The Hong Kong Polytechnic University Economic Operation

1. Introduction
• There are many factors involved in the successful operation of a power
system. Next to security, economy is the most important.
• Economic load dispatch concerns with the optimum load scheduling of the
different generating plants in the system and it involves 2 different problems
‘

1. Economic Dispatch – an on‐line problem to optimally distribute the load

among the generating units to minimise the total cost of supply while
the total demand and the losses at any instant is met by the total
generation.
2. Unit Commitment – a pre‐dispatch problem to select optimally out of
the available generating units to meet the expected load and provide a
specified margin of operating reserve over a specified period of time.
1

The Hong Kong Polytechnic University Economic Operation
Economic Dispatch (ED) Problem
• Given load
• Given set of on‐line (in service) units
A B C Load
• How much should each unit generate to meet
this load at minimum cost? ? MW ? MW ? MW

Unit Commitment (UC) Problem
GW
• Given load profile
(e.g. values of the load for each hour of a day)
• Given set of units available
• When should each unit be started, stopped and
how much should it generate to meet the load
at minimum cost?

? ? ?
Load Profile
G G G Hours
2

1
 The Hong Kong Polytechnic University Economic Operation

2. Characteristics of Thermal Units
Flexible Power Plants:
• Power output can be adjusted (within limits)
– Thermal power plants (Coal‐fired, oil‐fired, simple cycle gas turbines, combined cycle gas
turbines)
– Hydraulic power plant reservoir
• Status and power output can be optimized

Inflexible Power Plants:

• Power output cannot be adjusted for technical or commercial reasons
– Nuclear
– Run‐of‐the‐river hydro
– Renewables (wind, solar,…)
– Combined heat and power (CHP, cogeneration)
• Output treated as given when optimizing 3

The Hong Kong Polytechnic University Economic Operation

• H : Btu per hour heat input to the unit (or MBtu/h)

Steam Turbine
• C : per hour operating cost rate of a unit
Boiler Fuel (Gross) (Net)
C = (Fuel Cost) x H ($/h) Input B T G
Generator
• The operating cost of a thermal plant is mainly the
cost of the fuel. Other costs such as costs of
maintenance, labour and supplies contribute only A/P
to a small extent and are assumed to vary as a Auxilary Power System
fixed percentage of the fuel cost.
Boiler-turbine-generator unit

• The fuel cost is meaningful in case of thermal stations, but for hydro stations where the
energy storage is ‘apparently free’, the operating fuel cost as such is not meaningful. This
is the reason why only thermal plants are considered in the following sections

2
 The Hong Kong Polytechnic University Economic Operation

In practice, the operating cost Ci is constructed of piecewise ($/h)

Unit operating cost, Ci
non‐linear functions valid for ranges of output Pi based on
empirical data. By fitting a suitable degree polynomial, an
analytical expression for operating cost can be written as ∆Ci
Ci  Pi . It generally suffices to fit a second degree ∆Pi
polynomial, i.e. (MW)
Ci  B2 Pi  B1 Pi  B0
2
$/h Pi,min Unit Output, P Pi,max
i
where B0 , B1 and B2 are constants depending upon a
particular plant and

Incremental cost, dCi/dPi
($/MWh)
Pi ,min  Pi  Pi ,max
linear approximation
The slope of the cost curve, i.e. dCi dPi is called the
incremental fuel cost ( ICi ). If the cost is approximated as
a quadratic as in the above, then ICi is linear
(MW)
ICi  2 B2 Pi  B1 $/MWh
Pi,min Unit Output, P Pi,max
i 5

The Hong Kong Polytechnic University Economic Operation

3. Economic Dispatch (ED) Problem
P1 Objective function:
PD
C1 B T G C  C1  P1   C2  P2   ...  Cn  Pn    Ci  Pi 
n

i 1
P2
total fuel cost of the entire system
C2
B T G
Constraints:
n
PD  P1  P2  ...  Pn   Pi
Pn i 1

load / generation balance

C3 B T G (losses are neglected)

Pi ,min  Pi  Pi ,max
unit constraints
6

3
The Hong Kong Polytechnic University Economic Operation

3.1. Optimization Basics for ED Problem 
Gradient Search:
• Minimum of a function, f(x), can be found by a series of steps that always take us in
a downward direction.
• From any starting point, x0, we may find the direction of “steepest descent” by
noting that the gradient of f(x) always points in the direction of maximum ascent.
f   f x1 L f x n 
T

• Therefore, if we want to move in the direction of minimum descent, we negate the

gradient.
x1  x0   f

• Where α is a scalar to allow us to guarantee that the process converges (The best
value of α must be determined by experiment).
7

The Hong Kong Polytechnic University Economic Operation

Definitions:
• Since the objective is to maximize or minimize a mathematical function, we will call this
function the Objective Function.
• The constraint functions and simple variable limits will be lumped under the term
constraints.
• The region defined by the constraints is said to be the feasible region for the independent
variables.
• If the constraints are such that no such region exists, that is, there are no values for the
independent variables that satisfy all the constraints, then the problem is said to have an 
infeasible solution.
• When an optimum solution to a constrained optimization problem occurs at the boundary
of the feasible region defined by a constraint, we say the constraint is binding.
• If the optimum solution lies away from the boundary, the constraint is nonbinding.

4
 The Hong Kong Polytechnic University Economic Operation

Ex: simple elliptical objective function
Minimize:  f ( x1 , x2 )  0.25 x12  x22 Subject to the constraint:  ( x1 , x2 )  5  x1  x2  0

• At the optimum point, the gradient of f is

perpendicular (mathematicians say “normal”)
to ω and therefore there can be no
improvement in f by moving off this point.
• f and  are linearly dependent vectors.
• We can solve for this optimum point
mathematically by using this “normal”
property at the optimum:
f     0 Gradients near a constrained optimum
Lagrange multiplier: scaling variable > 0
• That is, the two gradients can be added together in such a way that they cancel each
other, as long as one of them is scaled
9

The Hong Kong Polytechnic University Economic Operation

• Instead of using the gradients, restate them as Lagrange equation:

L( x1, x2 ,  )  f ( x1, x2 )    ( x1, x2 )

 L x1  0
f      0  L x2  0
Necessary  L   0
gradient representation Conditions

 L x1  0.5 x1    0 x1  4
 L x2  2 x2    0 => x2  1
 L   5  x1  x2  0 2
10

5
The Hong Kong Polytechnic University Economic Operation

Optimization with Constraints:

f ( x)
i (x)  0 i  1, 2, ..., N  Binding constraints
g i (x)  0 i  1, 2, ..., Ng Not necessarily binding
x : vector of real numbers, dimension = N

Kuhn‐Tucker equations: complimentary slackness

N Ng
L(x, λ , μ)  f (x)   i i (x)   i gi (x) condition
Either g ( x )  0 or   0
0 0
i 1 i 1
i i

1. L(x0 , λ 0 , μ 0 )  0 for i  1, ..., N binding

2. i (x0 )  0 for i  1, ..., N 
i.e. binding if   0
0
i

3. gi (x 0 )  0 for i  1, ..., Ng
4. i0 gi (x0 )  0 for i  1, ..., Ng
i0  0 11

The Hong Kong Polytechnic University Economic Operation

Ex: simple elliptical objective function
Minimize:  f ( x1 , x2 )  0.25 x12  x22
 ( x1 , x2 )  5  x1  x2  0 g ( x1 , x2 )  x1  0.2 x2  3  0
Subject to:                                               &   

L   0.25 x12  x22     5  x1  x2     x1  0.2 x2  3

• First condition gives
 L x1  0.5 x1      0
 L x2  2 x2    0.2   0
• Second condition gives
5  x1  x2  0
• Third condition gives
x1  0.2 x2  3  0
• Forth condition gives
  x1  0.2 x2  3  0 ,   0
12

6
 The Hong Kong Polytechnic University Economic Operation

• let µ = 0, which implies that • let µ > 0, which implies that
g ( x1 , x2 )  x1  0.2 x2  3  0 g ( x1 , x2 )  x1  0.2 x2  3  0

by conditions 1 and 2 by conditions 2 and 3

x1  4 x1  2.5
x2  1 x2  2.5
2 by condition 1
but   5.9375
g ( x1 , x2 ) |x1  4  4  0.2  3  1.2  0
  4.6875
x2 1 and
g ( x1 , x2 ) |x1  2.5  2.5  0.5  3  0
x2  2.5

All conditions are met!
13

The Hong Kong Polytechnic University Economic Operation

3.2 ED Problem Solution with Lagrange Method
Recall ED Problem Objective function:
n

P1
C  C1  P1   C2  P2   ...  Cn  Pn    Ci  Pi 
PD i 1
C1 B T G
total fuel cost of the entire system

P2 Constraints:
C2
B T G n
PD  P1  P2  ...  Pn   Pi
i 1

load / generation balance

Pn (losses are neglected)
C3 B T G Pi ,min  Pi  Pi ,max
unit constraints
14

7
The Hong Kong Polytechnic University Economic Operation

Incremental cost, dCi/dPi
($/MWh)

First Necessary Condition:

linear approximation
L(x0 , λ 0 , μ 0 )  0 for i  1, ..., N

(MW)
dCi dPi   N equations Pi,min Unit Output, P Pi,max
i

Pi,min  Pi  Pi ,max 2N inequalities

($/h)

Unit operating cost, Ci
n
PD   Pi 1 constraint
i 1 ∆Ci
∆Pi

(MW)
Pi,min Unit Output, P Pi,max
i
15

The Hong Kong Polytechnic University Economic Operation

Example.1:
• Consider the following three generating units
1. H1  P1   510  7.2 P1  0.00142 P12 Mbtu/h 150 MW  P1  600 MW
2. H 2  P2   310  7.85 P2  0.00194 P2 2 Mbtu/h 100 MW  P2  400 MW
3. H 3  P3   78  7.97 P3  0.00482 P32 Mbtu/h  50 MW  P3  200 MW
• The fuel is coal in the first power plant (U1) and oil in the others (U2 and U3)
• Determine the economic operating point for these three units when delivering a total
of 850 MW where FC1 = 1.1 $/Mbtu & FC2 = FC3 = 1 $/Mbtu.
Cost functions:
C1  P1   1.1 x H1  P1   561  7.92 P1  0.001562 P12 R/h
C2  P2   1 x H 2  P2   310  7.85 P2  0.00194 P2 2 R/h
C3  P3   1 x H 3  P3   78  7.97 P3  0.00482 P32 R/h
16

8
 The Hong Kong Polytechnic University Economic Operation
Conditions for an optimum dispatch:
dC1 dP1  7.92  0.003124 P1     9.148 R / M Wh
dC2 dP2  7.85  0.00388 P2   P1  393.2 MW
dC3 dP3  7.97  0.00964 P3   P2  334.6 MW
P1  P2  P3  850 P3  122.2 MW
• Suppose the price of coal (i.e. FC1) decreased to 0.9 R/Mbtu.
  8.284 R / M Wh
P1  704.6 MW > 600 MW
dC1 dP1  6.48  0.00256 P1   P2  111.8 MW
P3  32.6 MW < 50 MW
Conditions for an optimum dispatch when P1 = 600 MW:
  8.576 R / M Wh
dC2 dP2  7.85  0.00388 P2  
P1  600 MW
dC3 dP3  7.97  0.00964 P3  
P2  187.1 MW
P2  P3  250
P3  62.9 MW 17

The Hong Kong Polytechnic University Economic Operation

3.3 ED Problem Solution with λ Iteration Method
dC1 dP1 dC 2 dP2 dC3 dP3

PD  P1  P2  P3
Graphical Solution
Solution with λ Iteration (λ is outside limit of G3) 18

9
The Hong Kong Polytechnic University Economic Operation

Consider Example‐1 where FC1 = 1.1 R/Mbtu & FC2 = FC3 = 1 R/Mbtu
Iteration‐1:
Set λ(0) = 8 7.92  0.003124 P1  8 P1  25.61 MW
7.85  0.00388 P2  8 P2  38.66 MW P1  P2  P3  67.38MW  850 MW
7.97  0.00964 P3  8 P3  3.11 MW
Iteration‐2:
Set λ(1) = 10 7.92  0.003124 P1  10 P1  600 MW
7.85  0.00388 P2  10 P2  554.12 MW P1  P2  P3  1274.7 MW  850 MW
7.97  0.00964 P3  10 P3  120.58 MW
Iteration‐3:
7.92  0.003124 P1  9 P1  345.71 MW
Set λ (2) = 9
(λ(0)+ λ(1))/2 7.85  0.00388P2  9 P2  296.39 MW P1  P2  P3  748.94MW  850 MW
7.97  0.00964 P3  9 P3  106.84 MW
19

The Hong Kong Polytechnic University Economic Operation

Iteration‐4:
Set λ (3) = 9.5 P1  505.76 MW
(λ(1)+ λ(2))/2 P2  452.26 MW P1  P2  P3  1089.32MW  850MW
P3  158.71 MW
Iteration‐5:
Set λ (4) = 9.25 P1  425.73 MW
(λ(2)+ λ(3))/2 P2  360.82 MW P1  P2  P3  919.33MW  850 MW
P3  132.78 MW
Iteration‐6:
Set λ (5) = 9.125 P1  385.72 MW
(λ(2)+ λ(4))/2 P2  328.61 MW P1  P2  P3  834.14 MW  850 MW
P3  119.81 MW

Continue Until P1  P2  P3  850 MW  

20

10
The Hong Kong Polytechnic University Economic Operation

21

The Hong Kong Polytechnic University Economic Operation

22

11
The Hong Kong Polytechnic University Economic Operation

• Large steam turbine generators have a number of

steam admission valves that are opened in sequence
to obtain ever‐increasing output of the unit. They
have discontinuous type of incremental heat rate
characteristic.
• This type of input‐output characteristic is nonconvex;
hence, we cannot use an equal incremental cost
methodology since there are multiple values of MW
output for any given value of incremental cost.
• Solution techniques include:
 Dynamic programming,
 Lagrangian relaxation,
 Mixed integer programming

2
3

The Hong Kong Polytechnic University Economic Operation

Gas Turbine
Pe
GT G
Generator
HRSG

Gas Turbine

GT G
Generator
HRSG
Steam Turbine

ST G
Generator

Combined Cycle Power Plant

(two Gas Turbines and a Steam Turbine)
2
4

12
The Hong Kong Polytechnic University Economic Operation

3.4 Economic Dispatch Including Transmission Losses
When transmission losses are included in the economic dispatch problem, the total load
demand equation becomes
 n 
PD    Pi   PL
 i 1 
where PL is the total transmission loss. In general, PL is not constant but depends on the 
unit outputs  Pi . The Lagrange function becomes
n
 n

L   Ci  Pi     PD  PL   Pi 
i 1  i 1 
Minimum of C is attained when 
dCi dPi   1  PL Pi   0 for i  1,..., n

 1  dC i dC  1  penalty factor
    Li i Li    for ith plant
 1  PL Pi  dPi dPi  1  PL Pi 
25

The Hong Kong Polytechnic University Economic Operation

Thus the optimum fuel economy is achieved when the product of the incremental
fuel times the penalty factor is the same for all plants. PL Pi is known as the
incremental transmission loss at plant i and λ is known as the incremental cost of
received power in $/MWh.

dC1 dC dC n
  L1  L2 2  ...  Ln
dP1 dP2 dPn

The above equations are known as the exact coordination equations because both
incremental fuel cost and incremental transmission loss are coordinated for best
economy.

26

13
The Hong Kong Polytechnic University Economic Operation

Focusing on the generator i and assuming that it undergoes an increase in generating 
Pi
power of amount        , which in turn manifests itself as an increased share of total load 
demand as well as associated line losses. That is

Pi  PD  PL

The loss penalty factor for this generator can be rewritten as

 1   1   Pi  Pi
Li      1
 1  PL Pi   1  PL Pi   Pi  PL  PD

Li  1 => generator is close to the loads and most of the increased generator power goes 
to supply the load demand.

Li  1 => generator is far away from loads and is required to supply an increased load 
demand in the presence of high line losses.
27

The Hong Kong Polytechnic University Economic Operation

Loss Formula Concept – B Coefficients:

The two generator units of this system are assumed to deliver power to a remotely 
located load with power factor pf1 and pf2. Three transmission line segments are 
involved with resistances of R1, R2 and R3.
Let the currents supplied by G1 and G2 to the 
load are I1 and I2 and are co‐phasal, the line 
losses PL is
PL  I1 R1  I 2 R2   I1  I 2 
2 2 2
R3
 R R  2  2 R3   R R  2
  21 3
2 1
P   P1 P2   2
2 3
2  2
P
V1  pf1    V1 V 2  pf 1  pf 2    V
 2  pf 2  

 B11 P1  B12 P1 P2  B22 P2
2 2

The B coefficients can be assumed constant if bus voltages and power factors remain 
constant. In practice, they are calculated for some average operating condition and can 
be used reliably for a fairly wide range around those conditions. 28

14
The Hong Kong Polytechnic University Economic Operation

Based on the experience with the simple case in last section, a general expression for 
the total system losses can be written, with several approximations, as
n n n
PL   PB
i ij Pj PL Pj   2 Bij Pj
i 1 j 1 i 1

Bij
where      is the loss coefficients which are constants under certain assumed operating 
Bij  B ji
conditions and               . Assuming quadratic plant cost curves as
Ci  B2i Pi 2  B1i Pi  B0i

The incremental cost is: dCi dPi  2 B2i P  B1i

PL Pj dCi dPi dCi dPi   1  PL Pi   0
Substituting                  and                  in                                                     and collecting all 
Pi
terms of     ,

Pi Pi
For any particular value of λ,     can be solved iteratively by assuming initial values of      .     
A convenient choice is  Pi  0 .
29

The Hong Kong Polytechnic University Economic Operation

Consider Example‐1 where FC1 = 1.1 $/Mbtu & FC2 = FC3 = 1 $/Mbtu
PL  0.00003P12  0.00009 P22  0.00012 P32
7.92  0.003124 P1   1  2  0.00003 P1 
dCi dPi   1  PL Pi   0  7.85  0.00388 P2   1  2  0.00009  P2 
7.97  0.00964 P3   1  2  0.00012  P3 
P1  P2  P3  850  PL  0

Step‐1: Pick a set of starting values for P1 , P2 , P3 that sum to the load i.e. P1  P2  P3  850 MW
Step‐2: Calculate the incremental losses PL Pi and the total losses PL. The incremental
losses and total losses will be considered constant until we return to Step‐2.
Step‐3: Calculate the value of λ that causes  P1 , P2 , P3 to sum to the total load plus losses.
Step‐4: Compare the  P1 , P2 , P3 from Step‐3 to the values used at the start of Step‐2. If 
there is no significant change in any one of the values, go to STOP, otherwise go back to 
Step‐2.
30

15
 The Hong Kong Polytechnic University Economic Operation

Step‐1: P1  400 MW
P2  300 MW P1  P2  P3  850 MW
P3  150 MW
Iteration‐1:
Step‐2: PL P1  2  0.00003  400  0.024
PL P2  2  0.00009  300  0.054 PL  15.6 MW
PL P3  2  0.00012  150  0.036

Step‐3:
7.92  0.003124 P1   1  0.024 P1  440.68MW
7.85  0.00388 P2   1  0.054  P2  299.12 MW
7.97  0.00964 P3   1  0.036 P3  125.77 MW
  9.5252
P1  P2  P3  865.6  0
Step‐4: max(P1 , P2 , P3 ) > ε, go to Step‐2
31

The Hong Kong Polytechnic University Economic Operation

Iteration‐2:
Step‐2: PL P1  2  0.00003  440.68  0.0264
PL P2  2  0.00009  299.12  0.0538 PL  15.78MW
PL P3  2  0.00012  125.77  0.0301
Step‐3:
7.92  0.003124 P1   1  0.0264 P1  433.94 MW
P2  300.11MW
7.85  0.00388 P2   1  0.0538 
P3  131.74 MW
7.97  0.00964 P3   1  0.0301
  9.5275
P1  P2  P3  865.78  0
Step‐4: max(P1 , P2 , P3 ) > ε, go to Step‐2

Iteration‐3:
Continue Until max(P1 , P2 , P3 )  
32

16
The Hong Kong Polytechnic University Economic Operation

3.5 Base Point and Participation Factors
This method assumes that the economic dispatch problem has to be solved repeatedly by 
moving the generators from one economically optimum schedule to another as the load 
changes by a reasonably small amount. We start from a given schedule‐the base point. 
Next, the scheduler assumes a load change and investigates how much each generating 
unit needs to be moved (i.e., “participate” in the load change) in order that the new load 
be served at the most economic operating point.
i    Ci Pi 0  Pi

Incremental cost, dCi/dPi
($/MWh)

n n
1
PD   Pi   
i 1 Ci Pi 
0 ∆λ
i 1 λ0

1 Ci Pi 0 
∆Pi
Pi
 n Pi0 (MW)
PD
1 Ci Pi 0 
i 1
Unit Output, Pi
Pi,max

33

The Hong Kong Polytechnic University Economic Operation

Consider Example‐1 where FC1 = 1.1 R/Mbtu & FC2 = FC3 = 1 R/Mbtu
Starting from the optimal economic solution found in for 850 MW, use the
participation factor method to calculate the dispatch for a total load of 900 MW.

P1 1 C1 P10   0.003124 

1

   0.47
PD 1 C1 P10   1 C2  P20   1 C3  P30   0.003124 1   0.00388 1   0.00964 1

P2 P3
 0.38 ,  0.15
PD PD

Pnew,1  393.2  0.47  50  416.7

 P 
Pnew,i  Pbase,i   i  PD  Pnew,2  334.6  0.38  50  353.6
 PD 
Pnew,3  122.2  0.15  50  129.7

34

17
The Hong Kong Polytechnic University Economic Operation

3.6 Interconnection for Interchange of Power
Power plants are connected together through tie‐line (transmission line) to
form a large interconnected system due to the following advantages.
1. Increased Reliability ‐ extra spinning reserve and redundant power paths
– The loss of generator or transmission line can be made up from spinning reserve
among generators throughout the interconnection or alternative transmission
paths.
2. Cost Reduction ‐ power interchange
– Power is bought from the lower cost generator and hence the cost of generation is
reduced
– Regional spinning reserve is reduced by relying on system reserve, thus saving
operation cost
3. Improved Regulation
– Changes of load can be catered by all generating units within the whole system and
not the generators in the region only 35

The Hong Kong Polytechnic University Economic Operation

Consider two power systems operating with different incremental costs (IC).
– Utility A is generating at a lower IC than utility B.
– If utility B buys power from utility A for its load at a price less than the cost of
generation by itself, utility B would save money.
– Utility A would benefit economically from selling power to utility B as long as utility
B is willing to pay a price greater than utility A’s cost of generation.
Steps for interchange evaluation:
1. Assume no power interchange, calculate the economic dispatch (ED) for each system
alone.
2. Determine which system has the lower IC. The system with lower IC runs a series of
ED, each having a greater demand. The system with higher IC runs a series of ED, each
having a lower demand.
3. Find out at which level of interchange energy will bring the two systems towards the
most economic operation
36

18
The Hong Kong Polytechnic University Economic Operation

37

The Hong Kong Polytechnic University Economic Operation

38

19
 The Hong Kong Polytechnic University Economic Operation

39

The Hong Kong Polytechnic University Economic Operation

4. Unit Commitment
• To ‘commit’ a generating unit is to ‘turn it on’; that is, to bring the unit up to
speed, synchronize it to the system, and connect it so it can deliver power to the
network.
• The problem with ‘commit enough units and leave them on line’ is one of
economics. A great deal of money can be saved by turning units off (decommitting
them) when they are not needed.
• Since the load varies continuously with time, the optimum combination of units
may alter during any period. To determine which units of a plant should operate
for a given load is the problem of unit commitment (plant ordering).
• Every electricity supply undertaking is normally under obligation to provide power
to its consumers with some degree of reliability. Therefore, it is necessary to
coordinate both the economy and reliability (security) of unit commitment.

40

20
The Hong Kong Polytechnic University Economic Operation

Example‐U1:

41

The Hong Kong Polytechnic University Economic Operation

42

21
 The Hong Kong Polytechnic University Economic Operation

Observations on the example:

• Far too few units committed:
 Can’t meet the demand
• Not enough units committed:
 Some units operate above optimum
• Too many units committed:
 Some units operate below optimum
• Far too few units committed:
 Minimum generation exceeds demand
• No‐load cost affects choice of optimal combination

43

The Hong Kong Polytechnic University Economic Operation
Load
• Optimal generation schedule for a load profile
• Decompose the profile into a set of period 1000

• Assume load is constant over each period

• For each time period, which units should be 500
committed to generate at minimum cost during that
period? Time
0 6 12 18 24

Load Unit 3
Load Unit 1 Unit 2 Unit 3
1100 On On On
1000 On On Off 1000

900 On On Off Unit 2

800 On On Off
500
700 On On Off Unit 1
600 On Off Off
500 On Off Off Time
0 6 12 18 24
44

22
The Hong Kong Polytechnic University Economic Operation

Issues:
• Must consider constraints
– Unit constraints
– System constraints
• Some constraints create a link between periods
• Start‐up Costs
– Cost incurred when we start a generating unit
– Different units have different start‐up costs
• Curse of dimensionality

45

The Hong Kong Polytechnic University Economic Operation

4.1 Unit Constraints
• Constraints that affect each unit individually:
• Maximum generating capacity
• Minimum stable generation
• Minimum “up time”
• Minimum “down time”
• Ramp rate
Notations:
u(i,t) : Status of unit i at period t
u(i,t)  1 : Unit i is on during period t
u(i,t)  0 : Unit i is off during period t
P(i, t ) : Power produced by unit i during period t
46

23
The Hong Kong Polytechnic University Economic Operation

• Minimum up time
– Once a unit is running it may not be shut down immediately:
If u(i,t)  1 and tiup  tiup,min then u(i,t  1)  1

• Minimum down time
– Once a unit is shut down, it may not be started  immediately
If u(i,t)  0 and tidown  tidown,min then u(i,t  1)  0
• Maximum ramp rates
– Due to boiler constraints, the electrical output of a unit cannot change by more than a
certain amount over a period of timeOnce a unit is shut down, it may not be started
immediately
Maximum ramp up rate constraint: P  i, t  1  P  i, t   Pi up ,max
Maximum ramp down rate constraint: P(i, t )  P(i, t  1)  Pi down ,max
• Crew constraints – if a plant consists of two or more units, they cannot both be turned
on at the same time since there are not enough crew members at attend both units
while starting up
47

The Hong Kong Polytechnic University Economic Operation

4.2 System Constraints
Many constraints can be placed on the unit commitment problem. Each individual 
power system, power pool, reliability council and so forth may impose different rules 
on the scheduling of units, depending on the generation makeup, load‐curve 
characteristics and such
1. Spinning Reserve
– Unanticipated loss of a generating unit or an interconnection causes
unacceptable frequency drop if not corrected rapidly.
– Need to increase production from other units to keep frequency drop within
acceptable limits.
– Rapid increase in production only possible if committed units are not all
operating at their maximum capacity.
N

 u (i, t )P i
max
 PLoad (t )  PSR (t )
i 1

PSR (t ): Spinning Reserve requirement at time t

48

24
The Hong Kong Polytechnic University Economic Operation

How much Reserve?
• Protect the system against “credible outages”
• Deterministic criteria:
– Capacity of largest unit or interconnection
– Percentage of peak load
• Probabilistic criteria:
– Takes into account the number and size of the committed units as well as their outage rate
Types of Reserve:
• Spinning reserve
– Primary: Quick response for a short time
– Secondary: Slower response for a longer time
• Tertiary reserve (scheduled or off‐line reserve)
– Replace primary and secondary reserve to protect against another outage
– Provided by units that can start quickly (e.g. open cycle gas turbines)
Reserve must be spread around the network, i.e. must be able to deploy reserve even if 
the network is congested
49

The Hong Kong Polytechnic University Economic Operation

Cost of Reserve
• Reserve has a cost even when it is not called
• More units scheduled than required
– Units not operated at their maximum efficiency
– Extra start up costs
• Must build units capable of rapid response
• Cost of reserve proportionally larger in small systems
– Important driver for the creation of interconnections between systems

50

25
The Hong Kong Polytechnic University Economic Operation

• With the exception of unit 4, the

loss of any unit on this system can
be covered by the spinning reserve
on the remaining units.
• If unit 4 were to be lost and unit 5
were to be run to its maximum of
600 MW, the eastern region would
still need 590 MW to cover the load
in that region.
• The western region can supply 590
MW from its reserves. However, the
tie capacity of only 550 MW limits
the transfer.

51

The Hong Kong Polytechnic University Economic Operation

2. Environmental constraints
• Scheduling of generating units may be affected by environmental constraints.
• Constraints on pollutants such SO2, NOx.
– Various forms:
 Limit on each plant at each hour
 Limit on plant over a year
 Limit on a group of plants over a year
• Constraints on hydro generation:
– Protection of wildlife
– Navigation, recreation

52

26
 The Hong Kong Polytechnic University Economic Operation

3. Network constraints
Transmission network may have an effect on the commitment of units.
– Some units must run to provide voltage support.
– The output of some units may be limited because their output would exceed the
transmission capacity of the network.
A B

Cheap generators More expensive generator

May be “constrained off” May be “constrained on”

4. Fuel constraints
Some units have limited fuel (for example due to late delivery, bad weather, high price,
forecast error, etc.) or else have constraints that require them to burn a specified
amount of fuel in a given time (for example due to contractual reasons).
53

The Hong Kong Polytechnic University Economic Operation

4.3. Start‐up Costs 
• Thermal units must be “warmed up” before they can be brought on‐line. Start‐up cost depends on
time unit has been off.
• There are 2 approaches to treating a thermal unit during its down period. The first is referred as
cooling. It allows the unit’s boiler to cool down and then heat back up when recommitted. The
second called banking requires that sufficient energy be input to the boiler to just maintain
operating temperature. The costs for the two can be formulated as

54

27
The Hong Kong Polytechnic University Economic Operation

4.4 Unit Commitment Problem
• Decision variables:
– Status of each unit at each period:
u(i,t) 0,1  i,t
– Output of each unit at each period:
 
P(i, t )  0,  Pi min ; Pi max   i, t
• Combination of integer and continuous variables
– Continuous variables
– Can follow the gradients or use LP
– Any value within the feasible set is OK
– Discrete variables
– There is no gradient
– Can only take a finite number of values
– Problem is not convex
– Must try combinations of discrete values 55

The Hong Kong Polytechnic University Economic Operation

How many combinations? How many solutions?
• Examples
• 3 units: 8 possible states
• N units: 2N possible states

111

110

101

100

011 T= 1 2 3 4 5 6

010  Optimization over a time horizon divided into intervals
001  A solution is a path linking one combination at each
interval
000
 There are 2NT paths
56

28
The Hong Kong Polytechnic University Economic Operation

• Example: 5 units, 24 hours

   
N T 5 24 35
2  2  6.2 10 combinations

• Processing 109 combinations/second, this would take 1.9 1019 years to solve
• There are 100’s of units in large power systems!
• Many of these combinations do not satisfy the constraints

Brute force approach wonʼt work!

• Need to be smart
• Try only a small subset of all combinations
• Canʼt guarantee optimality of the solution
• Try to get as close as possible within a reasonable amount of time
57

The Hong Kong Polytechnic University Economic Operation

4.5 Unit Commitment Problem Solution
• Characteristics of a good technique
– Solution close to the optimum
– Reasonable computing time
– Ability to model constraints

• Priority list / heuristic approach
• Dynamic programming
• Lagrangian relaxation
• Mixed Integer Programming

58

29
The Hong Kong Polytechnic University Economic Operation

Example‐U2: Unit Data
Min Min No-load Marginal Start-up
Pmin Pmax Initial
Unit up down cost cost cost
(MW) (MW) status
(h) (h) ($) ($/MWh) ($)
A 150 250 3 3 0 10 1,000 ON
B 50 100 2 1 0 12 600 OFF
C 10 50 1 1 0 20 100 OFF

Demand Data
Hourly Demand
350
300
250 Reserve requirements are not considered!
Load 200
150
100
50
0
1 2 3
Hours
59

The Hong Kong Polytechnic University Economic Operation

Feasible Unit Combinations (States)

Combinations 1 2 3
Pmin Pmax
A B C 150 300 200
1 1 1 210 400
1 1 0 200 350
1 0 1 160 300
1 0 0 150 250
0 1 1 60 150
0 1 0 50 100
0 0 1 10 50
0 0 0 0 0
60

30
The Hong Kong Polytechnic University Economic Operation

Transitions between feasible combinations
Combinations 1 2 3
Pmin Pmax
A B C
1 1 1 210 400
1 1 0 200 350
1 0 1 160 300
1 0 0 150 250 Initial State

0 1 1 60 150
0 1 0 50 100
0 0 1 10 50
0 0 0 0 0
61

The Hong Kong Polytechnic University Economic Operation

Infeasible transitions: Minimum down time of unit A
Combinations 1 2 3
Pmin Pmax
A B C
1 1 1 210 400
1 1 0 200 350
1 0 1 160 300
1 0 0 150 250 Initial State

0 1 1 60 150 TD TU
0 1 0 50 100 A 3 3
0 0 1 10 50 B 1 2
0 0 0 0 0
C 1 1
62

31
The Hong Kong Polytechnic University Economic Operation

Infeasible transitions: Minimum up time of unit B
Combinations 1 2 3
Pmin Pmax
A B C
1 1 1 210 400
1 1 0 200 350
1 0 1 160 300
1 0 0 150 250 Initial State

0 1 1 60 150 TD TU
0 1 0 50 100 A 3 3
0 0 1 10 50 B 1 2
0 0 0 0 0
C 1 1
63

The Hong Kong Polytechnic University Economic Operation

Operating Costs

Economic Dispatch
State Load PA PB PC Cost
1 150 150 0 0 1500
2 300 250 0 50 3500 4
3 300 250 50 0 3100 $3200
4 300 240 50 10 3200
3 7
5 200 200 0 0 2000 $3100 $2100
6 200 190 0 10 2100
2 6
7 200 150 50 0 2100 $2100
$3500

1 5
$1500 $2000
Initial State

64

32
The Hong Kong Polytechnic University Economic Operation

Accumulated Costs
$5400
1 1 1 4
$3200 • This example is intended to
$0 illustrate the principles of unit
$5200 $7300
1 1 0 $700 3 $0 7 commitment
$3100 $600 $2100 • Some constraints have been
$600 $5100 $7200 ignored and others artificially
1 0 1 2 $0 6 tightened to simplify the
$3500 $2100 problem and make it solvable
$100
$0 by hand
$1500 $7100
1 0 0 $0 1 5 • Due to simplifications, it does
$1500 $2000 not illustrate the true
complexity of the problem
• The solution method used in
Unit Start-up cost Lowest total cost this example is based on
A 1000 dynamic programming

B 600
C 100
65

The Hong Kong Polytechnic University Economic Operation

4.6 Priority List
For a optimised system, units must be shut down as
the load goes down and then recommitted as it goes
back up. A ‘shutdown rule’ can be obtained for
Example‐U1 simply by a brute‐force technique
wherein all combinations of units will be tried for each
load value taken in steps of 50 MW

66

33
The Hong Kong Polytechnic University Economic Operation

The shut‐down rule is quite simple: When load is above 1000 MW, run all 3 units; between
1000 MW and 600 MW, run units 1 and 2; below 600 MW, run only unit 1. Alternatively, this
rule can be expressed as a priority list of units: 1, 2, 3. It shows that unit 1 has the top priority
to be committed while unit 3 has the lowest priority. As the above priority list is derived by
brute‐force, it is very time consuming for a large system. Simpler methods for the production
of approximate priority lists have been used by utilities. For the above example a simpler
scheme is outlined below:
First derive the ‘full load average production cost’ (FLAPC) table for all units as follows and
then a commitment scheme based on FLAPC can be drawn

The priority order based on FLAPC is 2, 1, 3.
Now the above merit order loading scheme is used.
67

The Hong Kong Polytechnic University Economic Operation

4.7 Shut‐Down Algorithm
The following simple shut‐down algorithm can be used together with the priority list
a) Will dropping the next unit leave enough generation on‐line to meet demand
plus reserve requirements? If NO, operate as at present, if YES, go to (b).
b) Determine when the unit will be needed next (when load rises), say H hours
later.
c) If this is less than the minimum shut down time, keep the unit on‐line, if not
go to (d).
d) Calculate two costs. First is cost for next H hours with unit on‐line; second is
without unit (to choose between banking and cooling depending on cost). If
there is a saving, shut down the unit; otherwise, operate as at present.
e) Repeat this entire procedure for the each unit on the priority list.

68

34
The Hong Kong Polytechnic University Economic Operation

Example‐U3:

69

The Hong Kong Polytechnic University Economic Operation

70

35
The Hong Kong Polytechnic University Economic Operation

71

The Hong Kong Polytechnic University Economic Operation

72

36
The Hong Kong Polytechnic University Economic Operation

4.8 Other Types of Units
The systems treated so far have included only fossil‐fuel generating units. In practice,
however, systems have a mix of different types of units including fossil‐fuel, nuclear,
pumped‐storage hydro, hydro and other types such as wind generators, diesel and gas‐
turbine units, etc.
1. Nuclear Units
• Although the fixed costs of a nuclear unit may be high, their operating cost are low
due to inexpensive nuclear fuel.
• As such, nuclear units are normally base‐loaded at their rated outputs. That is, the
reference power settings of turbine‐governors for nuclear units are held constant at
rated output.
• Nuclear units, therefore, become large in size and they are committed to supply only
base loads and do not normally participate in economic dispatch.

73

The Hong Kong Polytechnic University Economic Operation

2. Pumped‐Storage Hydro Units
• Pumped‐storage hydro is a form of energy storage and is the only practical method
of storing electric energy on a large scale.
• During off‐peak hours these units are operated as sychronous motors to pump
water to a higher elevation. Then during peak‐load hours the water is released and
the units are operated as synchronous generators to supply power.
• Economic operation of the power system is improved by pumping during off‐peak
hours when the system λ (generation cost) is low, and by generating during peak‐
load hours when λ is high. However, coordinations are needed for incorporating
pumped‐storage into economic dispatch of fossil‐fuel units.
• Apart from demand regulation, a large proportion of the spinning reserve is
commonly put on the pumped‐storage hydro units because of its fast pick‐up. As a
results, pumped‐storage plant are often not included in economic dispatch analysis.

74

37
The Hong Kong Polytechnic University Economic Operation

3. Hydro Units
• The input‐output relation of a hydro unit is a function of both discharge rate and
effective head.
• Hydro units do not have any fuel cost.
• They can follow rapid load changes more easily than steam units.
• Further, hydro units can be committed within minutes while the commitment of
steam units requires several hours.
• The last two features make hydro plants specially suitable for cold reserve and for
working as peaking and regulating plants.
• However, while the fuel‐supply is not a limiting factor in the operation of thermal
plants, the operation of hydro‐electric plants is dependent on the availability of
water flow.
• In other words, energy output (MWh) is fixed and the MW dispatch is regulated by
adjusting the operating hours in the case of hydro‐electric plants.
75

38