K, ( K) = [Y.rl K )] z(K) = [Y(K)]

Y YR(K) , x(K)
There are two cases:
(a) If
~~y(K)~y

is satisfied, then z(K) is an optimal basic solution of the original problem

LPI.
(b) If subset
F(K) = {iERly.{K) > Yi' or ylK) < Yi}

is not empty, let

E(K) = K n J(K)
and
K* =K-E(K) +F(K)
and K* is used to replace K to construct a new small-scale linear program-
ming LP2(K*). This LP2(K*) is solved using the dual simplex method. The
initial basic solution of LP2(K*) is ZH* where H* = K* u S. The sUbscript
set of the basic variable components of ZH* is /* = /(K) - E(K) + F(K).
The above process is repeated (the old K* is replaced by a new K*)
until Case (a) is achieved.

0.2. MAXIMUM FLOW METHOD

The maximum flow method is applied to multi-area generating system
adequacy evaluation in Chapter 4. This section presents its basic concepts
and procedure.
OPTIMIZATION TECHNIQUES 345

0.2.1. Basic Concepts

(a) Definitions

Network (oriented graph): A set composed of node subset P and

oriented branch subset B, denoted by
G(P,B)
Arc: An oriented branch between two nodes i
and j, denoted by aij.
Arc parameters: Arc length Wij, arc flow fij, and arc capac-
ity cij'
Source: A node producing flow, denoted by s.
Sink: A node collecting flow, denoted by t.
Path: An oriented sequence of nodes and arcs.
An arc having the same direction as the
path is called a forward arc and an arc
having the opposite direction to the path
is called a backward arc.
Augmented path: If all arc flows on a path satisfy the fol-
lowing conditions, this path is called an
augmented path:
O~Jij<
I" cij if aijEjJ +
O<fij~cij if aijEjJ-
where jJ + and jJ - denote forward and
backward arc sets, respectively.
Cut set: A set of arcs connecting two mutually
complementary node subsets in a net-
work, denoted by {aijliEP"jEPj }. The
sum of all arc capacities in a cut set is
called the cut capacity. The one having
minimum cut capacity in all possible cut
sets is called the minimum cut set.

(b) Two Basic Theorems

Theorem 1: A feasible flow f = {fij} is the maximum flow if and only if

there is no augmented path between the source and the sink.

Theorem 2 (Ford-Fulkerson theorem): The flow capacity ofa maximum

flow from the source to the sink equals the cut capacity of the minimum cut
set separating the source and the sink.
346 APPENDIX D

0.2.2. Maximum Flow Problem

The maximum flow problem is to find a feasible flow f= {j;J which
has maximum flow capacity. It is described mathematically as follows:

subject to
O<I:·<C··
-Ji]- lJ (for all aij)

IJij-IJji=O (i ";:S, t)
j j

where sand t indicate the source and the sink, respectively; F is the flow
capacity.
There are different methods for solving the maximum flow problem.
The most commonly used method is the labeling algorithm. This algorithm
starts from a feasible flow and includes the two processes of labeling and
adjusting.

(a) Labeling. In this process, nodes are classified into the labeled and
the unlabeled. The labeled nodes are further divided into the checked and
the unchecked. Each label includes two numbers: The first number indicates
from which node the label is obtained. The second number gives the possible
maximum flow increment from the prior node to the present node.
Step 1: Label the source with (0, +00) so that the source is a labeled
but unchecked node and other nodes are unlabeled.
Step 2: Consider any labeled but unchecked node i. There are three
cases for all nodes j which are connected to i but unlabeled:
• If j;j= cij for arc aij, node j is ignored.
• Ifj;j<cij for arc aij, nodej is labeled with (+i, oU», where ou)=
min[o(i), cij-fij]' +i indicates that the flow can be increased from i
to j and oU) is the possible magnitude to be increased.
• Ifjji> 0 for arc aj;, nodej is labeled with (-i, oU», where ou)=
min[o(i),jjd. -i indicates that the flow can be decreased fromj to i
and oU) is the possible magnitude to be decreased.
OPTIMIZATION TECHNIQUES 347

After all possible j are labeled, i becomes a labeled and checked node
and all j become the labeled but unchecked nodes.
Step 3: Repeat the above process until the sink is labeled and then go
to the adjusting process. If all possible nodes are checked and the labeling
process cannot proceed further to label the sink, the algorithm ends. The
present flow is a maximum flow.

(b) Adjusting

Step 1: Find an augmented path J.l from the source to the sink in terms
of the first number in the labels. Assume that the first number in the label
for the sink tis +k (or -k). Arc akt (or atk) is an arc on the augmented
path. Then the first number in the label for node k is checked. If it is
+i (or -i), aile (or aki) is an arc on the augmented path. The pursuit is
conducted until the source is checked.
Step 2: Adjust the flow capacities of the arcs on the augmented path:

if aijEJ.l
if aijEJ.l+
if aijEJ.l-
where oCt) is the second number in the label for the sink.
Step 3: Erase all labels and return to the labeling process with the new
feasible flow {f Ii}·
A minimum cut is also obtained using the presented labeling method.
When labeling ends, the cut set between labeled and unlabeled node subsets
is the minimum cut set. It locates the "bottle neck" from the source to the
sink.

0.3. REFERENCES
1. D. G. Luenberger, Introduction to Linear and Nonlinear Programming, Addison-Wesley,
Reading, Massachusetts, 1973.
2. L. S. Lasdon, Optimization Theory for Large Systems, Macmillan, New York, 1970.
3. W. Li, Secure and Economic Operation of Power Systems-Models and Methods, Chongqing
University Publishing House, 1989.
4. W. Li, "A Penalty Linear Progranuning Model for Real-Time Economic Power Dispatch
with N -1 Security, International Symposium on Engineering Mathematics and Application,
Beijing, July, 1988.
348 APPENDIX D

5. J. C. Dodu and A. Merlin, "Une Application de la Programmation Lineaire a l'Etude des

Reseaux Electriques de Grande Taille," Bulletin de la DER, EDF, Serie C, No. I, 1974,
pp.29-56.
6. P. A. Jenson and J. W. Barnes, Network Flow Programming, Wiley, New York, 1980.
7. L. R. Ford and D. R. Fulkerson, Flows in Networks, Princeton University Press, 1962.
Index

ACCI,28 Coefficient of variation, 36

Active event, 235 Coherence, 199
Adequacy, 9 Common cause outage, 164
Adequacy indices, 22 Component operating cycle, 219
ADLC,25 Composite system, 131
AEIC,290 Composite system adequacy, 131
AEGC,290 Congruential generator, 39
AENS,28 Connection set, 233
AGCP,290 Contingency analysis, 332
AIOIC,290 Control variates, 53
Alert state, 188 Corrective actions, 190
Annual index, 27 Correlation function, 322
Annual load curve, 93 Covariance, 322
Annualized index, 27 Customer damage function, 258
Antithetic variates, 58 Customer survey, 256
Arc, 345 Cut set, 345
ASAI,28
ASUI,28
Augmented path, 345 Dagger sampling, 58
DC load flow, 329
Delivery point, 21
Base, 338 Density function, 318
Basic solution, 338 Derated state model, 79
Basic variable, 338 Distribution function, 318
BES,21 Distribution of index, 84
BPACI,26 Distribution system adequacy, 209
BPECI,26 Dual problem, 340
BPII,26 Dual simplex, 341
Bus load correlation, 171 Duality, 340

CAIDI,28 EDLC,25
CAIFI,27 EDNS,26
CCDF,259 EENS,26
Central limit theorem, 324 EEPG,286
Chronology, 180 EFLC,25
Cluster technique, 95 EGC,286

349
350 INDEX

EIC, 29 Load point unavailability, 214

ELC, 25 Load uncertainty, 110
Emergency state, 188 LOEE,23
ENLC,25 Log-normal distribution, 320
ENS, 28 LOLD,24
EOIC, 290 LOLE,23
EPSRA,20 LOLF,24
ERIS, 19
Estimate, 325
Expectation, 322 Multistep model, 93
Exponential distribution, 320 Maximum flow method, 344
Extreme emergency state, 188 MBPCI,26
Minimization model, 137
Fast decoupled model, 328 Minimum cost assessment, 284
Feasible solution, 338 Modulus, 39
Forced unavailability, 91 Monte Carlo, 33
Ford-Fulkerson theorem, 345
Frequency index, 179
Functional zone, 10 NCDE,200
NCDF,200
Gamma distribution, 321 NCDP, 201
Generalized simplex, 338 Network, 345
Generating system adequacy, 75 Newton-Raphson model, 328
Generating unit contingency, 135 Noncoherence, 199
Normal distribution, 320
Generating unit modeling, 78
Normal state, 187
Numerical characteristic, 321
Hierarchical level, II
Hybrid method, 191
Optimal and feasible solution, 338
IEAR,29 Optimal load flow, 331
IEEE RTS, 299 Optimization techniques, 337
Independence, 39
Importance sampling, 53
Interruption cost model, 269 Parameter estimation, 324
Inverse transform, 43 Passive event, 234
IOR,211 Path,346
Peaking unit model, 81
K-mean algorithm, 95 PLC, 24
Point, 233
Population, 324
Labeling, 346
Present value, 284
Law of large numbers, 323
Primary problem, 340
Limit theorem, 323
Probabilistic analysis, I
Linear programming, 337
Probability, 317
Linear programming relaxation, 342
Probability distribution, 318
Load curtailment, 137
Load flow, 327
Load point failure rate, 214
Load point outage duration, 214 Q-Ioad curtailment, 190
INDEX 351

Random number, 39 State sampling, 60

Random variate, 42 State transition sampling, 64
RBTS,306 Station component modeling, 233
Regional weather states, 153 Station reliability, 232
Reliability evaluation, 3 Stopping rules, 83, 96
Reliability worth, 16 Stratification, 197
RENS,I44 Stratified sampling, 56
RGTAI,I44 Supporting policy, 120
RLCP,I44 System operating cycle, 221
RNLC,I44 System operating states, 187
RPII,I44 System performance index, 230

SAIDI,27 Tabulating technique, 45

SAIFI,27 Transmission component outage, 135
Sample mean, 325 TTF,79
Sample variance, 325 TTR,79
Samples, 324
Sampling distribution, 324
SCDF,258 Unavailability, 35
Security, 9 Uniformity, 39
Security constraints, 188
Sensitivity index, 114
Service continuity, 22 Variance, 322
SI,26 Variance reduction technique, 52
SIC, 259
Sink, 345
Source, 345 Weibull distribution, 321
Standard deviation, 322 WTA,257
State duration sampling, 62 WTP,257