Stability of Power Systems

An introduction

Mehrdad Ghandhari

Royal Institute of Technology (KTH)

Electric Power and Energy Systems
2018
ii
Contents

1 Introduction 1

2 Mathematical modeling 7
2.1 Definitions and notations . . . . . . . . . . . . . . . . . . . . . . . . . . 7

3 Synchronous machines modeling 17

3.1 Swing equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
3.2 Electrical equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
3.3 Steady-state model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
3.4 Transient-state model . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
3.4.1 Classical model . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
3.4.2 One-axis model . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.4.3 Two-axis model . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
3.5 Per unit conversion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
3.6 Excitation systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
3.6.1 One-axis model with AVR and PSS . . . . . . . . . . . . . . . . 43

4 Transient stability of an SMIB system 47

4.1 Analysis of the swing equation . . . . . . . . . . . . . . . . . . . . . . . 48
4.1.1 Stability of the equilibrium points of the SMIB system . . . . . 49
4.2 Equal Area Criterion (EAC) . . . . . . . . . . . . . . . . . . . . . . . . 58
4.3 Transient Energy Function (TEF) method . . . . . . . . . . . . . . . . 67
4.4 Transient stability enhancement . . . . . . . . . . . . . . . . . . . . . . 71

iii
iv

5 Dynamic modeling of multi-machine power systems 73

5.1 Load modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
5.1.1 Static load model . . . . . . . . . . . . . . . . . . . . . . . . . . 74
5.1.2 Dynamic load model . . . . . . . . . . . . . . . . . . . . . . . . 75
5.2 Multi-machine power systems . . . . . . . . . . . . . . . . . . . . . . . 77
5.2.1 Structure preserving model . . . . . . . . . . . . . . . . . . . . . 77
5.2.2 Reduced Network Model . . . . . . . . . . . . . . . . . . . . . . 79

6 Rotor angle stability 91

6.1 Transient stability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
6.1.1 Transient Energy Function (TEF) method . . . . . . . . . . . . 92
6.1.2 SIngle Machine Equivalent (SIME) method . . . . . . . . . . . . 95
6.2 Small-signal stability . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
6.2.1 Modal analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
6.2.2 Small signal stability of power systems . . . . . . . . . . . . . . 103

7 Voltage stability 123

7.1 Voltage stability analysis . . . . . . . . . . . . . . . . . . . . . . . . . . 123
7.2 Voltage instability mechanisms and prevention . . . . . . . . . . . . . . 129
7.2.1 Fast voltage instability . . . . . . . . . . . . . . . . . . . . . . . 129
7.2.2 Slow voltage instability . . . . . . . . . . . . . . . . . . . . . . . 131
7.3 Prevention of voltage instability . . . . . . . . . . . . . . . . . . . . . . 134

8 Frequency control 143

8.1 System model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144
8.2 Turbine and turbine governor . . . . . . . . . . . . . . . . . . . . . . . 146
8.2.1 Hydro turbine model . . . . . . . . . . . . . . . . . . . . . . . . 146
8.2.2 Hydro turbine governor model . . . . . . . . . . . . . . . . . . . 150

A Load flow calculations based on Newton-Raphson method 163

A.1 Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163
 v

A.2 Application to power systems . . . . . . . . . . . . . . . . . . . . . . . 167

A.2.1 Newton-Raphson method for solving power flow
equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 172
A.3 The system data of the 3-generator system . . . . . . . . . . . . . . . . 173

B 175
B.1 Estimation of the stability region . . . . . . . . . . . . . . . . . . . . . 175
B.2 The Jacobian matrices of the linearized system . . . . . . . . . . . . . . 175
vi
Chapter 1
Introduction
This compendium should be considered as an introduction to power system stability and
control. A more detailed and comprehensive discussion and presentation may be found
in the references given in this compendium.

The function of an electric power system is to produce electricity and to transmit

it to customers. Thus, an electric power system may be divided into three parts, one
concerned with generation, one with transmission, and one with consumption as shown
in Figure 1.1. Electricity is produced at lower voltage at a generator. The generator
then feeds its electric power into the transmission system via a step-up transformer in
order to increase the voltage from the generation level (10 to 25 kV) to the transmission
level (hundreds of kV). Then at the load center, it is stepped-down to lower voltages
for distribution to customers.

Customer
Generating Transmission
Station System
(Load)

Generator step-up Substation step-down

transformer transformer

Figure 1.1. Basic structure of an electric power system.

Since modern society is strongly dependent on electricity, high reliability of supply and
high level of system security are of fundamental importance. However, in an electric
power system electricity cannot easily and economically be stored in large quantities.
This special property implies that electricity must be produced the instant it is used
that is the produced power must be in balance with the consumed power. Furthermore,
power systems are frequently subjected to various types of disturbances which may be
small, in the form of load changes and control actions, or large in the form of a short
circuit on a transmission line or loss of a large generator. The system must however
be able to adjust to the changing conditions and operate satisfactorily despite these
disturbances. Thus, to keep the high reliability of supply and high level of system
security will be a challenge for system operators.
By using computing and analysis tools, power system security analysis is a major
activity for power system operators to determine the robustness of the power system
to the occurrence of certain disturbances. This analysis normally concerns:

• Static security analysis which deals with operating and engineering constraints

1
2

such as overloading of transmission lines, transformers and other equipments, and

also bus voltage magnitudes in a post-disturbance state.

• Dynamic security analysis which deals with power system dynamic response
to disturbances. For instance, after a disturbance whether the system will survive
the ensuing transient and move into a secure state.

Therefore, the operation of a power system must be subject to security and reliable
standards developed by the system operator. A main principle underlying these stan-
dards is the so called N-1 criterion.

Definition 1.1 The N-1 criterion states that the power system must be operated at
all times such that after an unplanned loss of an important generator or transmission
facility it will remain in a secure state.
Furthermore, when a loss occurs the system must be returned to a new N-1 secure state
within a specified time (normally within 15-20 minutes) to withstand a possible new
loss.

The modern (interconnected) electric power systems are often considered as the most
complex man-made dynamical systems for use in daily life (far away as just “two
holes in the wall”). They consist of many individual elements connected together
to form a large, complex system capable of generating, transmitting and distributing
electrical energy over a large geographical area. Because of this interconnection of
elements, a large variety of dynamic interactions are possible, some of which will only
affect some of elements or parts of the system, while others may affect the system as a
whole. Therefore, it is crucially necessary to have knowledge about how a power system
behaves (or responds) to different disturbances, and also to understand the dynamics
of the system.
Power system dynamics are mainly initiated by a disturbance, and they occur in dif-
ferent time scales. Based on their physical character these dynamics may be fast,
relatively slow or very slow. The fast dynamics (known as electromagnetic dynamics)
are those associated with fast electromagnetic changes in the electrical machines or
operation of the protection system. The time frame of these fast dynamics is from
milliseconds to a second. The relatively slow dynamics (known as electromechanical
dynamics) are those associated with the oscillation of the rotating masses of the gen-
erators and motors. The time frame of the electromechanical dynamics is from around
a second to several seconds. Finally, the very slow dynamics are those associated with
the system frequency control (due to the turbine governing systems), and the ther-
modynamic changes (due to boiler control action in the steam power plants). In this
compendium, the main focus is on the electromechanical dynamics.
In general, mathematical models are used to analyze dynamic behavior of power sys-
tems. One interesting feature of classifying the power system dynamics, as introduced
above, is that it will be easier to derive an appropriate mathematical model including
components relevant to the specific dynamic (particularly due to the time frame of
 3

the specific dynamic). For instance, a disturbance may eventually excite all of the
above mentioned dynamics in a power system, but for the same disturbance we may
have different mathematical models to adequately describe the specific dynamic. If a
mathematical model which describes the electromagnetic dynamic will also be included
in the mathematical model of the electromechanical dynamic, then the model will ob-
viously result in unnecessary high computation time, and it will not give any useful
information when only the electromechanical dynamic is of concern.
Since the main focus of this compendium is on the electromechanical dynamic, only
the mathematical models of this dynamic will be derived to be able to study and
understand the concept of power system stability.
Power system stability is of fundamental importance concerning system security, and
it has been defined in many different ways. However, in this compendium we use the
definitions presented by IEEE/CIGRE Joint Task Force in [1].

Definition 1.2 Power system stability is the ability of an electric power system, for
a given initial operating condition, to regain a state of operating equilibrium after be-
ing subjected to a physical disturbance, with most system variables bounded so that
practically the entire system remains intact [1].

To facilitate analysis of stability, power system stability has been classified into different
categories in [1] as follows:

• Rotor angle stability refers to the ability of synchronous machines of an in-

terconnected power system to remain in synchronism after being subjected to
a disturbance. Instability that may result occurs in the form of increasing an-
gular swings of some generators leading to their loss of synchronism with other
generators. Loss of synchronism can occur between one machine and the rest of
the system, or between groups of machines, with synchronism maintained within
each group after separating from each other.
Rotor angle stability may be characterized as follows:

– Small-signal stability which is concerned with the ability of the power

system to maintain synchronism under small disturbances. The disturbances
are considered to be sufficiently small that linearization of system equations
is permissible for purposes of analysis.
– Transient stability which is concerned with the ability of the power system
to maintain synchronism when subjected to a large disturbance, such as a
short–circuit on a transmission line. Transient stability depends on the
initial operating conditions of the system as well as the type, severity and
location of the disturbance.

• Voltage stability refers to the ability of a power system to maintain steady

voltages at all buses in the system after being subjected to a disturbance from a
4

given initial operating condition. Instability that may result occurs in the form
of a progressive fall or rise of voltages of some buses.
Depending on the time scale the voltage instability may be classified as fast (a
couple of seconds) or slow (tens of seconds to minutes).

• Frequency stability refers to the ability of a power system to maintain steady

frequency following a severe system upset resulting in a significant imbalance
between generation and load. It depends on the ability to maintain/restore equi-
librium between system generation and load, with minimum unintentional loss of
load. Instability that may result occurs in the form of sustained frequency swings
leading to tripping of generating units and/or loads.

Since mathematical models are used to describe power system dynamics, power system
stability will also be defined based on the mathematical theory concerning stability
of dynamical systems. In the evaluation of stability the concern is to the dynamic
behavior of the power system when subjected to a disturbance.
Figure 1.2 shows the basic structure of a power generating unit. In this compendium,
the dynamics and mathematical models of the components shown in the figure will be
presented.

U ref

Exciter AVR
Ef
U
Valve/gate U
Power
Turbine
Water system
I
or Shaft
steam f
Governor

Pmset

Figure 1.2. Block diagram of a power generating unit.

Three mathematical models to describe the dynamics of a synchronous generator will

be presented. Those models are referred to as classical model, one-axis model and two-
axis model. In the classical model, the generator is represented by a voltage source
with constant magnitude behind a reactance. However, in the two other models the
magnitude of the voltage is considered as a state variable and also a function of the
field voltage Ef . This voltage can be controlled by the generator excitation system
which consists of an exciter and an Automatic Voltage Regulator (AVR). The primary
function of an exciter is to provide a dc source for field excitation of a synchronous
generator, and the AVR controls the excitation voltage. A control on excitation voltage
 5

results in controlling the field current (or Ef ) which in turn controls the terminal voltage
U and reactive power.
To get an insight into power system stability, the power system indicated in Figure 1.2
will be represented by an equivalent model to analyze the behavior a generator con-
nected to an infinite bus. This system is known as Single Machine Infinite Bus (SMIB)
system based on which transient stability will be discussed, and also the concept of
Equal Area Criterion will be presented.
Then, power system stability analysis will be applied to a general multi-machine power
system. Due to the load models (which will also be discussed) the dynamic of a multi-
machine power system may be described by a set of differential-algebraic equations
or only by a set of differential equations. The first model is refereed to as Structure
Preserving Model (SPM), and the second one is referred to as Reduced Network Model
(RNM).
For studying transient stability of a multi-machine power system, there are several
methods two of which will be briefly discussed in this compendium. The first method
is known as Transient Energy Function (TEF) method (termed also the direct method)
which is based on Lyapunov’s direct method. The second method is known as SIn-
gle Machine Equivalent (SIME) method which transforms the trajectories of a multi-
machine power system into the trajectory of a single machine equivalent system. Then,
by refreshing the parameters of the single machine equivalent system at each integra-
tion time–step, the SIME method numerically assesses the transient stability of this
equivalent system based on the Equal Area Criterion.
Small-signal stability analysis is also discussed in this compendium. This analysis deals
with small disturbances, and it is applied to linearized system models. It provides
valuable information about the inherent dynamic characteristics of the power system.
Moreover, by this analysis, the excitation system can be designed to enhance the small-
signal stability in a power system.
Finally, the dynamic of a hydro turbine and turbine governor will be discussed. The
objective of a turbine governing system installed in a generating unit is to produce a
desired power which is partly determined by the set value for the produced power and
partly by a contribution originating from the frequency control.
6
Chapter 2
Mathematical modeling
For being able to understand and analyze dynamic of a system, normally mathematical
models are derived to describe the dynamic of the system. Fortunately, for mechanical
and electrical systems (such as power systems) there are laws of nature which are basis
for deriving these mathematical models. For example, Kirchhoff’s laws, Newton’s laws,
Ohm’s law, induction law and etc. Applying these laws, the dynamic of the system
can then be described by a set of differential equations of the form
d
x(t) = ẋ = f (x) (2.1)
dt
where x is the nx –dimensional state vector belongs to the Euclidean space Rnx , and
f is a vector-valued function f : D → Rnx which is continuous and has continuous
first-order partial derivatives with respect to x on a domain D ⊂ Rnx into Rnx . The
solution to (2.1) is designated by x(t) with x(to ) as an initial state at initial time t = to .

2.1 Definitions and notations

Definition 2.1 Any xo for which ẋ = f (xo ) = 0, is termed equilibrium point (e.p).

Definition 2.2 A system is in a steady-state, if the system is settled to an e.p, i.e

ẋ = f (xo ) = 0.

Definition 2.3 A system is in a dynamic-state if ẋ = f (x) 6= 0.

One of the important issues regarding dynamical systems is to characterize and study
stability of equilibrium points since stability theory plays a central role in systems
theory and engineering. Stability of equilibrium points is usually characterized in the
sense of Lyapunov, a Russian mathematician and engineer (1857-1918) who laid the
foundation of the stability theory which now carries his name.
Three concepts are introduced to characterize stability of an equilibrium point, namely:

• Stability in the sense of Lyapunov, (Definition 2.4).

• Lyapunov’s indirect method, (Theorem 2.1).
• Lyapunov’s direct method, (Theorem 2.2).

Definition 2.4 An equilibrium point is Lyapunov stable (or stable in the sense of
Lyapunov) if solutions that start near the equilibrium point remain near the equilibrium
point for all time. More precisely, the equilibrium point xo of (2.1) is

7
8

• stable if, for each ǫ > 0 (no matter how small), there exists a γ = γ(ǫ) > 0 such
that, (see Figure 2.1)

kx(to ) − xo k < γ ⇒ kx(t) − xo k < ǫ , ∀ t ≥ to

t
xo (to ) x (t )
ε
xo (t )

γ ε

x (to )

Figure 2.1. The equilibrium point xo is stable.

• unstable if not stable.

• asymptotically stable if it is stable and γ can be chosen such that, (see Figure 2.2)

kx(t) − xo k → 0 as t→∞

t
xo (to )
x (t )
ε
xo (t )

γ ε

x (to )

Figure 2.2. The equilibrium point xo is asymptotically stable.

Note that in the above figures, xo = xo (t) , ∀ t ≥ to .

Example 2.1 Consider the system shown in Figure 2.3. The system has two equi-
librium points, xo = P 1 and xo = P 2. By applying Definition 2.4, characterize the
stability of these points.
 9

P2

P1

Figure 2.3. A simple system with two equilibrium points.

Assuming there is no friction in the system, and releasing the ball from any initial
position x(to ) sufficiently close to xo = P 1 within a circle of radius γ centered at
xo = P 1, the motion of the ball will then be limited within a circle of radius ǫ centered
at xo = P 1, see Figure 2.4. Since there is no friction in the system, the ball will move
(or oscillate) around xo = P 1 for ever within the circle. Therefore, xo = P 1 is a stable
e.p, and the ǫ − γ requirement is satisfied. However, if there is friction in the system,
the ball will eventually stay at xo = P 1 which is now asymptotically stable.
Next, releasing the ball from any initial position x(to ) sufficiently close to xo = P 2, the
ball will leave the circle of radius ǫ. Therefore, xo = P 2 is unstable since the ǫ − γ
requirement cannot be satisfied.

xo x (to )

γ
ε

xo x (t )
o

ε γ

Figure 2.4. Stability of equilibrium points based on Definition 2.4.

Stability of the equilibrium point of (2.1) can also be defined by examining the lin-
earized system. This approach is known as Lyapunov’s indirect method. Linearizing
the nonlinear system (2.1) around the equilibrium point xo , we obtain
∆ẋ = A∆x (2.2)
10

where  
∂f1 (x) ∂f1 (x)
  ∂x1
··· ∂xnx
∂f (x)  .. .. .. 
A= = . . .  (2.3)
∂x x=xo ∂fnx (x) ∂fnx (x)
∂x1
··· ∂xnx x=xo

which is also called the Jacobian matrix at xo . The eigenvalues of A (i.e., λ) are
obtained by solving
A −λ1 = 0 (2.4)
where 1 is an identity matrix.

Theorem 2.1 Let xo be the equilibrium point of the nonlinear system (2.1). Then,

(i) xo is asymptotically stable if all eigenvalues of matrix A have negative real parts.
(ii) xo is unstable if any eigenvalue of matrix A has a positive real part.

A more qualitative approach to stability analysis is Lyapunov’s direct method. The

method requires finding a scalar function which has some very special properties.

Theorem 2.2 The equilibrium point xo of (2.1) is stable if a continuously differen-

tiable scalar function V(x) can be found satisfying the following conditions:

(i) V(xo ) = 0,
(ii) V(x) > 0 for all x ∈ D, except at xo ,
∂V(x)
(iii) V̇(x) = ∂x
· f (x) ≤ 0 for all x ∈ D.

The equilibrium point xo is asymptotically stable if conditions (i)-(ii) hold, and

(iv) V̇(x) < 0 for all x ∈ D, except at xo where V̇(xo ) = 0,

or, alternatively

(v) V̇(x) ≤ 0 , provided that V̇(x) is not identically zero on any solution x(t) in D,
except at xo .

A function V(x) satisfying the above conditions is called Lyapunov function. The
proofs of the above theorems can be found in [2].
Thus by applying Theorem 2.1, the local stability or instability of an equilibrium point
of a nonlinear system (2.1) can be characterized based on the stability or instability
of the linearized system (2.2). However, it should be noted that the statement (i) in
Theorem 2.1 is generally not applicable if some (complex) eigenvalues have zero real
parts, and for these cases Lyapunov’s direct method (i.e. Theorem 2.2) can be applied
to characterize the stability of the equilibrium point.
 11

l
θ

m
mg

Figure 2.5. A simple pendulum system.

Example 2.2 Consider the simple pendulum system shown in Figure 2.5, where a
spherical mass with weight m is attached to a rod of negligible weight with length l.
Assuming a rigid pendulum (i.e. the length l is fixed), the mass moves along a circle
of radius l. The position along this circle is given by l θ. There is also a damping (or
frictional) torque TD resisting the motion which is assumed linearly proportional to the
speed of the mass, i.e. TD = D θ̇ where D ≥ 0 is a damping constant.
Describe mathematically the dynamic of this system. Also, find its equilibrium points,
and characterize their stability.

Applying Newton’s second law of motion, the following mathematical model (also
termed equation of motion) can then be obtained:

M θ̈ + mgl sin(θ) + D θ̇ = 0 (2.5)

where M = ml2 is moment of inertia and g is the acceleration due to gravity. Note
that θ̇ is the angular velocity.
Let  
x1
T
C1 = mgl , ω = θ̇ and x = [x1 x2 ] =   (2.6)
x2

Taking the state variables as x1 = θ and x2 = θ̇ = ω, then (2.5) and (2.6) give

       
ẋ1 θ̇ x2 f1 (x)
ẋ =   =   =  =  = f (x) (2.7)
1
ẋ2 θ̈ M
(−C1 sin(x1 ) − Dx2 ) f2 (x)

which is of the form given in (2.1) with nx = 2. According to Definition 2.1, the
equilibrium points of the system are given by f (xo ) = 0, i.e.

       
ẋ1 x2o f1 (xo ) 0
 = = =  (2.8)
1
ẋ2 M
(−C1 sin(x1o ) − Dx2o ) f2 (xo ) 0
12

Thus, x2o = 0 and x1o = ±kπ, where k is any integer. From the physical description
of the pendulum, it is clear that the pendulum has only two equilibrium positions
corresponding to the equilibrium points

xo = [x10 x2o ]T = [0 0]T and xo = [x10 x2o ]T = [π 0]T

Other equilibrium points are repetitions of these two positions which correspond to
the number of full swings the pendulum would make before it rests at one of the two
equilibrium positions.
Linearizing system (2.7) around its equilibrium point xo , the matrix A is obtained as
   
∂f1 (x) ∂f1 (x)
  0 1
∂f (x)  ∂x 1 ∂x 2

A= =  =  (2.9)
∂x x=xo ∂f2 (x) ∂f2 (x) K D
∂x ∂x
−M −M
1 2 x=xo

where K = C1 cos(x1o ).
Based on (2.4), the eigenvalues of A are given by
s 2
D D K
λ1 = − − −
2M 2M M
s 2 (2.10)
D D K
λ2 = − + −
2M 2M M

It is obvious that K = C1 > 0 with x1o = 0, and K = −C1 < 0 with x1o = π. Thus,
both eigenvalues of the matrix A have negative real parts with xo = [0 0]T . However,
the real part of λ2 is positive with xo = [π 0]T . Applying Theorem 2.1, it can be
found that xo = [0 0]T is asymptotically stable, but xo = [π 0]T is unstable.
By applying Theorem 2.2, it can also be shown that xo = [0 0]T is asymptotically
stable. Lyapunov’s direct method is closely related to the energy of the system. In
many physical systems the total energy of the system is a good candidate for a Lya-
punov function, and these energy-based functions are known as energy functions in the
literature. The energy function for the system in Example 2.2 is given by
Z x1
1 2
V(x) = WK + WP = M x2 + C1 sin(y) dy
2 0 (2.11)
1 2
= M x2 + C1 (1 − cos(x1 ))
2
where WK = 12 Mx22 is the kinetic energy and WP = C1 (1 − cos(x1 )) is the potential
energy of the pendulum.
Next, it will be shown that the function V(x) satisfies the conditions of Theorem 2.2
for xo = [0 0]T .

• V(xo ) = 0. Thus, condition (i) is satisfied.

 13

• The gradient of V(x) is given by

∂V(x) h ∂V(x) ∂V(x)

i  
∇ V(x) = = ∂x1 ∂x2
= C1 sin(x1 ) M x2
∂x

Since ∇ V(xo ) = 0, the function V(x) has a local extremum at xo = [0 0]T . It

will now be shown that this extremum is a local minimum, that is V(x) > 0.
Obviously, this extremum is a minimum if the Hessian matrix of V(x) is positive
definite. The Hessian matrix of V(x) is given by

 ∂ 2 V(x) ∂ 2 V(x)
  
2 ∂x21 ∂x1 ∂x2 C1 cos(x1 ) 0
∂ V(x)   
H= = = 
∂x2 ∂ 2 V(x) ∂ 2 V(x) 0 M
∂x2 ∂x1 ∂x22

which is a real symmetric matrix.

H is a positive definite matrix if all its eigenvalues are positive. It can be shown
that the eigenvalues of H are λ1 = M and λ2 = C1 cos(x1 ). Since M is a positive
π π
constant, H is positive definite if − < x1 < .
2 2
Thus, V(x) > 0 in a neighborhood (i.e. D) of xo . Therefore, condition (ii) is also
satisfied.

• The time derivative of V(x) is given by

∂V(x)
V̇(x) = · ẋ = ∇ V(x) · f (x) =
∂x  
x2 (2.12)
= [C1 sin(x1 ) M x2 ] ·   = −D x22 ≤ 0
1
M
(−C1 sin(x1 ) − Dx2 )

Since ẋ = f (x) 6= 0 (or more precisely ẋ2 = f2 (x) 6= 0 ) on x2 = 0 except at

xo , condition (v) is also satisfied. Thus, the equilibrium point xo = [0 0]T is
asymptotically stable.

Assume that D = 0. The eigenvalues are then given by

r
K
λ1,2 = ± −
M

By virtue of Theorem 2.1, it can still be shown that xo = [π 0]T is unstable. But,
Theorem 2.1 cannot characterize the stability of xo = [0 0]T when D = 0 (why?).
By applying Theorem 2.2, we may however be able to characterize the stability of
xo = [0 0]T .
14

It has already been shown that the energy function (2.11) satisfies conditions (i)-(ii).
With D = 0 the time derivative of V(x) is then given by
∂V(x)
V̇(x) = · ẋ = ∇ V(x) · f (x) =
∂x  
x2 (2.13)
= [C1 sin(x1 ) M x2 ] ·  =0
1
M
(−C1 sin(x1 ))
which satisfied condition (iii). Therefore, xo = [0 0]T is stable (but not asymptot-
ically). Henceforth, a stable (or an asymptotically stable) equilibrium point will be
designated by xso , and an unstable equilibrium point by xuo .
Assume that the pendulum is settled on its stable e.p xso = [0 0]T . At time t = tf , a
torque T (as a disturbance) is applied to the pendulum, and at time t = tc the applied
torque (or disturbance) is removed.
8

2
x2 (rad./s)

−2

−4

−6

−8
−100 −80 −60 −40 −20 0 20 40 60 80 100
x1 (deg.)

Figure 2.6. Phase portrait of the pendulum system.

Figure 2.6 illustrates phase portrait of the pendulum system due to the applied distur-
bance. The curve in the figure is termed the system trajectory (or motion). The stable
e.p of the system is indicated by “*”. During disturbance the system trajectory (i.e.
the pendulum) moves from its stable e.p towards the point indicated by “o” where the
disturbance is removed. This point is the initial point of the post-disturbance system.
Note that in this example, the pre-disturbance system and the post-disturbance system
have the same stable e.p.
When the system is frictionless (or undamped), i.e. D = 0, xso is not asymptotically
stable. Thus, the trajectory of the post-disturbance system (blue line) oscillates around
xso . However, when the system is damped, i.e. D > 0, xso is asymptotically stable.
Therefore, the trajectory (red line) tends to xso as t → ∞.
 15

Consider now the damped pendulum system. In Figure 2.7, the initial points are
indicated by “o”, and xso by “*”. Moreover, the unstable equilibrium points xu1
o =
[−180 0]T and xu2o = [180 0] T
are indicated by “+”.
As shown in the figure, the trajectories whose initial states (or points) lie within the
shaded region converge to xso . However, the trajectories whose initial states (or points)
lie outside the shaded region will never tend to xso .

xu1 xs xu2
x2

0 o o o

0
x1

Figure 2.7. Stability region of the damped pendulum system.

The shaded region is termed the stability region of xso . Boundary of the stability region
is termed the stability boundary.

Definition 2.5 The stability region of xso is a region in the state space from which
all trajectories converge to xso . Furthermore, the trajectory whose initial point lies on
the stability boundary will never leave the stability boundary and will converge to the
unstable equilibrium point on the stability boundary as time goes to infinity.

The stability region of the undamped pendulum system is illustrated in Figure 2.8.
Since xso is not asymptotically stable, trajectories starting within the stability region
will only oscillate around xso . However, trajectories starting outside the stability region
will diverge that means the pendulum rotates (not oscillates).
Figure 2.9 illustrates variation of the total energy (see equation (2.11)) in the pendulum
system. When the system is damped the total energy of the post-disturbance system
decreases that is the system is a dissipative system (see equation (2.12)). However,
when the system is undamped the total energy is constant that is the system is a
conservative system (see equation (2.13)).
16

x2

0
x1

Figure 2.8. Stability region of the undamped pendulum system.

D>0 D=0

Total Energy (black dashed line)

Kinetic Energy (blue solid line)

Potential Energy (red solid line)

Energy

Energy

0 0
0 0
Time Time

Figure 2.9. Variation of energy in the pendulum system.

Chapter 3
Synchronous machines modeling
Synchronous machines play an important roll in power systems. Nowadays almost all
power generators are synchronous generators. They convert mechanical power into
electrical form, and feed it into the transmission system. Many large loads are also
driven by synchronous motors. Furthermore, in large sizes, synchronous condensers
may be used to provide a very convenient and continuous control of reactive power and
voltage. These machines can be classified as either high-speed machines with cylindrical
rotors (or round-rotors) driven by steam or gas turbines, or low-speed machines with
salient-pole rotors driven by hydro turbines.
Synchronous machines are the most important components in the analysis of electrome-
chanical oscillations in power systems. Further, as discussed in Chapter 1, rotor angle
stability relies on keeping interconnected synchronous machines in synchronism after
a disturbance. Therefore, it is of fundamental importance to understand the behavior
and dynamics of synchronous machines in power system stability study.
Figure 3.1 illustrates simply a generating unit which is an electromechanical system
that can be divided into an electrical part (synchronous generator) and a mechanical
part (shaft and turbine).
Generator

I
Tm ω m Te Stator
+ Pe
Pm
Rotor U
Turbine Stator
Shaft I

Figure 3.1. A simple illustration of a generating unit.

The synchronous generator can also be divided into two parts, one static part called the
stator (or armature) and one rotating part called the rotor (or field). The stator has
a distributed winding which is connected to the ac power system. The stator winding
(called also the armature winding) consists of three identical phase windings which are
120 electrical degrees apart. These windings carry the load current and supply power
to the system. The rotor has a winding (called the field winding) which carries a direct
current (provided by an exciter) to produce a magnetic flux.
Figure 3.2 shows a simplified salient-pole synchronous generator. The ends of each
of the phase windings are denoted by a and a′ (phase a), b and b′ (phase b) and c
and c′ (phase c). The stator is represented by three magnetic axes a, b and c each
corresponding to one of the phase windings. However, the rotor is represented by two
axes, namely: the direct axis (d-axis), which is the magnetic axis of the field winding
(known as pole axis), and the quadrature axis (q-axis), which is the axis of symmetry

17
18

a-axis

a-axis ia
d-axis q-axis

d-axis
bm +
q-axis ua
bm

u + if
b¢ c

f
a a¢
n
if

c¢ c-axis
b-axis
b ib ub uc + c-axis

+
b-axis ic

Figure 3.2. Symbolic representation of a synchronous generator.

between two poles (known as the interpole axis). In this compendium, the q-axis is
located 90 electrical degrees behind the d-axis.
The rotor can also be equipped with additional short-circuited damper windings to
reduce the mechanical oscillations of the rotor. Figure 3.3 shows a salient-pole syn-
chronous generator with one damper winding on the d-axis of the rotor, and one damper
winding on the q-axis of the rotor.

d-axis a-axis

bm ia
rf

q-axis
,L

ra , Laa

+
u + f
i

Q
ff

ua
,L
rQ
rD
f

,L D
iD

0
=
D

iQ
uD

Q
u
=

n
0

L bb rc ,
rb, Lc
c
b-axis
ib ub uc + c-axis
+

ic
Figure 3.3. A simplified salient-pole synchronous generator with damper windings.

The mechanical rotor angle βm (also called the mechanical angle of the shaft) defines
the instantaneous position of the rotor d-axis with respect to a stationary reference.
The a-axis is here chosen as the reference. Thus, this angle is defined by
βm = ωm t + βmo (3.1)
where ωm is the mechanical speed (or angular velocity) of the rotor (or shaft), and βmo
is the initial position of the rotor, i.e. at t = t0 = 0 the position of the rotor is given by
 19

βm (t0 ) = βmo with respect to the a-axis. The rotor angle can be expressed in electrical
radians (or degrees) by
p
β = βm (3.2)
2
where β is the (electrical) rotor angle measured in electrical radians (or degrees), and
p is the number of the poles. Note that β = βm when p = 2.
Similarly, the electrical speed of the rotor (i.e. ωg ) can be obtained by
p
ωg = ωm (3.3)
2
dβ
where ωg = = 2πf , and f is the electrical frequency of the system given in (Hz).
dt
Based on equations (3.2)-(3.3), the following can therefore be obtained
p p p
β= βm = ωm t + βmo = ωg t + βo
2 2 2
dβ
= ωg (3.4)
dt
d 2β d ωg
2
=
dt dt

Positive direction of rotation is counter-clockwise, and positive angle βm (or β) is mea-

sured in the positive direction of rotation. Since β is measured with respect to a
stationary reference axis on the stator, it is an absolute measure of rotor angle. Conse-
quently, it continuously increases with time. In the steady-state, the rotor rotates with
a speed corresponding to the system nominal frequency fs (in Europe fs = 50 (Hz),
and in USA fs = 60 (Hz)) that is
dβ
= ωg = ωs = 2πfs
dt
where ωs = 2πfs is a constant normally called (electrical) synchronous speed. There-
fore, in the steady-state the rotor angle β increases uniformly with time. Due to
practical issues, it is more convenient to measure the angular position from the q-axis
with respect to a reference axis which rotates at synchronous speed, i.e ωs . This new
angular position is defined by
π π π
δ = (β − ) − ωs t = ωg t + βo − − ωs t = (ωg − ωs )t + (βo − )
2 2 2 (3.5)
= (ωg − ωs )t + δo

Note that in the steady-state ωg = ωs , and δ = δo which is a constant. Moreover,

dδ
= δ̇ = (ωg − ωs ) = ω
dt (3.6)
d 2δ d
2
= δ̈ = (ωg − ωs ) = ω̇g = ω̇
dt dt
that is, δ̇ = ω represents the deviation of the rotor speed from synchronism.
20

The rotor mechanical synchronous speed is however given by ωms = 2πno , where no is
given in revolution per minute. Note that
p
ωg ωm ωm
= p2 = (3.7)
ωs ω
2 ms
ωms

3.1 Swing equation

Consider again Figure 3.1. When the rotor is rotated by the turbine, a rotating mag-
netic flux is produced in the air gap. The magnetomotive force (mmf) produced by the
field current in the filed winding combines with the mmf produced by currents in the
stator winding. The resultant flux across the air gap between the stator and rotor in-
duces voltage in each phase of the armature winding. These induced voltages have the
same magnitudes, but are phase-shifted by 120 electrical degrees. The resultant flux
also provides the electromagnetic torque (or the electrical output torque denoted by
Te (Nm)) between the stator and rotor. This electromagnetic torque (developed in the
generator when it delivers power) opposes the torque of the turbine (or the mechanical
input torque denoted by Tm (Nm)).
One of the most important factors in studying electromechanical dynamics of a power
system is the motion of the rotor during and after a disturbance. The dynamic of this
motion is described by a set of differential equation based on Newton’s second law
d 2 βm d ωm
J 2
=J = Tm − Te (3.8)
dt dt
where J (kgm2 ) is the total moment of inertia of the turbine, shaft and generator.
Equation (3.8) is known as the swing equation. Multiplying the swing equation with
ωm the following can be obtained
d ωm
ωm J = Pm − Pe (3.9)
dt
where
Pm = ωm Tm is the mechanical power input given in (W)
Pe = ωm Te is the three-phase electrical power output given in (W)
Next, the inertia constant H of the generator is defined by
2
WKs 0.5Jωms
H= = (s) (3.10)
Sng Sng
where WKs is the total kinetic energy stored in the generator in the steady-state (or
stored kinetic energy in the rotating inertia at speed ωms ), and Sng is the generator
rated three-phase VA. Typically, H ranges between 3 and 6 seconds, depending on the
size and type of generator. The inertia constant states how many seconds it would take
to bring the generator from synchronous speed to standstill if rated power is extracted
from it while no mechanical power is supplied by the turbine.
 21

Substituting J from equation (3.10) into the swing equation (3.9) yields
2HSng d ωm
ωm 2
= Pm − Pe (3.11)
ωms dt
or by virtue of equations (3.3)-(3.7)
2HSng d ωg ωs
= (Pm − Pe ) (3.12)
ωs dt ωg
During the electromechanical dynamics of a power system, the rotor speed is near
ωg
synchronous speed, i.e. ≈ 1 and it has a negligible effect on the right hand side of
ωs
(3.12). Therefore, the following swing equation is commonly used for analysis of the
electromechanical dynamics (or transients) of a power system,
2HSng d ωg ωs
= (Pm − Pe ) ≈ Pm − Pe (3.13)
ωs dt ωg
3φ
Dividing both sides of (3.13) with an arbitrary three-phase base power Sbase results in
d ωg Pm − Pe
M = 3φ
= Pmpu − Pepu (3.14)
dt Sbase
where
2H Sng
M= 3φ
(3.15)
ωs Sbase
Applying equation (3.6), the motion of the rotor is then described by
δ̇ = (ωg − ωs ) = ω
1 (3.16)
ω̇ = (Pmpu − Pepu )
M
Furthermore, the contribution of mechanical friction in the bearings and/or damping
power provided by damper windings may be included in the swing equation (3.16).
′
This power is denoted by PD = D (ωg − ωs ) = D ω, where D = D ωs
, and D ′ is a small
positive constant. Including the damping, (3.16) is then rewritten as
δ̇ = ω
1 (3.17)
ω̇ = (Pmpu − Pepu − Dω)
M
However, with ω expressed in (pu), then (3.17) is rewritten as
δ̇ = ωs ωpu
1 (3.18)
ω̇pu = ′ (Pmpu − Pepu − D ′ ωpu )
M
′ ′
where, M = ωs M and D = ωs D.
ωg
Note that since the effect of on the right hand side of (3.12) is negligible, it has
ωs
ωg
been assumed that ≈ 1. However, the small variations of ωg are sufficient to
ωs
produce significant rotor angle deviation. Therefore, it is not justified to assume that
δ̇ = (ωg − ωs ) = ω ≈ 0.
22

3.2 Electrical equations

Consider again the synchronous generator shown in Figure 3.3. In developing the
mathematical model the following assumptions are made:

• The magnetic circuits are linear (i.e., no saturation).

• Magnetic hysteresis is negligible.

• The stator magnetomotive force (mmf) and flux are sinusoidal.

Since the windings are magnetically coupled with each other, the flux linkage of each
winding is due to its own current and the currents in all the other windings. The flux
linkage of each winding is therefore given by:

    
ψa Laa Lab Lac | Laf LaD LaQ ia
 ψb   Lba Lbb Lbc | Lbf LbD LbQ   ib stator
    
 ψc   Lca Lcb Lcc | Lcf LcD LcQ   ic 
 
  
— =  — — — —|— — — —  — (3.19)
 
  
 ψf   Lf a Lf b Lf c | Lf f Lf D Lf Q 
 if 
 
  
ψD  LDa LDb LDc | LDf LDD LDQ iD  rotor
ψQ LQa LQb LQc | LQf LQD LQQ iQ

or     
ψabc Labc,abc Labc,f DQ iabc
= (3.20)
ψf DQ Lf DQ,abc Lf DQ,f DQ if DQ

or in more compact form

ψsrf = Lsrf isrf (3.21)

where srf stands for stator reference frame that is the flux linkage equations are given
in the stator reference frame.
In (3.19), Laa , Lbb and Lcc represent the self-inductances of the stator three phase
windings, and Lf f , LDD and LQQ represent the self-inductances of the rotor field and
damper windings. The mutual inductance between two windings m and n is given by
Lmn . It should be noted that Lmn =Lnm , and Labc,f DQ = LTfDQ,abc .
It can be shown that all the elements of Labc,abc , Labc,f DQ and Lf DQ,abc vary as functions
of the rotor angular displacement β (i.e., they are functions of time). However, all the
elements of Lf DQ,f DQ are constants. Also, Lf Q = LDQ = 0, [3]-[6].
Using the voltage polarities and current directions of Figure 3.3, and also applying
Kirchhoff’s voltage law, the voltage equations for the stator and rotor windings are
 23

      
ψa ra 0 0 | 0 0 0 ia ua
 ψb   0 rb 0 | 0 0 0  ib   ub stator
      
 ψc   0 0 rc | 0 0 0  ic   uc 
d  
 —  = −— — — —|— — — —
  
— −  — 

(3.22)
dt  
 ψf 

0 0 0
   
   | rf 0 0  if  −uf 
   
ψD  0 0 0 | 0 rD 0 iD   0  rotor
ψQ 0 0 0 | 0 0 rQ iQ 0
or       
d ψabc Rabc 0 iabc uabc
=− − (3.23)
dt ψf DQ 0 Rf DQ if DQ uf DQ

or in more compact form

dψsrf
= ψ̇srf = − (Rsrf isrf + usrf ) (3.24)
dt
which is a set of differential equations describing the electrical behavior (or dynamic)
of a synchronous generator. However, (having replaced ψsrf in (3.24) with (3.21)) this
set of differential equations will be containing time-varying coefficients (since most of
the elements of Lsrf are functions of time) which make the analysis of the generator
dynamics more difficult.
Fortunately, this problem can be overcome by the so called Park’s transformation. This
transformation consists in transforming the phase quantities of the stator (for instance,
uabc , iabc and ψabc ) into three new quantities called dqo-components (i.e., udqo , idqo and
ψdqo ). The d- and q-components rotate together with the rotor, with the d-component
laying along the d-axis of the rotor, and the q-component laying along the q-axis of
the rotor. The o-component vanishes under balanced operation. Thus, by Park’s
transformation the quantities in the stator reference frame (srf ) are transformed to
the rotor reference frame (rrf ). This transformation is made by the matrix P , where

 2π

2π


cos (β) cos β − 3
cos β + 3
r 



2
 sin (β) sin β − 2π
sin β + 2π


P =  3 3  (3.25)
3 
 
√1 √1 √1
2 2 2

It should be noted that P −1 = P T , i.e. P is orthogonal.

The voltages, currents and flux linkages in the rotor reference frame are obtained by

udqo =P uabc
idqo =P iabc (3.26)
ψdqo =P ψabc
24

The quantities in the rotor reference frame can also be transformed to the stator ref-
erence frame by the inverse transformation as follows
uabc =P −1 udqo = P T udqo
iabc =P −1 idqo = P T idqo (3.27)
−1 T
ψabc =P ψdqo = P ψdqo

Applying (3.27) to (3.20), the following are obtained

 T    T 
P ψdqo L Labc,f DQ P idqo
= abc,abc ⇒
ψf DQ Lf DQ,abc Lf DQ,f DQ if DQ
 T       (3.28)
P 0 ψdqo L Labc,f DQ P T 0 idqo
= abc,abc
0 1 ψf DQ Lf DQ,abc Lf DQ,f DQ 0 1 if DQ
Let  
P 0
Pex =
0 1
Then,      
ψdqo Labc,abc Labc,f DQ −1 idqo
= Pex P (3.29)
ψf DQ Lf DQ,abc Lf DQ,f DQ ex if DQ

or in more compact form

−1
ψrrf = Pex Lsrf Pex irrf = Lrrf irrf (3.30)
where
 
Ld 0 0 | k Mf k MD 0
 0 Lq 0 | 0 0 k MQ 
 
   0 0 Lo | 0 0 0 
Ldqo,dqo Ldqo,f DQ  
Lrrf = = — — — —|— — — — 
Lf DQ,dqo Lf DQ,f DQ 
 k Mf

 0 0 | Lf f Lf D 0 

k MD 0 0 | LDf LDD 0 
0 k MQ 0 | 0 0 LQQ
q
with k = 3 . Note that the inductance matrix L is symmetrical, and all the
2 rrf
elements of this matrix are constants (i.e., independent of time) which is the main
advantage of Park’s transformation.
Equation (3.29) (or (3.30)) may be written in detail to give
    
ψd Ld k Mf k MD id
 ψf  =  k Mf Lf f Lf D  if  (3.31)
ψD k MD Lf D LDD iD
    
ψq Lq k MQ iq
= (3.32)
ψQ k MQ LQQ iQ
ψo = Lo io (3.33)
 25

These equations describe three magnetically decoupled winding sets. Under balanced
operation io is zero. Since we are dealing with symmetrical three-phase systems in this
compendium, the winding set (3.33) is omitted.
Figure 3.4 shows a pictorial representation of Park’s transformation. The winding set
(3.31) is laying on the d-axis of the rotor. The windings f and D (in red) are the
physical field and damper windings of the rotor shown in Figure 3.3, while the winding
d (in blue) is a fictitious winding representing the effect of the stator windings on
the d-axis, and it rotates with the field winding. The winding set (3.32) is laying on
the q-axis of the rotor. The winding Q (in red) is the physical damper winding of
the rotor shown in Figure 3.3, while the winding q (in blue) is also a fictitious winding
representing the effect of the stator windings on the q-axis. Note that these two winding
sets are magnetically decoupled (as shown in Figure 3.4) which is another advantage
of Park’s transformation.
xis
d-a i
d
+

ud
Ld
rd ,

iD Rotation
=0

wg xis
q-a
uD
D

Lq
,LD

r q, iq
rD

if
+

uf Q
+

, LQ uq
rQ iQ
, L ff
rf

=0
uQ

Figure 3.4. Representation of the salient-pole synchronous generator based on

Park’s transformation.

Equation (3.24) can also be transformed to the rotor reference frame by applying
−1 −1 −1
ψsrf = Pex ψrrf , isrf = Pex irrf and usrf = Pex urrf

Then, the following can be obtained

   
d −1
ψ̇rrf = − Rrrf irrf + urrf + Pex P ψrrf (3.34)
dt ex

Note that d Pex

−1 −1
6= 0, since Pex is a function of β (i.e. a function of time).
dt
Assuming ra = rb = rc = r (which in most cases is true), it can be shown that
−1
Rrrf = Pex Rsrf Pex = Rsrf with Rabc = Rdqo = r1
h i
Having determined Pex d Pex −1
, it can be shown that (by expanding equation (3.34))
dt
26

ψ̇d = − (ud + r id + ωg ψq )
ψ̇q = − (uq + r iq − ωg ψd ) (3.35)
ψ̇o = − (uo + r io )

and

ψ̇f =uf − rf if
ψ̇D = − rD iD (3.36)
ψ̇Q = − rQ iQ

Equation (3.35) is known as stator transients, since only stator quantities are involved
in this equation. Note that ud , uq , id , iq , ψd and ψq are the rotor-equivalents of the
stator voltages, currents and flux linkages. Once again, dealing with symmetrical three-
phase systems, then uo = io = 0. Therefore, ψ̇o can be omitted. Equation (3.36) is
known as rotor transients, since only rotor quantities are involved in this equation.
Equations (3.35)-(3.36) (i.e., equation (3.34) in compact form) together with equations
(3.31)-(3.33) (i.e., equation (3.30) in compact form) describe the electrical dynamic of
a synchronous generator in the rotor reference frame. These equations together with
equation (3.13) determine the behavior of the synchronous generator during different
disturbances in a power system.
In order to make a qualitative analysis of the behavior of a synchronous generator, it is
often meaningful to use models that comprise more simplifications and approximations
than those detailed model given by (3.35)-(3.36) and (3.31)-(3.33). Therefore, the
following assumptions are made:

1. Only balanced operation is considered that is the third equation in (3.35) is

removed, and also only positive-sequence quantities are considered.

2. The stator transients are neglected that is ψ̇d = ψ̇q = 0 in (3.35). This assumption
is justified since they are numerically small compared to ωg ψd and ωg ψq [3].

3. All the stator resistances are neglected that is r = 0 in (3.35). This assumption
is justified since they are very small.
ωg
4. During the transient-state the rotor speed is near synchronous speed, i.e. ≈1
ωs
(note however that these variations are sufficient to produce significant rotor
angle deviation).

5. The rotor transient saliency is neglected that is x′q ≈ x′d . It should be noted that
if there is no rotor winding on the q-axis of the rotor, then x′q = xq .

6. There are no damper windings on the rotor that is iD = iQ = 0.

 27

Based on the above assumptions, equations (3.31)-(3.32) may be rewritten as

ψd = Ld id + k Mf if
ψf = k Mf id + Lf f if (3.37)
ψq = Lq iq
and equations (3.35)-(3.36) may be rewritten as
ud = −ωs ψq = −ωs Lq iq
uq = ωs ψd = ωs (Ld id + k Mf if ) (3.38)
ψ̇f = uf − rf if

Figure 3.5 (a) shows a synchronous generator connected to a transmission network via
its terminal bus. The voltage at the terminal bus is Ū and the current injected to this
¯ Note that Ū is the generator line-to-neutral terminal voltage.
bus is I.
Imaginary-axis

U
dq-frame U Im

Generator U
I Transmission
SG Network θ
φ

Network-frame I γ
I Im
Real-axis
ωs
I Re U Re

(a) (b)

Figure 3.5. Single-line diagram of an SG connected to a transmission network,

and the phasor diagram of Ū and I¯ in the network reference frame.

In the transmission network, the voltages and currents are obtained and defined in the
network reference frame which rotates at synchronous speed ωs (see Figure 3.5 (b)).
However, as seen from the generator, the stator equivalent quantities are defined in the
rotor reference frame (dq-frame) based on Park’s transformation which rotates at the
rotor electrical speed ωg .
Using the Real-axis as the reference, the following can be obtained in the network
reference frame.
Ū = Uejθ = U cos(θ) + jU sin(θ) = URe + jUIm
I¯ = Iejγ = I cos(γ) + jI sin(γ) = IRe + jIIm (3.39)
φ= θ−γ
28

Under balanced operation, the instantaneous phase voltages and currents at the ter-
minal bus in the network reference frame are given by
√ √
ua = 2 U cos(ωs t + θ) , ia = 2 I cos(ωs t + γ)
√ √
ub = 2 U cos(ωs t + θ − 120) , ib = 2 I cos(ωs t + γ − 120) (3.40)
√ √
uc = 2 U cos(ωs t + θ + 120) , ic = 2 I cos(ωs t + γ + 120)

Next, an interface (or a relationship) between the components of Ū and I¯ in the rotor
reference frame and the network reference frame is derived.

3.3 Steady-state model

In equation (3.38), ud and uq may be rewritten as

ud = − ωs Lq iq = −xq iq
√
3 (3.41)
uq = ωs Ld id + kωs Mf if = xd id + √ eq
2
where eq = ωs Mf if is the stator equivalent electromotive force (emf) corresponding
to if , xd = ωs Ld is the d-axis synchronous reactance, and xq = ωs Lq is the q-axis
synchronous reactance. For the round-rotor generator xd = xq , and for the salient-
pole generator xd > xq . However, it may be assumed that xd ≈ xq if saliency is
neglected. Furthermore, note that ud , uq , id and iq in equation (3.41) are interpreted
as the instantaneous quantities of the fictitious windings in d- and q-axis due to Park’s
transformation.
Applying equation (3.26) (i.e. Park’s transformation) to equation (3.40), and measur-
ing the rotor position from the q-axis by the new angle δ, then by setting β = π2 + δ
the following can be obtained
√ √
uq = 3 U cos((ωg − ωs )t + δ − θ) , iq = 3 I cos((ωg − ωs )t + δ − γ)
√ √ (3.42)
ud = − 3 U sin((ωg − ωs )t + δ − θ) , id = − 3 I sin((ωg − ωs )t + δ − γ)

In the steady-state ωg = ωs , and equation (3.42) may be rewritten as

√ √
uq = 3 U cos(δ − θ) , iq = 3 I cos(δ − γ)
√ √ (3.43)
ud = − 3 U sin(δ − θ) , id = − 3 I sin(δ − γ)
Next, consider the following complex quantities
√
uq + j ud = 3 U [cos(δ − θ) − j sin(δ − θ)]
√ √ (3.44)
= 3 U e−j(δ−θ) = 3 Ū e−jδ
and
√
iq + j id = 3 I [cos(δ − γ) − j sin(δ − γ)]
√ √ (3.45)
= 3 I e−j(δ−γ) = 3 I¯ e−jδ
 29

It can be shown that [4]-[5]

ud uq id iq eq
Ud = √ , Uq = √ , Id = √ , Iq = √ and Eq = √ (3.46)
3 3 3 3 2

where Ud , Uq , Id are Iq the components of Ū and I¯ along the dq-axes. These components
and Eq are also referred to the per-phase rms stator equivalent of the rotor referenced
magnitudes ud , uq , id , iq and eq . Furthermore, Eq is also known as field excitation.
Note that, Ud , Uq , Id are Iq are different from ud , uq , id and iq .
Based on (3.46), the following is obtained from equation (3.41)

Ud
Ud = − xq Iq ⇒ Iq = −
xq
(3.47)
Uq − Eq
Uq = xd Id + Eq ⇒ Id =
xd

and from equations (3.44)-(3.45)

Ū = URe + jUIm = (Uq + j Ud ) ejδ = Uq ejδ + j Ud ejδ = Ūq + Ūd

(3.48)
I¯ = IRe + jIIm = (Iq + j Id ) ejδ = Iq ejδ + j Id ejδ = I¯q + I¯d

Equation (3.48) gives the relationship (or interface) between the components of Ū and
I¯ in the dq-axes reference frame (based on Park’s transformation) which is rotating at
speed ωg and in the network reference frame which is rotating at synchronous speed
ωs . This interface is also shown in Figure 3.6. The position of the q-axis with respect
to the synchronous reference frame is given by δ = (ωg − ωs )t + δ0 , see equation (3.5).
Note that in steady-state ωg = ωs , and δ = δ0 .

Im-axis q-axis

Uq
U Im U

δ
d-axis

θ Re-axis
U Re
( −) U d

Figure 3.6. Generator-network interface.

The interface between the generator reference frame and the network reference frame
may be given in matrix form as
      
Uq cos(δ) sin(δ) URe U cos(θ − δ)
= = (3.49)
Ud − sin(δ) cos(δ) UIm U sin(θ − δ)
30

The inverse relation is given by

    
URe cos(δ) − sin(δ) Uq
= (3.50)
UIm sin(δ) cos(δ) Ud

By virtue of (3.49), equation (3.47) can then be rewritten as

U sin(θ − δ)
Iq = −
xq
(3.51)
U cos(θ − δ) − Eq
Id =
xd

From equations (3.47)-(3.48), we have

Ū = (Uq + j Ud ) ejδ = xd Id ejδ + Eq ejδ − j xq Iq ejδ

= − j xd j Id ejδ + Eq ejδ − j xq Iq ejδ (3.52)
= − j xd I¯d + Ēq − j xq I¯q (V)
or
Ēq = Ū + j xd I¯d + j xq I¯q (V) (3.53)

Usually, the position of the q-axis is not known (i.e. δ is not known). However (based
on the load flow calculation), Ū and I¯ are known. Therefore, the position of the q-axis
with respect to the network reference frame can be obtained by some mathematical
manipulation in equation (3.53) as follows

Ēq = Ū + j xd I¯d + j xq (I¯ − I¯d )

= Ū + j xq I¯ + j (xd − xq )I¯d (3.54)
= ĒQ + j (xd − xq )I¯d (V)

The phasor diagram of equations (3.53) and (3.54) is shown in Figure 3.7.

q-axis

Eq jx
qI
q
d
)I
q
−x

EQ
d
j( x

d
dI
jx qI

jx

Uq
d-axis δ U
Iq
I θ Re-axis

( −) I d
( −) U d

Figure 3.7. Phasor diagram representing equations (3.53) and (3.54).

 31

In equation (3.54), ĒQ = Ū + j xq I¯ can be easily calculated since Ū and I¯ are known.
Subsequently, Ēq can be easily calculated. Note that ĒQ and Ēq both have the phase
angle δ. Another mathematical manipulation in equation (3.53) gives
Ēq = Ū + j xd (I¯ − I¯q ) + j xq I¯q
(3.55)
= Ū + j xd I¯ − j (xd − xq )I¯q (V)
or
Ē = Ēq + j (xd − xq )I¯q = Ū + j xd I¯ (V) (3.56)
where Ē is termed the generator internal voltage behind the d-axis synchronous reac-
tance. The circuit representation of equation (3.56) is shown in Figure 3.8.

jxd I

E U

Figure 3.8. Per-phase equivalent circuit of a synchronous generator in steady-state.

By virtue of equations (3.47)-(3.51), the three-phase (complex) electrical power output

of the generator is given by
S̄e = 3 Ū I¯∗ = 3 (Uq + j Ud ) (Iq − j Id ) = 3 (Uq Iq + Ud Id ) + j3 (Ud Iq − Uq Id )
(3.57)
=Pe + j Qe (VA)
where
Pe = 3 (Uq Iq + Ud Id )
U 2 sin(δ − θ) cos(δ − θ) Eq U sin(δ − θ) − U 2 sin(δ − θ) cos(δ − θ)
=3 +3
xq xd (3.58)
2
 
Eq U sin(δ − θ) U sin(2(δ − θ)) 1 1
=3 +3 − (W)
xd 2 xq xd
and
Qe = 3 (Ud Iq − Uq Id )
−U 2 sin2 (δ − θ) Eq U cos(δ − θ) − U 2 cos2 (δ − θ)
=3 +3
xq xd (3.59)
 2 2

sin (δ − θ) cos (δ − θ) Eq U cos(δ − θ)
= − 3 U2 + +3 (VAr)
xq xd xd
Since r = 0, the generator three-phase active power (Pe ) can also be obtained by
Pe = Real [3 Ē I¯∗ ]
 
Eq U sin(δ − θ) U 2 sin(2(δ − θ)) 1 1 (3.60)
=3 +3 − (W)
xd 2 xq xd
32

If saliency is neglected, i.e. xd ≈ xq , then Ēq = ĒQ = Ē, and

Eq U sin(δ − θ)
Pe = 3
xd
(3.61)
U 2 − Eq U cos(δ − θ)
Qe = − 3
xd

3.4 Transient-state model

3.4.1 Classical model

From (3.37), the following can be obtained

ψf − k Mf id
if = (3.62)
Lf f
Substituting if from (3.62) into uq in (3.38), the following is obtained
  √
k 2 Mf2 ωs Mf 3
uq = ωs Ld − id + k ′
ψf = xd id + √ e′q (3.63)
Lf f Lf f 2
ωs Mf
where, e′q = ψf is termed the q-axis transient emf which is proportional to the
Lf f
′
ωs k 2 Mf2 ωs k 2 Mf2
field flux linkage ψf , and xd = ωs Ld − = xd − is termed the d-axis
Lf f Lf f
transient reactance.
Note that x′d < xd , and x′q = xq since there is no rotor winding on the q-axis. Then,
for the classical model we have
ud = − xq iq
√
3 (3.64)
uq = x′d id + √ e′q
2
where, e′q is constant since ψf is assumed constant in the classical model. This implies
that in the classical model there is no electrical dynamic since the only dynamic in
equation (3.38) will be set to zero, ψ̇f = 0.
Based on the assumption 4 above, equations (3.43)-(3.53) will be also valid for the
classical model with the exception that Eq and xd will be replaced by Eq′ and x′d ,
respectively. Thus, for the classical model

• Equation (3.51) is modified to

U sin(θ − δ)
Iq = −
xq
(3.65)
U cos(θ − δ) − Eq′
Id =
x′d
 33

• Equation (3.53) is modified to

Ēq′ = Ū + j x′d I¯d + j xq I¯q (V) (3.66)

The phasor diagram of equations (3.53) and (3.66) is shown in Figure 3.9.

q-axis

Eq jx
qI
q

Eq′ jx
qI
q
{
d
Id

dI
jx ′
d

jx
Uq
d-axis δ
Iq U
I θ Re-axis

( −) I d
( −) U d

Figure 3.9. Phasor diagram representing equations (3.53) and (3.66).

Note that from the figure the following can be obtained

Eq = Eq′ − (xd − x′d ) Id (3.67)

• Equation (3.56) is modified to

Ē ′ = Ēq′ − j (xq − x′d )I¯q = Ū + j x′d I¯ (V) (3.68)

where Ē ′ is termed the generator transient internal voltage behind the d-axis
transient reactance. The phasor diagram of equation (3.68) is shown in Figure
3.10, where δ ′ is the phase angle of Ē ′ . Note that δ = δ ′ + α′ .

q-axis

Eq′

α ′ E′
jxd′ I
δ
d-axis
δ′ U
Re-axis

Figure 3.10. Phasor diagram representing equation (3.68)

34

Since ψf is assumed constant, E ′ and α′ are also constant with respect to the
rotor axes [5], and in the classical model the generator is represented by a voltage
source with a constant magnitude E ′ behind its d-axis transient reactance x′d as
shown in Figure 3.11 (a).

jxd′ I jxd′ I

E′ U Eq′ U

(a) (b)

Figure 3.11. Equivalent circuit of a synchronous generator for transient-state studies.

The generator three-phase active power is then given by

Pe = Real [3 Ē ′ I¯∗ ] = 3 (Uq Iq + Ud Id ) = 3(Eq′ Iq + (x′d − xq )Id Iq )

 
Eq′ U sin(δ − θ) U 2 sin(2(δ − θ)) 1 1 (3.69)
=3 +3 − (W)
x′d 2 xq x′d

Based on the assumption 5, x′d ≈ xq . This assumption results in Ēq′ = Ē ′ (i.e. δ = δ ′

and α′ = 0), and in Figure 3.11 (a) the voltage source (Ē ′ ) is replaced by Ēq′ = Eq′ ejδ
as shown in Figure 3.11 (b) based on which

Ēq′ = Ū + j x′d I¯ (V) (3.70)

and (since r = 0)

  Eq′ U
¯∗ ′
Pe = Real 3Ēq I = 3 Eq Iq = 3 ′ sin(δ − θ)
′ (W) (3.71)
xd

The dynamic of a synchronous generator represented by the classical model is then

described only by the swing equation (3.17), i.e.

δ̇ = ω
1 (3.72)
ω̇ = (Pmpu − Pepu − Dω)
M

where, Pmpu is assumed to be constant and Pepu is the per unit value of Pe in equation
(3.71) which is defined in Section 3.5.
 35

3.4.2 One-axis model

In the classical model, ψf is assumed constant which implies that e′q (or Eq′ ) is also
constant. This assumption is justified if the generator is located long away from the
disturbance point.
In the one-axis model (also known as flux-decay model), the variation of ψf is however
considered (although it changes slowly). This change can be determined by differenti-
ωs Mf
ating e′q = ψf (, and substituting ψ˙f from (3.38)) as follows:
Lf f
ωs Mf ˙ ωs Mf
e˙′q = ψf = (uf − rf if ) (3.73)
Lf f Lf f
which can be rewritten as
Lf f ˙′ ωs Mf
eq = uf − ωs Mf if (3.74)
rf rf
Let
′ Lf f ωs Mf
Tdo = , ef = uf and from (3.41) eq = ωs Mf if
rf rf
′
where, Tdo is termed the d-axis transient open-circuit time constant, and ef is an emf
proportional to the field voltage uf . Then, equation (3.74) is rewritten as
′ ˙′
Tdo eq = ef − eq (3.75)
Let also
e′q ef eq
Eq′=√ , Ef = √ and Eq = √ (3.76)
2 2 2
′ ′
where, Eq , Ef and Eq are the rms values of eq , ef and eq .
Substituting equations (3.76) and (3.67) into (3.75) and using (3.65), the dynamic of
Eq′ (with rms quantities) is obtained as follows
′
Tdo Ėq′ = Ef − Eq = Ef − Eq′ + (xd − x′d )Id
xd xd − x′d (3.77)
= Ef − ′ Eq′ + U cos(δ − θ)
xd x′d
The dynamic of the generator using the one-axis model is then given by (in (pu))
δ̇ = ω
1
ω̇ = (Pmpu − Pepu − Dω)
M ! (3.78)
′
1 xdpu ′
xdpu − x dpu
Ė ′ qpu = ′ Ef pu − ′ Eqpu + Upu cos(δ − θ)
Tdo xdpu x′dpu
where, Pepu is the per unit value of Pe in equation (3.71), and Ef pu is constant. The
per unit variables are defined in Section 3.5.
The one-axis model is widely used in voltage stability analysis, and also small-signal
analysis for designing Power System Stabilizer (PSS). The impact of PSS on damping
of electromechanical oscillation will be discussed further in this compendium.
36

3.4.3 Two-axis model

In the two-axis model, it is assumed that the q-axis of the rotor is equipped with
a short-circuited damper winding. This winding may also represent the effect of the
rotors of the high-speed generators which are driven by steam or gas turbines, and have
cylindrical (or round) rotors made up of solid steel forging. A high-speed generator
often does not have special damper windings. However, the solid steel rotor body acts
as a damper winding in the q-axis by offering paths for eddy currents [5]-[6].
Having a short-circuited damper winding in the q-axis of the rotor then iQ 6= 0 in the
assumption 6, and equations (3.37) and (3.38) are rewritten as
ψd = Ld id + k Mf if
ψf = k Mf id + Lf f if
ψq = Lq iq + k MQ iQ (3.79)
ψQ − k MQ iq
ψQ = k MQ iq + LQQ iQ ⇒ iQ =
LQQ
and
ud = − ωs ψq = −ωs (Lq iq + k MQ iQ )
uq = ωs ψd
(3.80)
ψ̇f = uf − rf if
ψ̇Q = − rQ iQ
Substituting iQ from equation (3.79) into ud in (3.80), the following is obtained
  √
ωs k 2 MQ2 ωs MQ 3
ud = − ωs Lq − iq − k ψQ = −x′q iq + √ e′d (3.81)
LQQ LQQ 2
ωs MQ ωs k 2 MQ2 ωs k 2 MQ2
where, e′d = − ψQ , and x′q = ωs Lq − = xq − . This reactance
LQQ LQQ LQQ
is termed the q-axis transient reactance. Note that x′q < xq .
Equation (3.64) is then modified as
√
3 √
ud = −x′q iq + √ e′d = −x′q iq + 3 Ed′
2
√ (3.82)
3 √
uq = x′d id + √ e′q = x′d id + 3 Eq′
2
or
Ud − Ed′
Ud = −x′q Iq + Ed′ ⇒ Iq = −
x′q
(3.83)
Uq − Eq′
Uq = x′d Id + Eq′ ⇒ Id =
x′d
e′
where, Ed′ = √d .
2
 37

Following a procedure similar to that for the one-axis model, the dynamic of Ed′ (with
rms quantities) is obtained as follows

′ xq ′ xq − x′q
Tqo Ėd′ = −Ed′ − (xq − x′q )Iq = − E − U sin(δ − θ)
x′q d x′q

′ LQQ
where, Tqo = is termed the q-axis transient open-circuit time constant.
rQ
Thus, the dynamic of the generator using the two-axis model is given in (pu) by

δ̇ = ω
1
ω̇ = (Pmpu − Pepu − Dω)
M !
′
1 x dpu ′
x dpu − x dpu (3.84)
Ė ′ qpu = ′ Ef pu − ′ Eqpu + Upu cos(δ − θ)
Tdo xdpu x′dpu
 ′ 
1 x qpu ′
x qpu − x qpu
Ė ′ dpu = ′ − ′ Edpu − Upu sin(δ − θ)
Tqo xqpu x′qpu

where, Pepu is the per unit value of

Pe = 3(Uq Iq + Ud Id ) = 3(Eq′ Iq + Ed′ Id + (x′d − x′q )Id Iq ) = 3(Eq′ Iq + Ed′ Id )

Eq′ U sin(δ − θ) + Ed′ U cos(δ − θ) (3.85)
=3
x′d

since x′q = x′d based on the assumption 5. The two-axis model is widely used in transient
stability analysis especially for modeling generators with round rotors.
Table 3.1 gives the typical values of generator parameters (depending upon the size of
generator). Reactances are expressed in (pu) based on the generator rated values, H
is given in (s) based on the generator rated three-phase MVA, and time constants are
given in (s).

Parameter Round Rotor Salient-Pole Rotor

xd 1.0 – 2.3 0.6 – 1.5
xq 1.0 – 2.3 0.4 – 1.0
x′d 0.15 – 0.4 0.2 – 0.5
x′q 0.3 – 1.0 —
′
Tdo 3.0 – 10.0 1.5 – 9.0
′
Tqo 0.5 – 2.0 –
H 2.5 – 10 2.0 – 4.0
Table 3.1. Typical values of generator parameters [6].
38

3.5 Per unit conversion

As described in [3]-[4], it is more convenient to normalize the generator quantities based
on the following base values.
1φ
Sbase = the generator rated per-phase VA
LN
Ubase = the generator rated line-to-neutral rms terminal voltage
1φ LN (3.86)
Sbase Ubase
Ibase = LN
and Zbase = ⇒
Ubase Ibase
1φ LN LN
Sbase = Ubase Ibase and Ubase = Zbase Ibase

Applying the above base values to equation (3.70), then

Ēq′ Ū x′d I¯
LN
= LN
+ j ⇒ Ē ′ qpu = Ūpu + j x′dpu I¯pu (3.87)
Ubase Ubase Zbase Ibase
where
Ēq′ Ū I¯ x′d
Ē ′ qpu = LN
, Ūpu = LN
, I¯pu = and x′dpu = (3.88)
Ubase Ubase Ibase Zbase

Also, (3.71) can be converted to (pu) as follows

Pe Eq′ U Eq′ U Zbase
1φ
= 3 ′ LN sin(δ − θ) = 3 LN LN sin(δ − θ) ⇒
Sbase xd Ubase Ibase Ubase Ubase x′d
′ (3.89)
Eqpu Upu
Pepu1φ = 3 sin(δ − θ)
x′dpu
3φ 1φ
If the generator rated three-phase VA (Sbase = 3 Sbase ) is used √
as the base power, and
LL LN
the generator rated line-to-line rms terminal voltage (Ubase = 3 Ubase ) is used as the
base voltage, then
3φ
√ LL
Sbase = 3 Ubase Ibase
3φ 1φ
Sbase 3Sbase
Ibase = √ LL = LN
(i.e., the same numerical value as in (3.86))
3Ubase 3Ubase (3.90)
LL 2 LN
(U ) Ubase
Zbase = base
3φ
= (i.e., the same numerical value as in (3.86))
Sbase Ibase
√
Also, if the line-to-line rms voltage (Ū LL = 3 Ū) of the terminal bus is considered
in Figure 3.11 (b), then the equation (3.70) is written as
U LL √ √
Ēq′ = √ + j x′d I¯ ⇒ 3 Ēq′ = U LL + j x′d 3 I¯ (3.91)
3
Applying the base values given in (3.90) to equation (3.91), then
√ √
3 Ēq′ Ū LL x′d 3 I¯
LL
= LL + j √ ⇒ Ē ′ qpu = Ūpu + j x′dpu I¯pu (3.92)
Ubase Ubase Zbase 3 Ibase
 39

where
√
3 Ēq′ Ēq′
Ē ′ qpu = LL
= LN
(i.e., the same numerical value as in (3.88))
Ubase Ubase
Ū LL Ū
Ūpu = LL
= LN (i.e., the same numerical value as in (3.88))
Ubase Ubase
¯
I x′
I¯pu = and x′dpu = d (i.e., the same numerical values as in (3.88))
Ibase Zbase

However, for the active power the following is obtained

√ ′√ √ ′
Pe 3 Eq 3 U 3 Eq U LL Zbase
3φ
= √ sin(δ − θ) = LL LL
sin(δ − θ) ⇒
Sbase LL
x′d 3 Ubase Ibase Ubase Ubase x′d
′
(3.93)
Eqpu Upu 1
Pepu3φ = ′
sin(δ − θ) = Pepu1φ
xdpu 3

Henceforward, the three-phase power and the line-to-line voltage at each bus will be
considered, and the superscripts ”3φ” and ”LL” will therefore be omitted. Moreover,
the normalization of the quantities will be based on
Sbase = an arbitrary three-phase base power
Ubase = a line-to-line base voltage
(3.94)
Sbase (Ubase )2
Ibase = √ and Zbase =
3 Ubase Sbase
Thus,
√ √ ′ √ ′ √
3Eq ′
3Eq ′ 3Ed 3Ef
Eqpu = , Eqpu = , Edpu = , Ef pu =
Ubase Ubase Ubase Ubase
′
xd x ′
x q xq
xdpu = , x′dpu = d , xqpu = , x′qpu = (3.95)
Zbase Zbase Zbase Zbase
U I Pe Pe
Upu = , Ipu = , Pepu = =√
Ubase Ibase Sbase 3 Ubase Ibase

Here-onward, if not otherwise explicitly stated, all used quantities in the compendium
(apart from ω) are expressed in (pu), and the subscript ”pu” will therefore be omitted.
Next, based on the above base values and considering the terminal voltage Ū as line-
to-line voltage, the electric power in (pu) for the classical model and the one-axis
model is given by
  Eq′ U
Pe = Real Ēq′ I¯∗ = Eq′ Iq = ′ sin(δ − θ) (pu) (3.96)
xd
since x′d = xq . Furthermore, the equivalent circuit of a generator represented by one of
those two models is shown in Figure 3.11 (b), where
Ēq′ = Eq′ ej δ = Ū + j x′d I¯ (pu) (3.97)
40

For the two-axis mode, the electric power in (pu) is given by

  Eq′ U sin(δ − θ) + Ed′ U cos(δ − θ)
Pe = Real Ē ′ I¯∗ = Eq′ Iq + Ed′ Id = (pu) (3.98)
x′d
since x′d = x′q . The equivalent circuit of a generator represented by the two-axis model
is similar to that shown in Figure 3.11 (a), but with

Ē ′ = (Eq′ + j Ed′ ) ej δ = Ū + j x′d I¯ (pu) (3.99)

3.6 Excitation systems

The generator excitation system consists of an exciter and an Automatic Voltage Reg-
ulator (AVR). The primary function of an exciter is to provide a dc source for field
excitation of a synchronous generator, and the AVR controls the excitation voltage. A
control on excitation voltage results in controlling the field current (i.e. if in Figure
3.3, or Ef in equation (3.78)) which in turn controls the generated voltage and reactive
power. This action contributes to the enhancement of system stability.
In general, there are two types of exciters, namely rotating and static exciters. The
rotating exciters make use of either dc generators or ac generators with rectifiers as
sources to provide dc current to the field winding of the generator. Figure 3.12 illus-
trates basic structure of the excitation systems with rotating exciters.
dc exciter Synchronous generator
Slip ring Field Stator
Field
+ if
rf , L ff

Stator uf

Input data

Voltage regulator
Input data
ac exciter Synchronous generator
Slip ring Field Stator
Field Stator Rectifier
+ if
rf , L ff

uf

Input data

Voltage regulator
Input data

Figure 3.12. Basic structure of the excitation systems with rotating exciters.

A dc exciter may be driven by a motor or the shaft of the generator. It supplies dc

current to the field winding of the synchronous generator through slip rings ( a slip
ring is a rotary continuity electromechanical device to transfer electrical power and
 41

electrical signals from a stationary to a rotating structure). These excitation systems

belong to early systems. Due to their obsolescence, and also the needs of today’s power
systems to high speed excitation systems on all grid connected generating units with
high-field forcing capability, many older dc exciters are being replaced by ac or static
type systems. In an ac excitation system, the ac output is rectified to provide the dc
current required by the field winding of the synchronous generator. If the rectifier is
stationary slip rings will be needed. With rotating rectifiers, the need for slip rings is
however eliminated.
Figure 3.13 shows the basic structure of static excitation systems. The rectifiers are
directly fed from the generator terminals (or the station auxiliary bus) via a step-down
transformer to provide the dc current required by the field winding of the synchronous
generator. In these excitation systems, slip rings are needed, and the rectifiers are
controlled directly by a voltage regulator. By far, most excitation systems installed
nowadays are of this type.
Synchronous generator
Slip ring Field Stator
Rectifier
+ if
rf , L ff

uf

Input data

Voltage regulator
Input data

Exciter transformer

Figure 3.13. Basic structure of static excitation systems.

As mentioned before, the above figures only illustrate the basic structure of these
different types of excitation systems. Each type has however different configurations
which are comprehensively described in [3]-[6], and related references therein. Figure
3.14 shows the block diagram of a simple excitation system which will be used in this
compendium (especially for small signal analysis).

U ref E f max

+ Ef
KA
¦
- 1 + Te s
U input
E f min

Figure 3.14. Block diagram of a simple excitation system.

In the figure, KA and Te represent the gain of an amplifier and the time constant of
exciter, respectively. Further, Uref is the set value and Uinput is the voltage which
42

will be controlled. Uinput is indeed the output of the block diagram shown in Figure
3.15, where Ū is the generator terminal voltage, I¯ is the current flowing out from the
generator, and U = |Ū |.

U UC 1 U input
U C = U ± ( RC + jX C ) I
1 + TR s

Figure 3.15. Block diagram of load compensation and transducer.

The first block is known as load compensator, and Z̄C = RC + jXC ≈ jXC is known
as the compensation impedance. If the voltage at the generator terminal will be con-
trolled, then Z̄C = 0 that is UC = U. However, if the voltage at a point beyond the
generator terminal will be controlled, then Z̄C 6= 0 which represents the electrical dis-
tance between that point and the generator terminal. The negative sign means that the
point at which the voltage will be controlled is closer to (or within) the transmission
system. The positive sign is however used if several parallel generators (equipped with
AVR) are connected to the same terminal bus via a common step-up transformer. By
this compensation, the voltage at an (artificial) point within each generator will be
controlled which results in an adequate and stable reactive power productions between
the generators. The second block is known as transducer which represents the delay
due to measuring, rectifying and filtering of the signal. TR is usually very small, and
this block can therefore be omitted. In this compendium, the load compensation is
also omitted since the control of the terminal voltage is of concern that is Uiput = U.
It will be shown that excitation system with high gain will introduce very poor or nega-
tive damping in electromechanical oscillations which can lead to angular instability. To
eliminate this effect and to improve the system damping in general, a supplementary
damping device known as Power System Stabilizer (PSS) is added in the excitation sys-
tem. This device provides an additional supplementary signal to the voltage regulator
at the summing junction as shown in Figure 3.16

U ref E f max

u pss ++ KA Ef
PSS å
- 1 + Te s
U
E f min

Figure 3.16. Block diagram of a simple excitation system with PSS.

Figure 3.17 shows the block diagram of a PSS.

uin Tw s 1 + T1s 1 + T3 s u pss

K PSS
1 + Tw s 1 + T2 s 1 + T4 s

Figure 3.17. Block diagram of a PSS.

 43

uin is the input signal. The generator speed deviation (i.e ω) or the generator real
power (i.e. Pe ) is the most commonly used input signal. KP SS is the stabilizer gain
which determines the magnitude of damping provided by PSS. The first block in the
figure represents a high-pass filter (known as washout block). The purpose of this filter
is to stop contribution from a steady-state deviation of the input signal. The second
and third blocks are known as phase compensation blocks which are indeed lead-lag
type transfer functions. The purpose of these blocks are to shift the phase by setting
T1 -T4 at values so that a positive contribution to damping is obtained. Normally, the
second phase compensation block will be used if the phase angle of the first phase
compensation block is greater than a given maximum value. Then, T1 = T3 and
T2 = T4 . However, if the phase angle of the first phase compensation block is less than
the given maximum value, then the second phase compensation block will not be used
(i.e. T3 = T4 ). The tuning of a PSS is discussed in Chapter 6.

3.6.1 One-axis model with AVR and PSS

It has already been shown that the dynamic of the generator k represented by the
one-axis model is given by

δ̇k = ωk
′
1 1 Eqk Uk
ω̇k = (Pmk − Pek − Dk ωk ) = (Pmk − sin(δk − θk ) − Dk ωk )
Mk Mk ′
xdk (3.100)
 
1 xdk ′ xdk − x′dk
Ė ′ qk = ′ Ef k − ′ Eqk + Uk cos(δk − θk )
Tdok xdk x′dk

where Ef k is constant.
Using the AVR shown in figure 3.16, then Ef k is not longer constant and its dynamic
is given by
1
Ėf k = (−Ef k + KAk (Ukref + upssk − Uk )) (3.101)
Tek

If no PSS is utilized then upssk = 0.

Using the PSS shown in figure 3.17, if the washout block and the second phase com-
pensation block are omitted, and also the generator speed deviation (i.e ωk ) is used
as the input signal, then the output signal upss can be considered as a state variable
whose dynamic is given by
1
u̇pssk = (−upssk + KP SSk ωk + T1k KP SSk ω̇k )
T2k
  (3.102)
1 Pmk − Pek − Dk ωk
= −upssk + KP SSk ωk + T1k KP SSk
T2k Mk

In general, by using the following equivalent block diagram shown in Figure 3.17 for a
PSS,
44

uin Tw s c1 c1 1 + T1s c3 c3 1 + T3 s u pss

K PSS 1 + Tw s 1 + T2 s 1 + T4 s

uin 1
s1 c1 T1 -
+ c2 1 s2 c3 T3 -
+ c4 1 s3 u pss
K PSS S S S S S
1 + Tw s - T2 1 + T2 s + T4 1 + T4 s +
+ + +

Figure 3.18. The equivalent block diagram of the PSS.

the following differential equations can be obtained and added to the dynamic model
of the generator k,
1
Ṡ1k = (uink KP SSk − S1k )
Twk
1
Ṡ2k = (c2k − S2k ) (3.103)
T2k
1
Ṡ3k = (c4k − S3k )
T4k
where,
c1k =uink KP SSk − S1k
T1k
c2k =c1k (1 − )
T2k
T1k (3.104)
c3k =c1k + S2k
T2k
T3k
c4k =c3k (1 − )
T4k
and finally
T3k
upssk = c3k + S3k (3.105)
T4k

Thus, the dynamic of the generator k with the one-axis model and the proposed AVR
and PSS models is given by equations (3.100)-(3.105).
Next, using (3.104) in (3.103) the following is obtained
Ṡ1k = s1uink uink − s1s1k S1k
Ṡ2k = s2uink uink − s2s1k S1k − s2s2k S2k (3.106)
Ṡ3k = s3uink uink − s3s1k S1k + s3s2k S2k − s3s3k S3k
where,
1
s1s1k = , s1uink = s1s1k KP SSk
Twk
 
1 T1k
s2s2k = , s2s1k = s2s2k 1 − , s2uink = s2s1k KP SSk (3.107)
T2k T2k
   
1 T3k T1k
s3s3k = , s3s2k = s3s3k 1 − , s3s1k = s3s2k , s3uink = s3s1k KP SSk
T4k T4k T2k
 45

Furthermore, (3.105) is rewritten as

T1k T3k T1k T3k T3k

upssk = KP SSk uink − S1k + S2k + S3k (3.108)
T2k T4k T2k T4k T4k

Substituting (3.108) in (3.101) gives

1 KAk KAk
Ėf k = − Ef k + Ukref − Uk +efuink uink −efs1k S1k +efs2k S2k +efs3k S3k (3.109)
Tek Tek Tek
where,
KAk T3k T1k
efs3k = , efs2k = efs3k , efs1k = efs2k , efuink = efs1k KP SSk (3.110)
Tek T4k T2k

Note that, if the first lead-lag filter is only used (i.e. T3k = T4k ), then the third equation
in (3.106) is removed, and also (3.108) is modified as follows

T1k T1k
upssk = KP SSk uink − S1k + S2k (3.111)
T2k T2k

To summarize, the dynamic of a generator k represented by the one-axis model, and

equipped with an AVR and a PSS with 2 lead-lag filters is given by

δ̇k = ωk
′
1 Eqk Uk
ω̇k = (Pmk − sin(δk − θk ) − Dk ωk )
Mk x′dk
 
1 xdk ′ xdk − x′dk
Ė ′ qk = ′
Ef k − ′ Eqk + Uk cos(δk − θk ) (3.112)
Tdok xdk x′dk
1 KAk KAk
Ėf k = − Ef k + Ukref − Uk + efuink uink − efs1k S1k + efs2k S2k + efs3k S3k
Tek Tek Tek
Ṡ1k = s1uink uink − s1s1k S1k
Ṡ2k = s2uink uink − s2s1k S1k − s2s2k S2k
Ṡ3k = s3uink uink − s3s1k S1k + s3s2k S2k − s3s3k S3k
46
Chapter 4
Transient stability of an SMIB system
Transient stability analysis of a power system is an extensive and complicated task.
However, it turns out that many of the most important phenomena and mechanisms
can be captured by simple systems. In large and complicated systems it is often
hard to distinguish the fundamental and decisive phenomena from the more irrelevant
ones. It is therefore of importance to study simple systems to get an insight into and
understanding of the basics, that can be used in the analysis of more complex systems.
Therefore, this chapter focuses on transient stability of an SMIB system.
Consider the Single-Machine-Infinite-Bus (SMIB) system shown in Figure 4.1.

Infinite
SG
Bus
I

Figure 4.1. SMIB system.

For the above SMIB system, the following assumptions are made:

1. The classical model is applied to the synchronous generator, i.e. the synchronous
generator is modeled as a constant emf (Eq′ ) behind its transient reactance x′d .

2. The system is lossless and the transmission line is modeled by a series reactance.

3. Voltages and currents are symmetrical.

4. The mechanical power Pm is constant.

5. The voltage at bus N is given by ŪN = UN ∠ θN where both UN and θN are fixed.

Based on the above assumptions, we redraw the SMIB system in Figure 4.1 as shown
in Figure 4.2, where xtot is the sum of the reactances of the transmission line and the
two transformers.

Eq′ ∠ δ U N ∠θ N
jxd′ jxtot Infinite
Bus
I
SG

Figure 4.2. Proposed SMIB system for transient stability study.

The dynamic of this system is given by the swing equation

47
48

δ̇ = ω
1 (4.1)
ω̇ = (Pm − Pe − Dω)
M
where
Eq′ UN
Pe = sin(δ − θN ) = Pemax sin(δ − θN )
x′d + xtot

4.1 Analysis of the swing equation

In order to make a qualitative analysis of the swing equation, some simplifications and
notations will be made as follows. Let
   
x δ
θN = 0 and x = 1 =
x2 ω

Then, the swing equation (4.1) is rewritten as

     
ẋ1 δ̇ ω
ẋ =   =   =  
1
ẋ2 ω̇ M
(Pm − Pemax sin(δ) − Dω)
(4.2)
  
x2 f1 (x)
=  =  = f (x)
1
M
(Pm − Pemax sin(x1 ) − D x2 ) f2 (x)

Thus, we are dealing with a nonlinear system of the form ẋ = f (x) defined in equation
(2.1) with nx = 2. Based on Definition 2.1, the equilibrium points (e.p) of equation
(4.2) are given by
     
f1 (xo ) x2,o 0
f (xo ) =   =   = 
 (4.3)
1
f2 (xo ) M
(Pm − Pemax sin(x1,o ) − D x2,o ) 0

Obviously, x2,o = 0 and x1,o is given by Pm = Pe (x1,o ) = Pemax sin(x1,o ).

In the steady-state (see Definition 2.2), the generator rotates with a speed that cor-
responds to the system frequency (i.e. ω = 0 ⇒ ωg = ωs ), and also the mechanical
power is equal to the electrical power (i.e. Pm = Pe ).
Figure 4.3 shows variations of Pe (δ) versus the rotor angle δ. Due to practical issues,
the interval of interest is 0 ≤ δ ≤ π.
Obviously, for

1. Pm > Pemax , there is no value of x1,o .

 49

Pe(δ)=Pemaxsin(δ)

P
m
Pe(δ)

π/2
0
0 δs δu π
δ

Figure 4.3. Variations of the electrical power versus the rotor angle.

π
2. Pm = Pemax , there is only one value of x1,o , that is x1,o = .
2
3. Pm < Pemax , there are two values of x1,o as shown in Figure 4.3, i.e. x1,o = δs
and x1,o = δu = π − δs .

Since point (3) is dealing with a normal operation it is of interest for the following
analysis. As was mentioned, for point (3) there are two equilibrium points, that is

       
x1,0 δs x1,0 δ
xo = xs = = and xo = xu = = u (4.4)
x2,o 0 x2,o 0

The question remained to be answered is which one is a stable equilibrium point (s.e.p).

4.1.1 Stability of the equilibrium points of the SMIB system

Stability of the equilibrium points (4.4) may be characterized by a similar argumenta-

tion as was given in Example 2.1. It may also be characterized by applying Theorem
2.1 as follows, see also Example 2.2.
The Jacobian of (4.2) at x = xo is given by

   
∂f1 (x) ∂f1 (x)
∂x1 ∂x2 0 1
 
A=  =  (4.5)
∂f2 (x) ∂f2 (x) K D
∂x1 ∂x2
−M −M
x=xo
50

where K = Pemax cos(x1,o ). The eigenvalues of (4.5) are given by

s 2
D D K
λ1,2 =− ± − (4.6)
2M 2M M

By analogy with the solution of Example 2.2, it can be shown that xo is asymptotically
π
stable if 0 < x1,o < . Therefore, xo = xs is asymptotically stable and xo = xu is
2
unstable. Note that D is a positive small constant.
If there is no damping in the system (i.e. D = 0), then the eigenvalues are given by
r
K
λ1,2 = ± − (4.7)
M

By virtue of Theorem 2.1, it can be shown that xo = xu is unstable (show that).

However, Theorem 2.1 cannot characterize the stability of xo = xs when D = 0 (why?).
Therefore, Theorem 2.2 should be applied to characterize the stability of xo = xs .
Using the following energy function [7]
Z δ
1 2
V(x) = WK + WP = M ω + (Pemax sin(x1 ) − Pm ) dx1
2 δs
1 (4.8)
= M ω 2 − Pm (δ − δs ) − Pemax (cos(δ) − cos(δs ))
2
1
= M ω 2 − Pm δ − Pemax cos(δ) + Co
2
it can be shown that xo = xs is stable if D = 0. It can also be shown that xo = xs
is asymptotically stable if D 6= 0 (see Example 2.2). Note that in equation (4.8),
Co = Pm δs + Pemax cos(δs ) is a constant such that V(xs ) = 0.
Stability of δs and δu may also be characterized as follows. Starting with δs (see
Figure 4.4), assume that due to some disturbance δ moves from δs to δ1 at which Pe
is greater than Pm (and therefore ω̇ < 0). This causes the generator to decelerate,
and therefore the rotor angle δ starts to decrease and moves back to δs . Furthermore,
the electric power Pe as a function of δ moves back to Pm . Next, assume that due to
some disturbance δ moves from δs to δ2 at which Pe is less than Pm (and therefore
ω̇ > 0). This causes the generator to accelerate, and therefore the rotor angle δ starts
to increase and moves back to δs . Also, Pe as a function of δ moves back to Pm .
To summarize, when δ = δ1 , Pe is greater than Pm . Then the generator decelerates,
and the rotor angle δ decreases and moves back to δs . As δ passes δs , Pe becomes
less than Pm . Then the generator accelerates. However, due to the generator inertia
(M or H) δ continues decreasing and after a while it swings and moves back to δs .
As δ passes δs again, Pe becomes greater than Pm . Then the generator decelerates.
However, due to the generator inertia (M or H) δ continues increasing and after a
while it swings and moves back to δs . If there is no damping in the system, δ oscillates
around δs . Therefore, δs is a stable e.p (but not asymptotically stable). If there is
 51

damping in the system, then δ will eventually be settled at δs . Therefore, the e.p. will
be asymptotically stable.
Next consider δu . Assume that δ moves from this e.p to δ3 at which ω̇ > 0 (why?).
This causes the generator to accelerate, and the rotor angle δ increases and moves
away from δu . At δ = δ4 , the rotor angle δ decreases and moves away from δu (why?).
Therefore, this e.p is unstable (see also Example 2.1).

P (δ)=P sin(δ) P (δ)=P sin(δ)

e emax
e emax

Pm Pm
Pe(δ)

Pe(δ)

0 0
0 δ2 δs δ1 0 δ4 δu δ3

Figure 4.4. Stable and unstable equilibrium points.

Up to now, the stability of the equilibrium points of the SMIB system was discussed.
Next, the transient stability of the SMIB system will be discussed.
Due to the dynamic and structure of the swing equation (4.2), only transient stability
(or first swing stability) is feasible in the SMIB system, and a physically based definition
of its stability may be given as follows:
Definition 4.1 The SMIB system is transiently stable if the generator remains in
synchronism with the infinite bus after being subjected to a large disturbance. Instability
that may result occurs in the form of increasing rotor angle (or speed) of the generator
leading to its loss of synchronism.
However, a mathematically based definition may be given as follows:
Definition 4.2 The SMIB system is transiently stable if the initial point of the post-
disturbance system lies within the stability region of the stable e.p of the post-disturbance
system.

Example 4.1 Consider the SMIB system shown in Figure 4.5. At time t = tf a three-
phase fault occurs at point F very close to BUS 2 (i.e. it can be assumed that the fault
occurs at Bus 2). At time t = tf + tc the fault is cleared by opening the faulted line.
The system data is given as follows (Sbase = Sng ):
Generator: x′d = 0.15 (pu), H = 4 (s), D = 0, and fs = 50 (Hz)
Transformer: xT 1 = xT 2 = 0.10 (pu)
Line: xL = 0.50 (pu) and BUS N: ŪN = UN ∠ θN = 1∠0 (pu)
52

BUS T BUS 1 jxL BUS 2 BUS N

Infinite
SG jxL F Bus
I jxT 1 jxT 2

Figure 4.5. The SMIB system of Example 4.1.

Prior to the fault, the voltage at BUS 1 was Ū1 = U1 ∠ θ1 = 1∠ 20.4873o (pu). Examine
transient stability of the system for the clearing times tc = 70 (ms), tc = 100 (ms) and
tc = 120 (ms), respectively.

Obviously, the system goes through three states, namely a pre-fault state (t < tf ), a
during-fault state (tf ≤ t < tf + tc ), and a post-fault state (t ≥ tf + tc ). The pre-fault
system is usually in a steady-state. Dynamic of the system in each state is given by
Pre-fault state:
Z
pre pre
ẋ = f (x) = f (xso ) =0 ⇒ x pre
(t) = xso + f pre (x) dt = xso (4.9)

During-fault state:
Z
f f
ẋ = f (x) ⇒ x (t) = x(tf ) + f f (x) dt (4.10)

Post-fault state:
Z
post post
ẋ = f (x) ⇒ x (t) = x(tf + tc ) + f post(x) dt (4.11)

where x(t) is the solution of the system. The question is if the post-fault system with
initial point x(tf + tc ) is transiently stable. To answer this question we must first
identify f pre (x), f f (x), f post(x), and xso based on the given system data.
Pre-fault state:
In this state, the system is in a steady-state. Since Ū1 is known prior to the fault, the
following can be obtained.
Ū1 − ŪN
I¯ = = 1.0162∠ 10.2437o (pu)
j(0.5 ∗ xL + xT 2 )

Ēq′ = Eq′ ∠ δspre = j (x′d + xT 1 ) I¯ + Ū1 = 1.0747∠ 33.9393o (pu)

Pm = Real [Ēq′ I¯∗ ] = 1 (pu) or

" !∗ #
pre pre
Ē ′ − Ū
q N
Pm = Pe (δs ) = Real Ēq′
jxpre
tot

Eq′ UN
= sin(δspre ) = Pemax
pre
sin(δspre )
xpre
tot

= 1.7911 sin(33.9393o) = 1 (pu)

 53

In this state, ẋ = f pre (x) is given by (in f (x), the angle is given in radian)

       
δ̇ ω ω 0
 = = = 
1 1
ω̇ M
(Pm − Pepre) M
pre
(Pm − Pemax sin(δspre )) 0

where
2H
xpre ′
tot = xd + xT 1 + 0.5 ∗ xL + xT 2 and M =
ωs
The stable e.p of the pre-fault system is given by xso = [δspre 0]T
During-fault state:
In this state, because of the fault the voltage at BUS 2 is zero. Thus,

Ēq′ − 0 Ēq′
I¯f = =
j(x′d + xT 1 + 0.5 ∗ xL ) jxftot
" !∗ #  ′ 2
Ē ′ (Eq )
= Real [Ēq′ I¯f ] = Real Ēq′
q
Pef ∗
= Real =0
jxftot −jxftot

and ẋ = f f (x) is given by

     
δ̇ ω ω
 = =  (4.12)
1 Pm
ω̇ M
(Pm − Pef ) M

Post-fault state:
In this state, since the faulted line is removed, there is only one transmission line. We
have then
Ēq′ − ŪN Ēq′ − ŪN
I¯post = =
j(x′d + xT 1 + xL + xT 2 ) jxpost
tot
" !∗ #
Ē ′ − ŪN Eq′ UN
Pepost ¯ ∗
= Real [Ēq Ipost ] = Real Ēq
′ ′ q
= post sin(δ)
jxpost
tot xtot
post
= Pemax sin(δ) = 1.2643 sin(δ)
 
post Pm
Pm = Pemax sin(δspost ) ⇒ δspost = arcsin post = 52.2740o
Pemax
δupost = 180 − δspost , since θN = 0

and ẋ = f post (x) is given by (in f (x), the angle is given in radian)
     
δ̇ ω ω
 = =  (4.13)
1 post 1 post
ω̇ M
(Pm − Pe ) M
(Pm − Pemax sin(δ))
54

The stable e.p of the post-fault system is given by xso = [δspost 0]T , and the unstable e.p
of the post-fault system is given by xuo = [δupost 0]T .
Figure 4.6 illustrates variations of Pepre (δ) and Pepost (δ) as functions of δ. As shown in
the figure, the e.p of the system is the point at which the electric power Pe (δ) intersects
the mechanical power. Obviously, the pre-fault system and the post-fault system have
different equilibrium points. The reason is that the post-fault system has only one
transmission line which results in a higher total reactance than the pre-fault system.

Ppre
e

Ppost
e

P
m
Pe(δ)

0
0 δpre δpost δpost
u
180
s s
δ (deg.)

Figure 4.6. Variations of Pepre and Pepost versus rotor angle δ.

Next, by using some simulation tool (like MATLAB) we are able to study the stability
of the system for the given clearing time tc . Figure 4.7 illustrates phase portrait of the
SMIB system , and variations of the rotor angle δ versus time for tc = 70, tc = 100,
and tc = 120 (ms). The stability region of the stable e.p of the post-fault system is
also shown in the figure (the shaded region).
0.018 160
t =120 (ms)
c

140 t =120 (ms), red line

c
tc=100 tc=100 (ms), blue line
t =70 (ms) 120 t =70 (ms), black line
c c

100
ω (p.u)

δ (deg.)

0 ← xso=(δpost
s
,0) 80
↑ ↑u post
xs =(δpre,0) x =(δ ,0)
o s o u
60

40

20

−0.018 0
0 160 0 1 2 3 4 5 6
δ (deg.) Time (s)

Figure 4.7. Phase portrait, and variations of the rotor angle δ versus time.

Based on Definition 4.2, the SMIB system is transiently stable for this fault with tc = 70
and tc = 100 (ms) since the initial point of the post-fault system lies within the stability
region of the δspost . However, the system is unstable for this fault with tc = 120 (ms)
(why?). Next, in order to study the dynamic performance of the system for the given
fault the case tc = 70 (ms) is discussed below in details.
 55

Figure 4.8 shows variations of Pe (δ) as a function of δ (the Pe − δ figure), and the phase
portrait of the system (the ω − δ figure) during the fault. The shaded region indicates
the stability region of the stable e.p of the post-fault system. In pre-fault state (i.e.
in the steady-state), the system is at xso = [δspre 0]T where Pe (δ) = Pepre(δspre ) = Pm .
When the fault occurs the electric power (as determined above) becomes zero (i.e.
Pe (δ) = Pef = 0), and the dynamic of the system is given by equation (4.12). Therefore,
the generator starts accelerating (i.e. ω is increasing) since ω̇ = PMm > 0, and so does
the rotor angle δ since ω > 0 as shown in the ω − δ figure.

← p1=x(tc)=(δc,ωc)

← xf(t)=(δ(t),ω(t))
P

ω (p.u)
m
P (δ)

0
↑
e

xs =(δpre,0)
o s

f
Pe
0 160
δpre
s
δc δ (deg.)
δ (deg.)

Figure 4.8. Pe − δ curve and phase portrait of the system during fault.

The fault is cleared at clearing time t = tc = 70 (ms) which corresponds to point p1

in the ω − δ figure, and also δ = δc in the Pe − δ figure. The angle δc is termed the
clearing angle which is the corresponding rotor angle to the clearing time tc .
The dynamic of the post-fault system is given by equation (4.13) with the initial point
p1 at which Pm > Pepost as shown in the Pe − δ figure of Figure 4.9. Thus, the generator
continues accelerating since ω̇ = Pm − Pepost > 0, and so does the rotor angle δ since
ω > 0 as shown in the ω − δ figure of Figure 4.9.

p2

p1
Ppost post
e x (t)
Pm
ω (p.u)

p3 xuo
P (δ)

0
xso
e

post 0 160
δc δs δmax δ (deg.)

Figure 4.9. Pe − δ curve and phase portrait of the post-fault system (δc → δmax ).

When δ reaches δspost (i.e. when δ = δspost ), then Pm = Pepost which corresponds to
point p2 in the ω − δ figure at which ω > 0. Therefore, the rotor angle δ continues to
increase, and Pepost as a function of δ follows the direction indicated in the figure. When
56

δ passes δspost , then Pm < Pepost (i.e. ω̇ < 0). Thus, the generator starts decelerating,
and ω decreases until it becomes zero which corresponds to point p3 in the ω − δ figure.
However, since ω is positive between points p2 and p3 the rotor angle δ continues to
increase. Note that δ̇ = ω. Since ω = 0 at point p3 , then δ̇ is also zero at this point
which means the rotor angle δ has reached its maximum (indicated as δmax in Figure
4.9, see also the δ−time figure of Figure 4.7).
Since Pm < Pepost at δ = δmax , the generator continues to decelerate which results in
decreasing of ω. When ω decreases and passes zero the rotor angle δ starts decreasing
(or swings back), and Pepost follows the direction indicated in Figure 4.10. When δ
reaches δspost , then Pm = Pepost which corresponds to point p4 at which ω < 0. Therefore,
the rotor angle δ and Pepost continue to decrease. When δ passes δspost , then Pm > Pepost
(i.e. ω̇ > 0). Thus, the generator starts accelerating, and ω increases until it becomes
zero which corresponds to point p5 . However, since ω is negative between points p4 and
p5 the rotor angle δ continues to decrease until it reaches its minimum (indicated as
δmin in Figure 4.10 which corresponds to point p5 , see also the δ−time figure of Figure
4.7).

Ppost
e

Pm
ω (p.u)

p5 p3 xuo
Pe(δ)

0
xso
xpost(t)

p4

0 160
δmin δpost δmax δ (deg.)
s

Figure 4.10. Pe − δ curve and phase portrait of the post-fault system (δmax → δmin ).

As shown in Figure 4.11, Pm > Pepost at δ = δmin . Therefore, the generator continues to
accelerate which results in increasing of ω, and the system will have a similar behavior
as described based on Figure 4.9.

p2

Ppost
e
xpost(t)
P
ω (p.u)

m
p5 p xu
Pe(δ)

0 s 3 o
xo

0 160
δ
min
δpost
s
δ
max δ (deg.)

Figure 4.11. Pe − δ curve and phase portrait of the post-fault system (δmin → δmax ).

Since there is no damping in the system (i.e. D = 0) the system trajectory will oscillate
 57

around the stable e.p of the post-fault system as shown in Figure 4.12. Therefore, the
system is transiently stable (or first-swing stable) for this fault with tc = 70 (ms). A
similar behavior can be found in case tc = 100 (ms).

p2

post
ω (p.u) x (t)

0 p p xu
5 xso 3 o

p4

0 160
δ (deg.)

Figure 4.12. The dynamic behavior of the SMIB system for the given fault
with tc = 70 (ms).
Figure 4.13 shows the dynamic behavior of the SMIB system for the given fault with
tc = 120 (ms). The fault is cleared at point p1 where δ = δc . Since Pm > Pepost at
this point the generator accelerates which results in increasing of ω, and also the rotor
angle δ (since ω > 0). When δ passes δspost , then Pm < Pepost . Thus, the generator
starts decelerating, and ω decreases. However, since ω is still positive the rotor angle
increases, and Pepost follows the direction indicated in the figure until it reaches the
intersection point between Pepost and Pm which corresponds to point p2 at which δ =
δupost and ω > 0. Therefore, the rotor angle δ continues to increase, and Pepost follows
the direction indicated in the figure. When δ passes δupost , then Pm > Pepost. Thus, the
generator starts accelerating which results in increasing of ω. Since ω > 0 the rotor
angle δ also continues to increase which leads to transient instability (or first-swing
instability). In power system literature, it is often said that the generator falls out of
step or loses synchronism (see also Definition 4.1).

p1 → post
←x (t)

p2
post
Pe ↓
← xf(t)
P
ω (p.u)

m
s,post
← xo ← xu,post
Pe(δ)

0 o
↑
xs,pre
o

Pf
e
0 160
δpre
s
δ
c
δ (deg.) δpost
u δ (deg.)

Figure 4.13. The dynamic behavior of the SMIB system for the given fault
with tc = 120 (ms).

Next, assume that the fault is cleared at time tc = tcc at which the trajectory of the
58

during-fault system intersects the stability boundary of the stable e.p of the post-fault
system as shown in Figure 4.14.

0.018
p
1
↓

post
Ppost ←x (t)
e

f
← x (t)
P
m

ω (p.u)
Pe(δ)

s,post
0 ← xo ← xu,post
o
↑
s,pre
xo

Pf
e −0.018
pre post 0 160
δ δ δ (deg.) δu δ (deg.)
s cc

Figure 4.14. The dynamic behavior of the SMIB system for the given fault with tc = tcc .

In the figure, the intersection point is indicated as p1 at which δ = δcc . Since p1 (which
is the initial point of the post-fault system) lies on the stability boundary, the trajectory
of the post-fault system converges to xu,post
o as time goes to infinity (see Definition 2.5).
Time tcc is termed critical clearing time, and δcc is termed critical clearing angle cor-
responding to tcc .
Obviously, for any clearing time tc which is less than the critical clearing time (i.e.
tc < tcc ) the system will be transiently stable (or first-swing stable) since the initial
point (p1 ) of the post-fault system will lie within the stability region of the stable
e.p of the post-fault system. Consequently, if tc ≥ tcc the system will be transiently
unstable (or first-swing unstable). Thus, the transient stability of the SMIB system can
be stated for a given clearing time tc without any numerical simulation if the critical
clearing time tcc is known.
For an SMIB system, the critical clearing angle δcc can analytically be determined by
applying Equal Area Criterion (EAC). When the critical clearing angle δcc is known,
the critical clearing time tcc can also analytically be determined if the electric power
during fault is constant. However, if the electric power during fault is a function of
the rotor angle δ, the critical clearing time tcc can be determined only by numerical
integration.

4.2 Equal Area Criterion (EAC)

From Figure 4.7 (the δ-time figure) it may be concluded that the SMIB system is
transiently stable (or first swing stable) if ω becomes zero. Since δ̇ = ω, the condition
ω = 0 implies that the rotor angle reaches a maximum, and then swings back (hence the
name first-swing stability) as shown in Figure 4.15 for the cases tc = 70 and tc = 100
(ms). This observation may be expressed as
 59

A necessary condition for transient stability (or first-swing stability) of the SMIB sys-
dδ
tem is that at some time (tm ), = ω(tm ) = 0.
dt

t =70 (ms) t =100 (ms)

c c
140 140
dδ
dt =ω= 0
120 120
↓
dδ
100 dt =ω= 0 100
↓

80 80
δ (deg.)

δ (deg.)
60 60

40 40

20 20

0 0
0 1 2 3 4 5 6 0 1 2 3 4 5 6
Time (s) Time (s)

Figure 4.15. Variations of δ versus time.

Now, consider the swing equation

dω 1
= (Pm − Pe (δ))
dt M
dδ
Multiplying both sides of the swing equation by = ω, the following is obtained
dt
dω 1 dδ
ω = (Pm − Pe (δ)) (4.14)
dt M dt

Integration of the (4.14) gives

Z ω(tm ) Z δmax
1
ω dω = (Pm − Pe (δ)) dδ ⇒
0 M δspre
Z δmax
1 1
(ω(tm ))2 = (Pm − Pe (δ)) dδ ⇒
2 M δspre
s Z δmax
2
ω(tm ) = (Pm − Pe (δ)) dδ
M δspre

The stability condition is that ω(tm ) = 0 which leads to

Z δmax
(Pm − Pe (δ)) dδ = 0 (4.15)
δspre

The SMIB system is transiently stable (or first-swing stable) if there exists an angle
δ = δmax such that equation (4.15) holds.
60

Let δ = δc be clearing angle which corresponds to clearing time tc . Then, equation

(4.15) can be rewritten as
Z δc Z δmax
f
(Pm − Pe ) dδ + (Pm − Pepost) dδ = 0 ⇒
δspre δc
Z δc Zδmax
(Pm − Pef ) dδ = (Pepost − Pm ) dδ ⇒ (4.16)
δspre δc
Aa = Ad
where Z Z
δc δmax
Aa = (Pm − Pef ) dδ and Ad = (Pepost − Pm ) dδ (4.17)
δspre δc

Thus, the SMIB system is transiently stable (or first-swing stable) if there exists an
angle δ = δmax such that Aa = Ad (hence the name equal area criterion) as illustrated
in Figure 4.16, (Pef is assumed zero).

Ppre
e

Ppost
e

Ad
Pm
Pe(δ)

Aa

Pfe
0
0 δpre δc δmax δpost 180
s
δ u

Figure 4.16. The equal area criterion (clearing angle, δc ).

But, what are Aa and Ad ? (or what do they represent?)

Consider again equation (4.14) which is rewritten as
dω dδ
= (Pm − Pe (δ))
Mω
dt dt
Integrating the above equation, the following is obtained
Z Z
dω dδ
Mω dt = (Pm − Pe ) dt
dt dt
where Z ω2 Z ω2
dω 1
L.H.S = Mω dt = Mω dω = M[ω 2 ]ωω21
ω1 dt ω1 2
and
Z δ2 Z δ2
dδ
R.H.S = (Pm − Pe ) dt = (Pm − Pe )dδ
δ1 dt δ1
 61

For δ1 = δspre , δ2 = δc , ω1 = 0 and ω2 = ωc , i.e. during the fault, we have

1 1
L.H.S = M(ω22 − ω12) = Mωc2
2 2
Z δ2 Z δ2
f
R.H.S = (Pm − Pe )dδ = Pm dδ = Pm (δc − δspre ) = Aa
δ1 δ1
1
Aa = Mωc2
2
that is

• Aa represents the kinetic energy injected into the system during the fault. It is
also called accelerating area.

In a similar way with δ1 = δc , δ2 = δmax , ω1 = ωc , ω2 = ω(tm ) = 0 and Pe = Pepost , it

can be shown that

• Ad represents the ability of the post-fault system to absorb energy, i.e. potential
energy. It is also called decelerating area.
Setting δmax = δupost , the maximum value of Ad is obtained which is denoted by Admax
as shown in Figure 4.17. Then, Aa = Admax gives the critical angle δcc , and the SMIB
system is transiently stable if Aa < Admax (or δc < δcc ).

Ppre
e

Ppost
e

Admax
Pm
Pe(δ)

Aa

Pfe
0
0 δpre δc δmax= δpost 180
s
δ u

Figure 4.17. The equal area criterion (critical clearing angle, δcc ).

In general, an SMIB system is transiently stable if A1 < A2 (or δc < δcc ) where
Z δc
A1 = (Pm − Pef ) dδ
pre
δs
Z δmax (4.18)
post
A2 = (Pe − Pm ) dδ
δc

with δmax = δupost . Furthermore, A1 = A2 gives δcc .

62

Example 4.2 Consider the system and the case in Example 4.1. Determine the crit-
ical clearing angle (δcc ), the critical clearing time (tcc ).
The critical clearing angle δcc is given by A1 = A2 where
Z δcc
A1 = (Pm − Pef )dδ = Pm [δ]δδcc pre
pre = Pm (δcc − δs
s
)
δspre
Z δmax
A2 = post
(Pepost − Pm )dδ = −Pemax [cos(δmax ) − cos(δcc )] − Pm (δmax − δcc )
δcc

Pm pre
A1 = A2 ⇒ δcc = arccos[cos(δmax ) + post (δmax − δs )] = 0.8192(rad.) = 46.9370(deg.)
Pemax
Since the electric power during the fault is constant (Pef = 0), the critical clearing time
tcc can be obtained as follows. During the fault, the dynamic of the system is given by
dω 1 Pm
δ̈ = = (Pm − Pef ) =
dt M M
Since Pm /M is a positive constant, the system has a constant acceleration during the
fault. The critical clearing time tcc can be obtained by integrating the above equation
twice. The first integration gives:

Z Z
Pm Pm dδ Pm
dω = dt ⇒ dω = dt ⇒ ω = = t
M M dt M
The second integration gives:
Z δcc Z t=tcc r
Pm Pm 2 2M
dδ = pre
t dt ⇒ δcc −δs = tcc ⇒ tcc = (δcc − δspre ) ≈ 107.49(ms)
pre
δs t=0 M 2M P m

The system is stable for this fault if tc < tcc as shown in Fig. 4.18.
160

tc=107.5 (ms), transiently unstable →

140

120

100
δ (deg.)

80 tc=107.48 (ms), transiently stable →

60

40

20

0
0 0.5 1 1.5 2 2.5 3
Time (s)

Figure 4.18. Critical clearing time, tcc = 107.49 (ms).

 63

Gen3

T3

BUS 3
S L3
BUS 1 BUS 4 BUS 2

Gen1 Gen2
T1 S L1 SL4 SL2 T2

Figure 4.19. The 3-machine test system.

Example 4.3 Consider the lossless 3-machine test system shown in Figure 4.19. The
system data is given in Appendix A (Section A.3).
a) Based on the system data given in Appendix A (Section A.3), perform the power
flow calculations.
b) As seen from BUS 1, find a Thévenin equivalent model of the system as shown in
Figure 4.20 such that with Gen 1 connected to this equivalent system we get the same
Pg1 , Qg1 and Ū1 as obtained in task a).

BUS 1 BUS 1 BUS 2

Eq¢Ð d U Th ÐqTh Eq¢Ð d 0.3 ZTh U Th ÐqTh
jxd¢ ZTh jxd¢ 0.85 ZTh
Þ ITh 0.3 ZTh
ITh
Gen1 Gen1
a) b)

Figure 4.20. The Thévenin equivalent model of the test system as seen from BUS1.

Now consider the system shown in Figure 4.20 b), where x′d = x′d1 + xT 1 . Assume
that ŪT h and Z̄T h are fixed for the following cases. For each case find the critical
clearing angle and time (correct to four decimal places). Also, plot variations of δ for
tc = tcc − 0.0001 (stable case) and tc = tcc + 0.0001 (unstable case). Furthermore, plot
variations of Pepre, Pef and Pepost as a function of δ, and identify Aa and Admax in the
figure.
c) Case 1: A three-phase fault occurs at BUS 1, and the fault is cleared by opening one
of the lines between BUS 1 and BUS 2.
d) Case 2: A three-phase fault occurs at BUS 2, and the fault is cleared by opening
one of the lines between BUS 1 and BUS 2.
e) Case 3: A three-phase fault occurs in the middle of one of the lines between BUS 1
and BUS 2, and the fault is cleared by opening the faulted line.
64

a) Running the Load Flow Calculations (LFC), (see the system data in Appendix A
and the MATLAB files for ”Project D1” in Canvas), the following are obtained (all
obtained angles are expressed in degrees).
Ū10 = 1∠0 , Ū20 = 1∠ − 27.9852 , Ū30 = 1∠ − 27.9852 , Ū40 = 0.9545∠ − 31.5891
and
Pg1 = 1.0000 , Qg1 = 0.3738
Also, in a similar way as shown in Example 4.1 (see also point ”4.” in Example 5.1),
we find that
E¯q1
′ ′
= Eq1 pre
∠δ1s = 1.0932∠10.5414
E¯q2
′ ′
= Eq2 pre
∠δ2s = 1.0491∠−24.7065
E¯q3
′ ′
= Eq3 pre
∠δ3s = 1.0324∠−25.7646

b) To fulfill the requirements in task b), the Thévenin equivalent model is found as
follows

1. After LFC, convert all given loads to impedance loads and disconnect the selected
generator from its terminal bus as shown in Figure 4.21 where xT i is included in
x′di and Z̄L4 = U420 /S̄L4
∗
.

Eq¢ 3
BUS 6
jxd¢ 3
I6
BUS 3

jx34
BUS 1 BUS 4 BUS 2 Eq¢ 2
Eq¢1
jxd¢ 1 jx14 jx24 jxd¢ 2
I5 BUS 5
ZL4

Figure 4.21. The 3-machine test system for applying the Thévenin theorem.

2. Then, in a similar way as discussed in the course EG2100 and by keeping Ē ′ q2

and Ē ′ q3 constant, we find the Thévenin voltage (i.e. the voltage at BUS 1) by
   −1  ¯′   −1  ¯′ 
I¯5 Z(5, 5) Z(5, 6) Eq2 0.0220 − j0.2866 0.0146 + j0.0756 Eq2
= ¯ =
I¯6 Z(6, 5) Z(6, 6) E ′
q3 0.0146 + j0.0756 0.0098 − j0.2829 E¯q3
′
 
4.9167∠55.8516
=
4.9287∠57.0998
ŪT h = Ū1 = Z(1, 5) I¯5 + Z(1, 6) I¯6 = 0.9476∠−44.6127
where the angles are given in degrees, Z = Y −1 and Y is the admittance matrix
of the system shown in Figure 4.21.
 65

3. The Thévenin impedance Z̄T h can be found when all voltage sources (i.e. E¯q2
′
and
¯′
Eq3 ) are short circuited. This can be done by removing all the rows and columns
corresponding to BUS 5 and BUS 6 in the admittance matrix Y . Thus,
Y∆ = Y (1 : 4, 1 : 4)
Z∆ = Y∆−1
Z̄T h = Z∆ (1, 1) = RT h + jXT h = 0.0673 + j0.6906

4. Next, connect Gen 1 to BUS 1 as shown in Figure 4.20 where E¯q1 ′

has been
obtained in a) above. With a correct obtained ŪT h and Z̄T h , the voltage at BUS
1 in Figure 4.20 must be equal to Ū1 obtained in a), and also the obtained Pg1
and Qg1 from the load flow calculations in a) must be equal to Ū1 I¯T∗ h in Figure
4.20.

c) As shown in Exercise 1 and Exercise 3 (see the file ”Exercise EG2110.pdf”), the
electric power can be expressed by
Pe = K1 + K2 sin(δ − K3 )
where, K1 , K2 , and K3 can be easily determined based on the known data. For all
three cases, the following are obtained (see Exercise 17)
Pm = Pg1 = 1
K1pre = 0.1008 ; K2pre = 1.1598 ; K3pre = −0.7032
K1post = 0.0930 ; K2post = 1.0388 ; K3post = −0.7010
pre
δspre = δ1s = 10.5414 (deg.)
δspost = 20.6665 (deg.)
δmax = 79.0081 (deg.)

Furthermore, δcc can be found by

A1 = (Pm − K1f )(δcc − δspre ) + K2f [cos(δcc − K3f ) − cos(δspre − K3f )]
A2 = (K1post − Pm )(δmax − δcc ) − K2post [cos(δmax − K3post ) − cos(δcc − K3post )]
g(δcc ) = A1 − A2 = 0
where g(δcc ) = 0 is solved for δcc by applying numerical methods such as Newton-
Raphson method (see Appendix A and the MATLAB files for ”Project D1”).
In this case (case 1), we have
K1f = 0 , K2f = 0 ⇒ δcc = 15.5200 (deg.)

Since Pef = 0 is constant, the critical clearing time can be determined in a similar way
as shown in Example 4.2. Thus,
s r
2M pre 2M
tcc = f
(δcc − δs ) = (δcc − δspre ) = 0.0744 (s)
Pm − Pe P m
66

120
Case 1 Ppre
e
Case 1
100 Ppost
e

Pm A dmax
80 1

60
δ (deg.)

Pe(δ)
40

δ post
s
Aa
20

δ pre
s
0

-20 Pfe
0 1 2 3 4 5 6 0
Time (s) δ pre
s
δ cc δ post
s
δ δ max= δ u

Figure 4.22. Variations of δ and Pe for case 1.

d) In this case (case 2), we have

K1f = 0.1307 , K2f = 0 ⇒ δcc = 16.3096 (deg.)

Since Pef = K1f = 0.1307 is constant, the critical clearing time can be determined in a
similar way as in c). Thus,
s
2M
tcc = (δcc − δspre ) = 0.0859 (s)
Pm − Pef

120
Case 2 Ppre
e
Case 2
100 Ppost
e
A dmax
80 1

60
δ (deg.)

Pe(δ)

40

δ post
s
Aa
20

δ pre
s
0
Pfe

-20
0 1 2 3 4 5 6 0
Time (s) δ pre
s
δ cc δ post
s
δ δ max= δ u

Figure 4.23. Variations of δ and Pe for case 2.

e) In this case (case 3), we have

K1f = 0.1165 , K2f = 0.1458 , K3f = −0.7516 ⇒ δcc = 17.1689 (deg.)

 67

Since Pef = K1f + K2f sin(δ − K3f ) is a function of δ (i.e. it is not constant), the critical
clearing time should numerically be determined (for instance by trial and error). For
this case, it was found that
tcc = 0.0981 (s)

120
Case 3 Ppre
e
Case 3
100 Ppost
e
A dmax
80 1

60
δ (deg.)

Pe(δ)
40

δ post
s
Aa
20

δ pre
s
Pfe
0

-20
0 1 2 3 4 5 6 0
Time (s) δ pre
s
δ ccδ post
s
δ δ max= δ u

Figure 4.24. Variations of δ and Pe for case 3.

4.3 Transient Energy Function (TEF) method

Another method to estimate the critical clearing time is known as Transient Energy
Function (TEF) method. This method (also known as the direct method) is based on
Lyapunov’s direct method which gives an estimate of the actual stability region. The
simplest estimate is provided by the set

Ωc = {x ∈ Rnx : V(x) < c}

i.e., for any initial point within Ωc the system is stable, and the system trajectory will
tend to the stable equilibrium point xs .

Example 4.4 Consider the following nonlinear system

ẋ1 = −8x1 + 2x1 x22

ẋ2 = −18x2 + 2x2 x21

By applying the direct method give an estimate of the actual stability region.

By setting ẋ1 = ẋ2 = 0, we first find the equilibrium points of the system. It can easily
be shown that the equilibrium points are

x01 = (−3, −2) , x02 = (3, 2) , x03 = (−3, 2) , x04 = (3, −2) , x05 = (0, 0)
68

Based on Theorem 2.1, x01 , x02 , x03 , x04 are unstable, and x05 = xs is (asymptotically)
stable.
Consider the positive definite function V(x) = x21 + x22 whose time derivative along the
trajectories of the system is given by

V̇ = −4(4x21 + 9x22 − 2x21 x22 )

Since our interest is in estimating the stability region, we need to determine a domain
D about the origin where V̇ is negative definite, and a set Ωc ⊂ D which is bounded.
We can find c by minimizing V subject to V̇ = 0. Doing this, we easily find c = 12.5,
see Section B.1 in Appendix B. Thus, V < 12.5 is the estimate of the stability region
as shown in Figure 4.25. Note that V = c = 12.5 is a level surface of Lyapunov function
V(x) = x21 + x22 .

+
x2

0 x
s

0
x
1

Figure 4.25. Estimate of the stability region of xs .

In the figure, the stability boundary of the actual stability region is indicated in blue,
and the stability boundary of the estimated stability region (i.e. Ωc = {x ∈ Rnx :
V(x) < c}) is indicated in red.
Let Vcr = c be the critical level surface, and xini be an initial point whose position
is known. If V(xini ) < Vcr , then the initial point lies within the estimated stability
region, and the system is therefore predicted by the direct method as stable. However,
if V(xini ) ≥ Vcr , then the system is predicted as unstable.
Let xini = (2, 2) be an initial point. At this point V(xini ) = 8 which is less than Vcr =
12.5, and based on the direct method the system with this initial point is predicted as
stable. Next, let xini = (−1, 4) be an initial point. At this point V(xini ) = 17 which is
greater than Vcr = 12.5. Thus by the direct method, the system with this initial point
is predicted as unstable.
 69

The initial points xini = (2, 2) and xini = (−1, 4) are indicated with “+” and “*”,
respectively, in the figure. Obviously, xini = (2, 2) lies within Ωc and the direct method
correctly predicts the stability of the system. The initial point xini = (−1, 4) does
not lie within Ωc , but it lies within the actual stability region that is the system
will be stable with this initial point. However, the direct method incorrectly predicts
the instability of the system. This is a disadvantage of this method which gives a
conservative prediction. However, an important feature of this method is that it does
not predict an unstable case as a stable case.
This method has received considerable attention for assessment of power system tran-
sient stability. In power system literature this method is more known as the Transient
Energy Function (TEF) method, since energy functions have been used as Lyapunov
function candidates. The energy function for a power system has normally the form

V(x) = Wk + Wp + C0 (4.19)

where, Wk and Wp are known as kinetic and potential energy, respectively. C0 is a

constant such that V(xs ) = 0 at the stable equilibrium point.
Using the TEF method, the assessment of transient stability of a power system after a
fault is performed by the following steps [7]:

1. Knowing the structure of the post-fault system, compute the stable equilibrium
point (xs ) of the post-fault system.

2. Formulate the energy function of the post-fault system, and calculate Vcr which
is the critical energy function.

3. Simulate the system during the fault (i.e. the on-fault system), and compute xtc
at the specified clearing time tc . The point xtc is considered as the initial point
of the post-fault system.

4. Calculate V(xtc ). If it is less than Vcr , the system is stable.

5. The critical clearing time can also be computed by simulating the on-fault system
until at a time tcc , V(xtcc ) = Vcr . This time is the critical clearing time.

Example 4.5 Consider the energy function below.

1  
V(x) = Mω 2 − (Pm − K1post )δ + K2post cos(δ − K3post ) + C0
2 (4.20)
C0 =(Pm − K1post )δspost + K2post cos(δspost − K3post )

a) Show that the function V(x) in (4.20) satisfies the conditions of Theorem 2.2 for
x0 = xpost
s .

b) By using the function V(x) in (4.20), find (or estimate) the critical clearing time
of each case in Example 4.3.
70

a) It can be shown in a similar way as described in Example 2.2.

b) The critical clearing time for each case can be found as follows.

• Find Vcr which is given by

 
Vcr = − (Pm − K1post )δmax + K2post cos(δmax − K3post ) + C0

• Simulate the system during the fault with appropriate clearing time tc and inte-
gration time step, see the MATLAB files for ”Project D1”.

• Let the vectors δf and ωf be the simulation results during the fault. Then,
substitute these results in V(x) in (4.20) to obtain the vector

1  
V(x) = Mωf2 − (Pm − K1post )δf + K2post cos(δf − K3post ) + C0
2

• The intersection between V(x) and Vcr gives the estimated critical clearing time.

Case 1 Case 2

V cr V cr
V(x)

V(x)

0.0744 0.0859
Time (s) Time (s)

Case 3

V cr
V(x)

0.0981
Time (s)

Figure 4.26. Estimation of critical clearing time based on TEF.

 71

4.4 Transient stability enhancement

In this section, some of the most common methods to improve the transient stability
of power systems will be presented. For the purpose of illustration how the proposed
methods enhance the transient stability, these methods will be applied to an SMIB
system. However, it is to be noted that these methods are general, and can be applied
to any real multi-machine power system.
In the previous section, it has been shown that the SMIB system is transiently stable
for a given fault and a given clearing time tc , if tc < tcc (or A1 < A2 or V(xc ) < Vcr ).
Using this relationship, a stability measure may be defined as follows

tcc − tc tc
Mt = 100 % = (1 − ) 100 %
tcc tcc
Vcr − V(xc ) V(xc )
MV = 100 % = (1 − ) 100 % (4.21)
Vcr Vcr
A2 − A1 A1
MA = 100 % = (1 − ) 100 %
A2 A2

Increasing Mt (or MA or MV ) results in enhancing the transient stability. Obviously,

an important way to improve the transient stability is the use of protection equipments
and circuit breakers that quickly detect and clear the faults (which results in shortening
of tc or reduction of A1 ), and also the use of automatic line re-closing following the
fault clearing (which results in enlargement of A2 as shown in Exercises).
Consider the SMIB system shown in Figure 4.27. In this study, the clearing time is
specified based on the properties of protection equipments and circuit breakers. The
objective of the proposed methods is therefore to extend the critical clearing time
(which is similar to enlargement of A2 and/or reduction of A1 ).

U1 U3 U2 U N ∠0
jx L1 jx L 2
Infinite
SG
Bus
jxt1 jxt 2

Figure 4.27. An SMIB system.

For this system H = 4, x′d = 0.15, Pm = 1, xt1 = xt2 = 0.1, xL1 = xL2 = 0.2 and
UN = 1 (all in (pu)). A three-phase fault occurs at bus 2, and the fault is cleared after
100 (ms) (i.e. tc = 0.1 (s)). Prior to the fault, U1 = 1 (pu). Since Pef = 0 during this
fault, A1 = A2 gives the critical clearing time
r
2M
tcc = (δcc − δspre ) = 0.1458 (s)
Pm

Therefore, Mt = 31.4%.
72

Braking resistors
One method to improve the transient stability is to switch a shunt resistor (normally
close to the generator) for a short time following a fault. This action is known as braking
resistors by which the rotor acceleration due to the fault will be braked. Switching a
shunt resistor R at bus 1 during the fault, then Pef 6= 0 (but a positive constant), and
Anew
1 = Anew
2 gives s
2M new − δ pre )
tnew
cc = f
(δcc s
Pm − Pe
For a clearing angle equal to δcc when Pef = 0, the accelerating area with Pef 6= 0
will be less than A1 with Pef = 0. To make Anew 1 = Anew new
2 , then A2 must be less
than A2 which implies that δcc must be greater than δcc . Furthermore, P 2M
new 2M
f > P .
m −Pe m
Therefore, tnew
cc > tcc which implies that M tR > M t .
For R = 1 (pu), it can be shown that tnew
cc = 0.2370 (s), and MtR = 57.8%.

Series capacitor compensation

Another method to enhance transient stability is reduction of transmission system
reactance. In the SMIB system,

Eq′ UN
Pe = ′ sin(δ) = Pemax sin(δ)
xd + xt1 + xt2 + xL1 + xL2

By compensating each line with a series capacitor, Pemax becomes higher which results
in enlargement of A2 .
Let each line be compensated with 50% that is xnew new
L1 = xL2 = 0.2 − xc = 0.2 ∗ (1 − 0.5).
Thus, by this method tnew
cc = 0.2050 (s), and Mtxc = 51.2%.

Shunt capacitor compensation

The main function of shunt capacitor compensation is to keep the voltage profile of the
(heavily loaded) transmission system within acceptable levels (i.e. close to the nominal
value). This compensation also increases the maximum power transfer capability which
results in enlargement of A2 .
Let a shunt capacitor with Bsh = 0.58 (pu) be installed at bus 3 to keep the steady-
state voltage magnitude of this bus at 1 (pu).
Thus, by this compensation tnew
cc = 0.1505 (s), and MtBsh = 33.5%.

Obviously, compared to the shunt capacitor compensation, the braking resistors and
series capacitor compensation are more effective of achieving transient stability en-
hancement.
Other methods
There are other actions that not only significantly enhance transient stability, they also
improve damping of electromechanical oscillations, for example high-speed excitation
system with Automatic Voltage Regulator (AVR) and Power System Stabilizer (PSS)
which will be discussed in this compendium, or power electronics based controllable
devices with appropriate control strategies.
Chapter 5
Dynamic modeling of multi-machine power
systems
In Chapter 4, the dynamic of a very simple power system (the SMIB system) has been
derived and studied. This simple model has of course a limited application to realistic
power systems. The analysis of such a simple model is primary motivated by the
insight it gives. In this chapter, dynamic modeling of a multi-machine power system is
however presented. It will be shown that depending on load modeling this dynamic will
be described by either a set of differential-algebraic equations (DAE) or only by a set
of differential equations. The dynamic model given by differential-algebraic equations
is termed as Structure Preserving Model (SPM), and the dynamic model given by
differential equations is termed as Reduced Network Model (RNM).
Before deriving the dynamic model of a multi-machine power system, we start this
chapter with presenting different load models.

5.1 Load modeling

As presented in [8], the term “load” can have several meanings. In this compendium
the following definition of the load is however of concern

• “A portion of the system that is not explicitly represented in a system model, but
rather is treated as if it were single power-consuming power device connected to
a bus in the system model” [8].

Based on the above definition, the load at a (high voltage) bus represents the aggre-
gation of hundreds or thousands of individual commercial, industrial and residential
power-consuming devices such as motors, heating, lighting, and electrical appliances
as shown in Figure 5.1.
This aggregate (or composite) load model may be represented by static or dynamic
load models, or a combination of both as described in (5.1)

S̄L = κ S̄st + (1 − κ) S̄dyn where 0 ≤ κ ≤ 1 (5.1)

where, κ is the fraction of the load represented by static load model. S̄L = ŪL I¯L∗ =
PL + j QL is a mathematical representation of the composite load model. It gives the
relationship between bus voltage and current flowing into the (composite) load.

73
74

Transmission Transmission
system system

High voltage bus UL High voltage bus UL

IL IL

S L = U L I L* Substation S L = U L I L*
transformer
Power-consuming

Static Dynamic
Load Load
devices

Capacitor

Changers
Commertial

Residential
Banks

Industrial

Tap

Figure 5.1. A composite load model for representing physical loads.

5.1.1 Static load model

This model expresses the characteristics of the load at any instant of time as functions of
the bus voltage magnitude and frequency at the same instant [8]. These characteristics
are mathematically described by only algebraic equations. This model has usually been
used for transient stability analysis.
The exponential and ZIP models are two types of the static load models which have
been widely used to represent the voltage dependency of loads.
In the exponential model, active and reactive components of the load are expressed as
(it is assumed that κ = 1, i.e. PL = Pst and QL = Qst )
 mp  mq
UL UL
PL = PEXP = PL0 and QL = QEXP = QL0 (5.2)
UL0 UL0
where, UL is the actual bus voltage magnitude, UL0 is the initial value of the voltage,
and PL0 and QL0 are the active and reactive powers at UL0 . The voltage exponents mp
and mq represents the parameters of this model.

• With mp = mq = 0, the model represents constant power characteristic.

• With mp = mq = 1, the model represents constant current characteristic.
• With mp = mq = 2, the model represents constant impedance characteristic.

Based on the nature of the composite load characteristics at a given bus these exponents
may have different values.
 75

In the ZIP model, active and reactive components of the load are expressed as
"  2   #
UL UL
PL = PZIP = PL0 kpz + kpi + kpp
UL0 UL0
"  2   # (5.3)
UL UL
QL = QZIP = QL0 kqz + kqi + kqp
UL0 UL0

This model is composed of constant impedance (Z), constant current (I), and con-
stant power (P) components. The parameters kp and kq define the fraction of each
component. Note that kpz + kpi + kpp = kqz + kqi + kqp = 1.
A more general representation of the static loads is given by (5.4) where the frequency
dependency of load characteristics is also included [9]-[10].
2  mpk !
X UL
PL = PZIP + κ PL0 kp k (1 + Dpk ∆f )
k=1
UL0
2  mqk ! (5.4)
X UL
QL = QZIP + κ QL0 kq k (1 + Dqk ∆f )
k=1
UL0

where Dpk and Dqk are damping constants, and ∆f = f − fs is the bus frequency
deviation (f is the actual frequency of the bus, and fs is the nominal frequency of the
bus). It is obvious that the load models (5.2)-(5.3) are derivatives of the general model
(5.4) in which kpz + kpi + kpp + kp1 + kp2 = kqz + kqi + kqp + kq1 + kq2 = 1.

5.1.2 Dynamic load model

It is well known that load characteristics have a significant impact on power system
dynamics. Therefore, accurate load modeling is vitally important for power system
utilities to predict more precisely the power system operating limits and stability mar-
gins. Thus, in many stability studies such as long-term stability and voltage stability
it is necessary to account for the dynamics of loads.
Electrical motors consume a large amount of the total electrical energy supplied by a
power system, and a large number of these motors are induction motors which affect
damping of oscillations. Thus, in addition to studies of long-term stability and voltage
stability, for improving damping prediction it has also been recommended that major
blocks of induction motor load should be represented by dynamic models including
both inertial and rotor flux dynamics (known as third-order model) [8]-[10]. For a
composite load, the motor components are aggregated into a single dynamic induction
motor model as shown in Figure 5.2 which has been derived by applying an appro-
priate transformation of phase variables into components along rotating axes. For an
induction machine the preferred reference frame is one with the axes rotating at syn-
chronous speed [11], i.e. the stator and rotor quantities are transferred to a reference
frame which rotates at synchronous speed ωs .
76

Is jx ′

Us E′

Figure 5.2. Induction motor transient-equivalent circuit (third-order model).

In Figure 5.2, x′ is the transient reactance (the stator resistance has been neglected),
and based on dq-components of the new reference frame we have the following (all
variables are expressed in (pu)):
Ūs = Uq + jUd
I¯s = Iq + jId
Ē ′ = Eq + jEd

Moreover, from the figure the following can be obtained

Ud − Ed′ Uq − Eq′
Iq = and Id = − (5.5)
x′ x′
Let xm be the magnetizing reactance, rr be the rotor resistance, and xs and xr be the
leakage reactances of the stator and the rotor, respectively. Then, it can be shown that
based on the dq-components of the new reference frame the dynamic of this induction
motor is given by [4]
1
ṡ = (Tm − Te )
2H
1

Ėq′ = ωs s Ed′ − ′ Eq′ + (x − x′ )Id (5.6)
T0
1
Ėd′ = − ωs s Eq′ − ′ (Ed′ − (x − x′ )Iq )
T0
where,
ωs − ωmotor
s = is the slip of the induction motor
ωs
Tm = A + Bs + Cs2 is the mechanical load torque in terms of the slip
Te = Eq′ Iq + Ed′ Id is the electrical torque
xr + xm
T0′ = is the transient open-circuit time constant
ω s rr
xm xr
x′ = xs + and x = xs + xm
xm + xr

Note that Iq and Id in (5.6) are given by (5.5). Furthermore, ŪL = Ūs , and if κ = 0 we
have I¯L = I¯s , i.e. S̄L = S̄dyn .
There are also other dynamic load models which are summarized in [10].
 77

5.2 Multi-machine power systems

Figure 5.3 shows a multi–machine power system. The transmission network has a total
of N buses including the generators terminal buses (the first ng buses). Voltages at
these buses are given by Ūi = Ui ∠θi .

U1 U ng +1

GEN 1
S L1
SL n
U2 U ng + 2 g +1

GEN 2
SL2
SL n
g +2

Transmission
Network

U ng UN
GEN ng
SL n
g SL N

Figure 5.3. A multi–machine power system.

For this system the following assumptions are made:

1. Dynamic of each generator is described by the one-axis model.

2. Mechanical power of each generator (i.e. Pm ) is constant.

3. Inherent damping of each generator (i.e. D) is zero.

4. Loads are represented by static load model.

5. Transmission network is lossless.

Based on the assumption 1, the equivalent circuit of the k:th generator is shown in
Figure 5.4 where the reactance of the k:th transformer is included in x′dk . The voltage
′
at the generator internal bus is given by Eqk ∠δk .

′ ∠δk
Eqk U k ∠θk
′
jxdk I gk

Figure 5.4. Synchronous generator one-axis dynamic circuit.

5.2.1 Structure preserving model

Based on the assumptions 1-3, the dynamic of the k:th generator is given by
78

for k = 1 · · · ng
δ̇k = ωk
 ′ 
1 Eqk Uk
ω̇k = Pmk − sin(δk − θk )
Mk x′dk (5.7)
 ′

1 xdk ′ xdk − xdk
Ė ′ qk = ′ Ef k − ′ Eqk + Uk cos(δk − θk )
Tdok xdk x′dk
where the reactance of the k:th transformer is also included in xdk .
Next, let Ybus of order (N × N) be the admittance matrix of the transmission network,
and the kl-th element of the admittance matrix be defined by Ȳbuskl = Gkl +jBkl , where
Gkl represents solely the resistances of the respective transmission lines. However, based
on the assumption 5 (since R ≪ X) Gkl = 0, and therefore Ȳbuskl = jBkl .
The real and reactive powers injected into bus k are given by
for k = 1 · · · ng
N ′
X Eqk Uk sin(θk − δk )
Pk = Bkl Uk Ul sin(θk − θl ) +
l=1
x′dk
N
(5.8)
X Uk2 − Eqk
′
Uk cos(θk − δk )
Qk = − Bkl Uk Ul cos(θk − θl ) +
l=1
x′dk

and for k = (ng + 1) · · · N

N
X
Pk = Bkl Uk Ul sin(θk − θl )
l=1
N
(5.9)
X
Qk = − Bkl Uk Ul cos(θk − θl )
l=1

Let PLk and QLk be the active and reactive loads at bus k. Then, for k = 1 · · · N the
power flow equations (5.8) and (5.9) can be written as
P k + P Lk = 0
(5.10)
Qk + QLk = 0
which is a set of algebraic equations.
Let
′
x = [δ1 · · · δng , ω1 · · · ωng , Eq1 ′
· · · Eqn g
]T
(5.11)
y = [θ1 · · · θN , U1 · · · UN ]T

Then, equations (5.7) and (5.10) can be rewritten as

ẋ = f (x, y)
(5.12)
0 = g(x, y)
 79

Equation (5.12) is a set of differential-algebraic equations which describe the dynamic of

the multi-machine system. The differential equations ẋ = f (x, y) describe the dynamic
of the generators. In case of having dynamic loads and other dynamical components,
their contributions will be included in these differential equations. The algebraic equa-
tions consist of the network equations based on Kirchhoff’s current law, i.e. the sum of
all currents (or powers) flowing into bus k must be equal zero. Equation (5.8) (which
is included in (5.10), and gives the power balance equations at the generator terminal
buses) is the interface between the algebraic variables (i.e. y) and the state variables
(i.e. x).
This model is termed Structure Preserving Model (SPM) since the structure of the
system model is preserved. An advantage of using SPM is that from a modeling view-
point, it allows more realistic representations of power system components, especially
load behaviors.

5.2.2 Reduced Network Model

The Reduced Network Model (RNM) can be obtained by replacing assumption 4 with
the following assumption

Uk2
4. Loads are represented as constant impedances Z̄Lk = RLk + jXLk = .
P Lk − jQLk

Impedance loads can also be given in the form of admittance loads as

1 PL − jQL
ȳLk = = k 2 k (5.13)
Z̄Lk Uk
Let Y of order (N × N) be the admittance matrix of the transmission network in
which the admittance loads are included, i.e. Y = Ybus + YL , where YL is a diagonal
matrix of order N × N whose diagonal entries are ȳLk . Then, the kl-th element of
the admittance matrix is given by Ȳkl = Gkl + jBkl where Gkl = 0 for k 6= l due to
assumption 5. However, Gkk 6= 0 because of the admittance loads.
Next, the multi-machine system is augmented with ng buses which represent the gen-
erators internal buses as shown in Figure 5.5, i.e. the total number of buses is ng + N.

The current of the k-th generator (i.e. I¯gk ) is given by

Ē ′ qk − Ūk
I¯gk = for k = 1 · · · ng (5.14)
jx′dk
Based on Kirchhoff’s current law, the following can be obtained.
N
X
0 = Ȳkl Ūl − I¯gk for k = 1 · · · ng (5.15)
l=1
N
X
0 = Ȳkl Ūl for k = ng + 1 · · · N (5.16)
l=1
80

Eq¢1Ðd1 I g1 U1 U ng +1
jx¢d 1

y L1 yL n
g +1

Transmission network
¢ g Ðd ng
Eqn U ng UN
jx¢dng I gng

yL n yL N
g

Figure 5.5. The augmented multi–machine power system .

or in compact form     
IG YA YB EG
= (5.17)
0N YC YD UN
where

• IG = [I¯g1 · · · I¯gng ]T , and 0N is a zero vector of order N × 1.

• EG = [Ē ′ q1 · · · Ē ′ qng ]T , and UN = [Ū1 · · · ŪN ]T .

1
• YA is a diagonal matrix of order ng × ng whose diagonal entries are ȲAkk = .
jx′dk
−1
• YB is a zero matrix of order ng × N whose non-zero entries are ȲBkk = .
jx′dk

• YC = (YB )T

• YD is a matrix of order N × N which is given by

 
Y A 01
YD = Y +
02 03

where, Y is the admittance matrix of the transmission network (including the

admittance loads), and 01 , 02 and 03 are zero matrices with appropriate orders.

Equation (5.17) can be rewritten as

IG = YA EG + YB UN (5.18)
0N = YC EG + YD UN (5.19)
 81

By solving UN from equation (5.19), and substituting it in equation (5.18), the following
is obtained
UN = − YD−1 YC EG


IG = YA EG + YB UN = YA − YB YD−1 YC EG (5.20)
= YRN M EG = (G + jB)EG
Thus, for k = 1 · · · ng
ng
X
I¯gk = (Gkk + jBkk )Ē ′ qk + (Gkl + jBkl )Ē ′ ql (5.21)
l=1
l6=k

where,

• YRN M = G + jB is the admittance matrix of the reduced network model which

is of the order ng × ng .
• Gkk and Bkk represent the equivalent short-circuit conductance and susceptance
of the k–th generator.
• Gkl and Bkl represent the transfer conductance and susceptance between internal
buses k and l.

Note that Gkk and Gkl are non-zero in case of having any active load in the system,
i.e. PL 6= 0.
By virtue of equation (3.48), the generator current I¯gk can also be written as
ng
X
I¯gk = (Iqk + j Idk ) ejδk
= (Gkk + jBkk )E qk e ′ jδk
+ (Gkl + jBkl )E ′ ql ejδl (5.22)
l=1
l6=k

from which Iqk and Idk can be solved for k = 1 · · · ng , as

ng
X
′
Iqk = Gkk Eqk + E ′ ql (Gkl cos(δk − δl ) + Bkl sin(δk − δl ))
l=1
l6=k
ng (5.23)
X
′
Idk = Bkk Eqk + E ′ ql (Bkl cos(δk − δl ) − Gkl sin(δk − δl ))
l=1
l6=k

Based on equations (3.77) and (3.96), the dynamic of the k:th generator is given by

δ̇k = ωk
1 1 ′


ω̇k = (Pmk − Pek ) = Pmk − Eqk Iqk
Mk Mk (5.24)
1 ′


Ė ′ qk = ′ Ef k − Eqk + (xdk − x′dk )Idk
Tdok
82

Substituting (5.23) into (5.24), the following is obtained

δ̇k = ωk
 
ng
1  ′2
X 
ω̇k = Pmk − Gkk Eqk − E ′ qk E ′ ql (Gkl cos(δkl ) + Bkl sin(δkl )) (5.25)
Mk l=1
l6=k
  
ng
1  ′ ′  ′
X 
Ė ′ qk = ′  Ef k − Eqk + (xdk − xdk ) Bkk Eqk + E ′ ql (Bkl cos(δkl ) − Gkl sin(δkl ))
Tdok l=1
l6=k

where, δkl = δk − δl .
It is obvious that equation (5.25) contains only state variables and constants. The
reason is that with the loads represented by constant impedances all N network buses
are eliminated by equation (5.20), and there are no algebraic variables (i.e. the voltages
at the network buses) in (5.25). Therefore, there is no need of (5.10) to calculate the
algebraic variables y.
Using x in (5.11), equation (5.25) can be written as

ẋ = f (x) (5.26)

i.e., the dynamic of the multi-machine system shown in Figure 5.5 is described by only
a set of differential equations. Furthermore, the (ng + N)-bus system in Figure 5.5 is
reduced to an ng -bus system containing only the generators internal buses.

Example 5.1 Consider again the lossless system shown in Figure 5.6. Based on as-
sumptions 1- 5, describe mathematically the dynamic of the system, if

a) the load model given by (5.2) is used, with mp = 1 and mq = 2. Also, plot the
variations of the electric powers Pg when 10% of the active load at BUS 4 is
disconnected during 0.1 (s).

b) the loads are considered as constant impedances. Also, plot the variations of the
electric powers Pg for a three-phase fault at BUS 4. The fault is cleared after 100
(ms), i.e. tc = 0.1 (s).

(See the MATLAB files for ”Project D2” and ”Project D4”)
a) Let
1 1 1
bd1 = , bd2 = , bd3 =
x′d1 x′d2 x′d3
1 1 1
b14 = , b24 = , b34 =
x14 x24 x34

where, the transformer reactance (xT ) is included in x′d .

 83

Gen3

T3

BUS 3
S L3
BUS 1 BUS 4 BUS 2

Gen1 Gen2
T1 S L1 SL4 SL2 T2

Figure 5.6. A simple multi–machine power system.

The loads are modeled as

   2
U1 U1
S̄L1 = PL1 + jQL1 = PL10 + jQL10
U10 U10
   2
U2 U2
S̄L2 = PL2 + jQL2 = PL20 + jQL20
U20 U20
   2
U3 U3
S̄L3 = PL3 + jQL3 = PL30 + jQL30
U30 U30
   2
U4 U4
S̄L4 = PL4 + jQL4 = PL40 + jQL40
U40 U40

The dynamic of the system is described by a set of differential- algebraic equations.

The differential equations are given by (xT is included in x′d and xd )

δ̇1 = ω1
δ̇2 = ω2
δ̇3 = ω3
1
ω̇1 = (Pm1 − Pe1 )
M1
1
ω̇2 = (Pm2 − Pe2 )
M2
1
ω̇3 = (Pm3 − Pe3 )
M3
 
1 xd1 ′ xd1 − x′d1
Ė ′ q1 = ′
Ef 1 − ′ Eq1 + U1 cos(δ1 − θ1 )
Tdo1 xd1 x′d1
 
1 xd2 ′ xd2 − x′d2
Ė ′ q2 = ′
Ef 2 − ′ Eq2 + U2 cos(δ2 − θ2 )
Tdo2 xd2 x′d2
 
1 xd3 ′ xd3 − x′d3
Ė ′ q3 = ′
Ef 3 − ′ Eq3 + U3 cos(δ3 − θ3 )
Tdo3 xd3 x′d3
84

The algebraic equations are given by

0 = −Pg1 + b14 U1 U4 sin(θ1 − θ4 ) + PL1

0 = −Pg2 + b24 U2 U4 sin(θ2 − θ4 ) + PL2
0 = −Pg3 + b34 U3 U4 sin(θ3 − θ4 ) + PL4
0 = b14 U4 U1 sin(θ4 − θ1 ) + b24 U4 U2 sin(θ4 − θ2 ) + b34 U4 U3 sin(θ4 − θ3 ) + PL4

0 = −Qg1 + b14 (U12 − U1 U4 cos(θ1 − θ4 )) + QL1

0 = −Qg2 + b24 (U22 − U2 U4 cos(θ2 − θ4 )) + QL2
0 = −Qg3 + b34 (U32 − U3 U4 cos(θ3 − θ4 )) + QL3
0 = b14 (U42 − U4 U1 cos(θ4 − θ1 )) + b24 (U42 − U4 U2 cos(θ4 − θ2 ))
+b34 (U42 − U4 U3 cos(θ4 − θ3 )) + QL4

where,
′
Pek = bdk Eqk Uk sin(δk − θk )
′
Pgk = −bdk Eqk Uk sin(θk − δk )


Qgk = −bdk Uk2 − Eqk
′
Uk cos(θk − δk )

By defining x and y as follows

′ ′ ′ T
x = [δ1 δ2 δ3 ω1 ω2 ω3 Eq1 Eq2 Eq3 ]
y = [θ1 θ2 θ3 θ4 U1 U2 U3 U4 ]T

the dynamic of the system is described in compact form by

ẋ = f (x, y)
0 = g(x, y)

with initial conditions x0 and y0 which are computed based on the load flow calcula-
tions, and by setting ẋ = f (x0 , y0) = 0.
In the load flow calculations, the loads are normally considered as constant loads (i.e.
S̄L0 = PL0 + jQL0 ). The initial conditions are then computed by the following steps:

1. Run the load flow calculations.

2. From load flow, y0 = [θ10 θ20 θ30 θ40 U10 U20 U30 U40 ]T is known.

3. Calculate Pg10 , Qg10 , Qg20 and Qg30 .

4. Calculate I¯g10 , I¯g20 and I¯g30

Pg10 − jQg10 Pg20 − jQg20 Pg30 − jQg30

I¯g10 = , I¯g20 = , I¯g30 =
Ū1∗0 Ū2∗0 Ū3∗0
 85

5. Calculate Ē ′ q10 , Ē ′ q20 and Ē ′ q30

′
Ē ′ q10 = Eq10
∠δ10 = Ū10 + j x′d1 I¯g10
′
Ē ′ q20 = Eq20
∠δ20 = Ū20 + j x′d2 I¯g20
′
Ē ′ q30 = Eq30
∠δ30 = Ū30 + j x′d3 I¯g30
′
6. x0 = [δ10 δ20 δ30 0 0 0 Eq10
′
Eq20
′
Eq30
]T is now known.

7. Finally, Pm1 , Pm2 , Pm3 , Ef 1 , Ef 2 and Ef 3 can be computed by setting

ẋ = f (x0 , y0 ) = 0, i.e.
Pm1 = Pe10 = Pg10
Pm2 = Pe20 = Pg20
Pm3 = Pe30 = Pg30
xd1 ′ xd1 − x′d1
Ef 1 = ′ Eq1 − U10 cos(δ10 − θ10 )
xd1 0
x′d1
xd2 ′ xd2 − x′d2
Ef 2 = ′ Eq2 − U20 cos(δ20 − θ20 )
xd2 0
x′d2
xd3 ′ xd3 − x′d3
Ef 3 = ′ Eq3 − U30 cos(δ30 − θ30 )
xd3 0
x′d3

The dynamics of the system in the pre-disturbance state, during-disturbance state and
post-disturbance state are given by
Pre-disturbance state:
ẋ = f pre (x0 , y0 ) = 0
0 = g pre (x0 , y0)
 
U4
During-disturbance state with PL4 = 0.9 PL40 U40
:

ẋ = f f (x, y)
0 = g f (x, y)
 
U4
Post-disturbance state with PL4 = PL40 U40
:

ẋ = f post (x, y)
0 = g post(x, y)

Figure 5.7 shows the variations of the electric power of each generator for this distur-
bance. As shown in the figure, Gen 1 oscillates against Gen 2 and Gen 3. Although,
D1 = D2 = D3 = 0, there is indeed a poor damping in the system (why?).
86

0.021

∆ Pe
0

−0.022
0.5 1 2 3 4 5
Time (s)

Figure 5.7. Variations of ∆Pe1 (blue), ∆Pe2 (red) and ∆Pe3 (black).

b) In a similar way as described in task a), the initial conditions are firstly obtained.
Next, the loads are modeled as admittances as follows
1 PL10 − jQL10 1 PL20 − jQL20
ȳL1 = = , ȳL2 = =
Z̄L1 U120 Z̄L2 U220
1 PL30 − jQL30 1 PL40 − jQL40
ȳL3 = = , ȳL4 = =
Z̄L3 U320 Z̄L4 U420

to make the following YL matrix.

 
ȳL1 0 0 0
 0 ȳL2 0 0 
YL = 
 0

0 ȳL3 0 
0 0 0 ȳL4

Let
1 1 1
b̄d1 = , b̄d2 = , b̄d3 =
j x′d1 j x′d2 j x′d3
and
 
    −b̄d1 0 0
b̄d1 0 0 −b̄d1 0 0 0  0 −b̄d2 0 
YA = 0 b̄d2 0 , YB =
   0 −b̄d2 0 0 , Y C = 
 0

0 −b̄d3 
0 0 b̄d3 0 0 −b̄d3 0
0 0 0

Next, with Y = Ybus + YL , and

   
  ȳ11 + ȳL1 0 0 ȳ14 b̄d1 0 0 0
Y 0  0 ȳ22 + ȳL2 0 ȳ24   0 b̄d2 0
  0
YD = Y + A 1 =  + 
02 03  0 0 ȳ33 + ȳL3 ȳ34   0 0 b̄d3 0
ȳ14 ȳ24 ȳ34 ȳ44 + ȳL4 0 0 0 0
 87

the admittance matrix of the Reduced Network Model (RNM) is obtained by

 
G11 + j B11 G12 + j B12 G13 + j B13
YRN M = YA − YB YD−1 YC = G + j B = G21 + j B21 G22 + j B22 G23 + j B23 
G31 + j B31 G32 + j B32 G33 + j B33
 
0.0843 − j 1.1164 0.1476 + j 0.5463 0.1181 + j 0.4370
= 0.1476 + j 0.5463 0.2583 − j 1.5441 0.2066 + j 0.7648
0.1181 + j 0.4370 0.2066 + j 0.7648 0.1653 − j 1.3882

Based on equation (5.21) and the structure of the YRN M , the figure below shows the
reduced model of the system shown in Figure 5.6. In this model, all generator buses
are replaced by the corresponding generator internal buses, and all load buses (BUS 4
in this example) are removed. (What are ȳiiRN M and ȳijRN M in the figure?)

Eq¢1Ð d1 Eq¢ 2 Ð d 2
I g1 y12RNM Ig2

y11RNM y22RNM
RNM RNM
y13 y 23

Eq¢ 3Ð d 3
y33RNM Ig3

Figure 5.8. The reduced network model of the system shown in Figure 5.6.

For a fault at BUS 4, YRN M must be modified. This is done by adding a large ad-
mittance to the diagonal element of YD corresponding to the faulty bus (BUS 4) as
follows

YDf (4, 4) = YD (4, 4) + (1 − j)1012 ⇒

f f −1 f f
YRN M = YA − YB (YD ) YC = G + j B

 
0.0000 − j 1.4286 0.0000 + j 0.0000 0.0000 + j 0.0000
= 0.0000 + j 0.0000 0.0000 − j 2.5000 0.0000 + j 0.0000
0.0000 + j 0.0000 0.0000 + j 0.0000 0.0000 − j 2.0000
Next, Iqk and Idk with k = 1, 2, 3 are obtained by
3
X
′
Iqk = Gkk Eqk + E ′ ql (Gkl cos(δk − δl ) + Bkl sin(δk − δl ))
l=1
l6=k

3
X
′
Idk = Bkk Eqk + E ′ ql (Bkl cos(δk − δl ) − Gkl sin(δk − δl ))
l=1
l6=k
88

Thus, the dynamic of the system is given by

δ̇1 = ω1
δ̇2 = ω2
δ̇3 = ω3
1 ′


ω̇1 = Pm1 − Eq1 Iq1
M1
1 ′


ω̇2 = Pm2 − Eq2 Iq2
M2
1 ′


ω̇3 = Pm3 − Eq3 Iq3
M3
1 ′


Ė ′ q1 = ′ Ef 1 − Eq1 + (xd1 − x′d1 )Id1
Tdo1
2 ′


Ė ′ q2 = ′ Ef 2 − Eq2 + (xd2 − x′d2 )Id2
Tdo2
3 ′


Ė ′ q3 = ′ Ef 3 − Eq3 + (xd3 − x′d3 )Id3
Tdo3
f f f
Note that in the during-fault state, YRN M = G + j B is replaced by YRN M = G +j B .

Figure 5.9 shows the variations of the electric power of each generator for this fault.
Obviously, the power variations are greater for this fault compared to the disturbance
in the task a) above.

0.19

0
∆ Pe

−0.28
0.5 1 2 3 4 5
Time (s)

Figure 5.9. Variations of ∆Pe1 (blue), ∆Pe2 (red) and ∆Pe3 (black).

Example 5.2 Consider again the lossless system shown in Figure 5.6. Let the two-axis
model given by (3.84) be used for the generators. Describe mathematically the dynamic
of the system, if the load model given by (5.2) is used, with mp = 1 and mq = 2. Also,
find the initial values x0 and y0 .
 89

Since two-axis model is used, the following three differential equations are added into
ẋ = f (x, y), (xT is included in x′q and xq )
 
1 ′

xq1 ′
1 xq1 − x′q1
Ė ′ d1 = ′
−Ed1 − (xq1 − x′q1 )Iq1
= ′ − ′ Ed1 − U1 sin(δ1 − θ1 )
Tqo1 Tqo1 xq1 x′q1
 
1 ′ ′

1 xq2 ′ xq2 − x′q2
Ė ′ d2 = ′
−Ed2 − (xq2 − xq2 )Iq2 = ′ − ′ Ed2 − U2 sin(δ2 − θ2 )
Tqo2 Tqo2 xq2 x′q2
 
1 ′ ′

1 xq3 ′ xq3 − x′q3
Ė ′ d3 = ′
−Ed3 − (xq3 − xq3 )Iq2 = ′ − ′ Ed3 − U3 sin(δ3 − θ3 )
Tqo3 Tqo3 xq3 x′q3

and the electric power is given by

′ ′


Pek = bdk Eqk Uk sin(δk − θk ) + Edk Uk cos(δk − θk )

The algebraic equations are unchanged, however

Pgk = Pek


Qgk = −bdk Uk2 − Eqk
′ ′
Uk cos(θk − δk ) − Edk Uk sin(θk − δk )

Note that in the one-axis model, the voltage behind the transient reactance is given by
′
Eqk ejδk (see Figure 5.4). However in the two-axis model, the generator internal voltage
′
is Ē ′ = (Eqk ′
+ jEdk ) ejδk (see Figure 3.11 (a) and equation (3.99)). Furthermore, from
Figure 3.6 and equations (3.48) and (3.83) we have

Udk = Uk sin(θk − δk )
Uqk = Uk cos(θk − δk )
Ūk = Uk ejθk = (Uqk + j Udk ) ejδk
I¯k = I¯gk = (Iqk + j Idk ) ejδk (5.27)
′
Udk − Edk
Iqk = −
x′qk
′
Uqk − Eqk
Idk =
x′dk

The initial conditions are computed as follows:

• Points ”1.” to ”4.” in the task a) above.

• Prior to the disturbance

Ė ′ dk = 0 ⇒ E ′ dk0 = −(xqk − x′qk ) Iqk0 = −(xqk − x′dk ) Iqk0 (5.28)

since it is assumed x′dk = x′qk .

90

From equation (3.99), we have

′
(Eqk 0
′
+ jEdk 0
) ejδk0 = Ūk0 + jx′dk I¯gk0 = Ūk0 + jx′dk (Iqk0 + j Idk0 ) ejδk0 (5.29)

Substituting E ′ dk0 from equation (5.28) into equation (5.29), the following is
obtained
′
Eqk 0
ejδk0 + x′dk Idk0 ejδk0 = Ūk0 + jxqk Iqk0 ejδk0 (5.30)

Subtracting xqk Idk0 ejδk0 from the both sides of (5.30) results in
′
(Eqk 0
−(xqk −x′dk ) Idk0 ) ejδk0 = Ūk0 +jxqk (Iqk0 +jIdk0 ) ejδk0 = Ūk0 +jxqk I¯gk0 (5.31)

Thus, δk0 can be calculated from equation (5.31). When δk0 is known, Uq0 , Ud0 ,
′ ′
Iq0 , Id0 , and finally Edk 0
and Eqk 0
can then be calculated from (5.27).

• Point ”7.”.
Chapter 6
Rotor angle stability
As was presented in Chapter 1, rotor angle stability refers to the ability of synchronous
machines of an interconnected power system to remain in synchronism after being
subjected to a disturbance. Instability that may result occurs in the form of increasing
angular swings of some generators leading to their loss of synchronism with other
generators. Loss of synchronism can occur between one machine and the rest of the
system, or between groups of machines, with synchronism maintained within each
group after separating from each other. This stability is characterized as:

• Transient stability which is concerned with the ability of the power system to
maintain synchronism when subjected to a large disturbance, such as a short–
circuit on a transmission line. Transient stability depends on the initial operating
conditions of the system as well as the type, severity and location of the distur-
bance.

• Small-signal stability which is concerned with the ability of the power system
to maintain synchronism under small disturbances. The disturbances are consid-
ered to be sufficiently small that linearization of system equations is permissible
for purposes of analysis.

Transient stability has already been discussed for an SMIB system. Due to the sim-
plicity of the SMIB system, it was possible to study its transient stability by means
of Equal Area Criterion without running time-domain simulation. However, since this
method is not analytically applicable to multi-machine power systems, some techniques
applicable to transient stability analysis of multi-machine power systems are presented
in this chapter.

6.1 Transient stability

The main aim of transient stability analysis of a power system is to study whether
the system after a large disturbance will settle to an acceptable steady-state as time
passes. To ensure power system stability (or security), power system utilities would like
to assess the performance of either current or postulated power system configuration
under a variety of actual or hypothesized operating conditions and disturbances. Then
based on the results of the stability studies, they take preventive control action if
necessary.
For transient stability study, power system utilities broadly apply time-domain simu-
lation programs to predict the response of the system(s) to various large disturbances.
In these programs, the dynamic of the system may be described by a set of differential-
algebraic equations of the form (5.12) (or only differential equations) which are solved

91
92

by using step-by-step numerical integration methods. Based on simulation results, the

behavior of the system is evaluated to predict the system stability and operating limits.
An advantage of applying time-domain simulations is that it is possible to have more
detailed models for generators and other power system components. By these detailed
models, the dynamic behaviors of the actual power system components are then more
accurately represented.

Stable case Unstable case

28 80
∆δ (deg.)

∆δ (deg.)
0 0

−21 −60
1 10 1 2
Time (s) Time (s)

Figure 6.1. Rotor angle deviations of each generator.

Figure 6.1 shows the dynamic behaviors of the generators in the Nordic32 test system
proposed by CIGRE after a large disturbance for different clearing times. For the
unstable case, the system loses its synchronism and it is transiently unstable for the
specified disturbance and clearing time. Simulations were performed by using the
simulation program SIMPOW, and the results were plotted in MATLAB.
The time-domain approach has however disadvantages such as heavy and time-consuming
computations (especially for large interconnected power systems). Moreover (as shown
in the above figure), it does not provide any information regarding the stability margin.
To overcome these disadvantages, other transient stability analysis methods have been
developed. In this compendium two methods will be presented, namely the Transient
Energy Function (TEF) method and the SIngle Machine Equivalent (SIME) method.

6.1.1 Transient Energy Function (TEF) method

As explained in Section 4.3, TEF method has received considerable attention for assess-
ment of power system transient stability. The most challenge of the application of TEF
method to a multi-machine power system is calculation of Vcr to have a less conserva-
tive estimation. There are different techniques for calculating Vcr . These techniques
have been presented and discussed in [7] and related references therein.
In the development of energy functions for multi-machine power systems, and also
in order to clearly distinguish between the forces that accelerate the whole system
 93

and those that tend to separate the system into different parts, it is convenient to
transform the system (5.12) into the so called Center Of Inertia (COI) reference frame.
The position of the COI is defined by
ng ng ng
1 X 1 X X
δCOI = Mk δk , ωCOI = Mk ωk where MT = Mk (6.1)
MT k=1 MT k=1 k=1

Furthermore,
ng ng
1 X 1 X PCOI
ω̇COI = Mk ω̇k = (Pmk − Pek ) = (6.2)
MT k=1 MT k=1 MT

Next, the state variables δk and ωk are transformed to the COI variables as

δ̃k = δk − δCOI , ω̃k = ωk − ωCOI (6.3)

These COI variables are constrained by

ng ng
X X
Mk δ̃k = 0 , Mk ω̃k = 0 (6.4)
k=1 k=1

The load buses angles are also transformed to the COI reference frame by

θ̃k = θk − δCOI

Time derivation of equation (6.3) gives

˙
δ̃k = δ̇k − δ̇COI = ωk − ωCOI = ω̃k
 
1 Mk
ω̃˙ k = ω̇k − ω̇COI = Pmk − Pek − PCOI
Mk MT

The system (5.12) in the COI reference frame is then expressed as (for k = 1 · · · ng )

δ̃˙k = ω̃k
 ′ 
1 Eqk Uk M k
ω̃˙ k = Pmk − sin(δ̃k − θ̃k ) − PCOI (6.5)
Mk x′dk MT
 
1 xdk ′ xdk − x′dk
Ė qk = ′
′ Ef k − ′ Eqk + Uk cos(δ̃k − θ̃k )
Tdok xdk x′dk

and
P k + P Lk = 0
(6.6)
Qk + QLk = 0

where, Pk and Qk are given by (5.8) and (5.9) in which θ and δ are replaced by θ̃ and
δ̃. Note however that θ̃k − θ̃l = θk − θl and δ̃k − θ̃k = δk − θk .
94

Let the loads (i.e. PLk and QLk ) be modeled based on (5.2) with mp = 0 and an
arbitrary mq. Then, the following energy function is given for the DAE system (6.5)
and (6.6)
V(ω̃, δ̃, Eq′ , U, θ̃) = WK + WP + Co (6.7)
where
ng 7
1X X
WK = Mk ω̃k2 and WP = V2p
2 k=1 p=1

with
ng N N Z
X X X QLk
V21 = − Pmk δ̃k , V22 = PLk θ̃k , V23 = d Uk
k=1 k=1 k=1
Uk
ng
X 1 2
V24 = ′
[E ′ qk + Uk2 − 2Eqk
′
Uk cos(δ̃k − θ̃k )]
k=1
2xdk
N N
1X X
V25 = − Bkl Uk Ul cos(θ̃k − θ̃l )
2 k=1 l=1
ng ng
X ′
Ef dk Eqk X E ′ 2qk
V26 = − and V27 =
k=1
xdk − x′dk k=1
2(xdk − x′dk )

Using the notation

dV ∂V dω̃
]ω̃ for [
dt ∂ ω̃ dt
and similarly for the other states, the following are then obtained
dV1 dV21 dV24
[ ]ω̃ + [ + ] =0 (6.8)
dt dt dt δ̃

dV22 dV24 dV25 X

[ + + ]θ̃ = (Pk + PLk )θ̃˙k = 0 (6.9)
dt dt dt

dV23 dV24 dV25 X U̇k

[ + + ]U = (Qk + QLk ) =0 (6.10)
dt dt dt Uk

ng ′
dV24 dV26 dV27 X Tdok
[ + + ]Eq′ = − ′
(Ė ′ qk )2 (6.11)
dt dt dt k=1
xdk − xdk

which results in the following time derivative of the energy function

ng ′
dV X Tdok
=− ′
(Ė ′ qk )2 ≤ 0 (6.12)
dt k=1
xdk − xdk
 95

6.1.2 SIngle Machine Equivalent (SIME) method

The SIME method is a hybrid direct-temporal transient stability method, which trans-
forms the trajectories of a multi-machine power system into the trajectory of a single
machine equivalent system of the form
δ̇SIM E = ωSIM E
(6.13)
ω̇SIM E = M −1 [PmSIM E − PeSIM E ]
whose parameters (which are derived from multi-machine power system) are time-
varying [12].
Basically, the SIME method deals with the post-fault configuration of a power system
subjected to a disturbance which presumably drives it to instability. Under such con-
dition, the SIME method uses a time-domain program in order to identify the mode
of separation of its machines into two groups, namely critical (subscript C) and non-
critical machines (subscript NC) which are replaced by successively a two-machine
equivalent. Then, this two-machine equivalent is replaced by a single machine equiv-
alent system. By definition, the critical machines are the machines responsible of the
loss of synchronism.
The parameters of (6.13) are given by
δSIM E = δC − δN C
ωSIM E = ωC − ωN C
!
X X
PmSIM E = MT−1 MC Pmi − MN C Pmj (6.14)
i∈C j∈N C
!
X X
PeSIM E = MT−1 MC Pei − MN C Pej
i∈C j∈N C
MC MN C
M = and MT = MC + MN C
MT
where
X X
MC = Mi , MN C = Mj
i∈C j∈N C
X X
δC = MC−1 Mi δi , δN C = MN−1C Mj δj (6.15)
i∈C j∈N C
X X
ωC = MC−1 Mi ωi , ωN C = MN−1C Mj ωj
i∈C j∈N C

By refreshing the parameters of the single machine equivalent system at each inte-
gration time–step and numerically assessing the transient stability of this equivalent
system based on the equal area criterion, the SIME method provides accurate and
fast transient stability assessment of multi-machine power systems, and also additional
interesting information such as stability margins, identification of the mode of insta-
bility and corresponding critical machines, sensitivity analysis and control techniques
[12]-[13].
96

6.2 Small-signal stability

As was mentioned, small-signal stability analysis deals with small disturbances, and
it is applied to linearized system models. This analysis provides valuable information
about the inherent dynamic characteristics of the power system. It also provides a
valuable complement to information gained by time-domain simulation.
Small disturbances, such as the normal small fluctuations in the system loads or small
changes of the set values of some parameters, are always present in a power system and
normally the resulting power (or electromechanical) oscillations are stable that is the
oscillations are positively damped and decay with time. However, these spontaneous
oscillations may due to insufficient damping occasionally grow in amplitude with time,
and result in sustained low frequency oscillations that cause loss of synchronism.
The change in electrical torque (∆Te ) of a synchronous generator following a distur-
bance can be resolved into two components, namely the synchronizing torque compo-
nent (∆TS ) and the damping torque component (∆TD ) as follows [6]:

∆Te = ∆TS + ∆TD = KS ∆δ + KD ∆ω (6.16)

where,

• ∆TS is in phase with ∆δ, and KS is the synchronizing torque coefficient.

• ∆TD is in phase with ∆ω, and KD is the damping torque coefficient.

Lack of sufficient synchronizing torque results in non-oscillatory instability in the first

few seconds following a fault, i.e. loss of synchronism between interconnected gener-
ators. This type of instability is essentially caused by the non-linear nature of the
dynamics of interconnected generators [14]. Therefore, (fast) automatic voltage reg-
ulators have been used to increase the synchronizing torques between interconnected
generators. However, they have also an effect of reducing the damping torques which
may result in oscillatory instability. This phenomenon is a typical small-signal stability
problem in today’s power systems. Thus to ensure system stability, a power system
should be designed and planned such that both synchronizing and damping torques
(with sufficient positive KS and KD ) exist for each of the synchronous machines.
In this section, application of small-signal stability analysis to power systems and the
impact of AVR and PSS on the synchronizing and damping torques will be presented.
But first, the essential characteristics (also known as modal analysis) of a Linear Time-
Invariant (LTI) system of the form

ẋ(t) = A x(t) + B U(t)

(6.17)
Y(t) = C x(t)

are reviewed in terms of the eigen-properties of matrix A.

 97

6.2.1 Modal analysis

Consider the LTI system (6.17), where

• x is a vector of order nx × 1 containing the system state variables.

• Y is a vector of order m × 1 containing the system outputs or measured variables.
• U is a vector of order r × 1 containing the system inputs (or control) variables.
• A is a matrix termed as the state matrix of order nx × nx .
• B is a matrix termed as the input matrix of order nx × r.
• C is a matrix termed as the output matrix of order m × nx .

For now, let U be zero and consider the unforced LTI system
ẋ(t) = A x(t) (6.18)

Eigenvalues

The eigenvalues of A are defined as the nx solutions of λ = λ1 · · · λnx which satisfy

det( A − λ 1 ) = 0 (6.19)
where “det” stands for determinant, and 1 is the identity matrix. The eigenvalues
may be real or complex. It is common to associate each eigenvalue λi with a mode
of the system. The stability of an equilibrium point can be determined based on the
eigenvalues of the system by applying Theorem 2.1. A real eigenvalue corresponds to a
non-oscillatory mode. A negative real eigenvalue represents a decaying mode whereas
a positive real eigenvalue monotonic instability.
If A is real, complex eigenvalues always occur in conjugate pairs. Each pair corresponds
to an oscillatory mode, and is expressed (for the i-th mode) by
λi = σi ± jωpi (6.20)
The real component σi gives the damping of the i-th mode. A negative σ represents
a damped oscillatory mode, however a positive σ represents an oscillatory instability.
The imaginary component ωpi gives the oscillation frequency of the i-th mode, and is
expressed by
ωp
fpi = i (6.21)
2π
The damping ratio of the i-th mode is given by
−σi −σi
ζi = q = (6.22)
σ2 + ω2 |λi |
i pi

A positive damping ratio determines the decay rate of the oscillation amplitude.
In this compendium we assume that the eigenvalues are distinct, i.e. λi 6= λj .
98

Eigenvectors

Any non-zero vector Vir which satisfies

A Vir = λi Vir (6.23)

is termed the right eigenvector of A corresponding to the eigenvalue λi . Note that Vir
is a column vector of order nx × 1.
Similarly, any non-zero vector Vil which satisfies

Vil A = λi Vil (6.24)

is termed the left eigenvector of A corresponding to the eigenvalue λi . Note that Vil is
a row vector of order 1 × nx .
It can be shown that
Vil Vjr = 0
(6.25)
Vil Vir = Ci 6= 0

Note that Vir (or Vil ) is not a unique solution, k Vir (where k is a scalar) can also be a
solution. Due to this property, it is possible to normalize the right and left eigenvectors
so that  
r
v1i
r 
l r
 l l l
 v2i 
Vi Vi = vi1 vi2 · · · vinx  ..  = 1 (6.26)
 . 
vnr x i

Modal matrices

For the purpose of modal analysis, it is convenient to introduce the following modal
matrices:

 
r r r
v11 v12 ··· v1n x
r r r
  v21 v22 ··· v2n
 
V R = V1r V2r r x 
··· Vnx =  .. .. .. ..  (6.27)
 . . . . 
vnr x 1 vnr x 2 · · · vnr x nx

 
l l l
v11 v12 ··· v1n x
l l l

iT  v21 v22 ··· v2n
h
T
T 
L l T
V1l V2l
x 
V = ··· Vnx =  .. .. .. .. 
 . . . .  (6.28)
vnl x 1 vnl x 2 ··· vnl x nx


R −1
= V (by virtue of equations (6.25)-(6.26))
 99

and the diagonal matrix

 
λ1 0 · · · 0
 0 λ2 · · · 0 
 
Λ =  .. .. . . ..  (6.29)
. . . . 
0 0 · · · λn x

In terms of the above matrices, equation (6.23) can also be expressed as

AV R = VRΛ ⇒ (V R )−1 A V R = (V R )−1 V R Λ = Λ (6.30)
that is, the matrix A can be diagonalized by its modal matrix V R .

System modes

Since the state matrix A is (normally) not a diagonal matrix, the dynamic of each
state variable of the unforced LTI system (6.18) is a linear combination of the other
state variables. Therefore, it may be difficult to analytically identify the parameters
that have significant impact on the dynamic of each state variable. To overcome this
difficulty, the system (6.18) is (based on the modal matrices) transformed to an LTI
system whose state matrix is diagonalized as follows.
First, a new state vector ξ is established by the transformation
ξ(t) = V L x(t) = (V R )−1 x(t) (6.31)
which implies that
x(t) = V R ξ(t) (6.32)

Substitution of this transformation (i.e. (6.32)) in the original system (6.18) results in
the following transformed system
˙ = A V R ξ(t) ⇒
V R ξ(t)
˙ = (V R )−1 A V R ξ(t)
ξ(t) (6.33)
= Λ ξ(t) (by virtue of equation (6.30))
Based on the above transformation, we are now dealing with an LTI system whose
state matrix is diagonal, i.e. Λ. The dynamic system (6.33) represents nx uncoupled
first-order differential equations of the form
ξ˙i (t) = λi ξi (t) for i = 1 · · · nx (6.34)
whose solutions with respect to time t are given by
ξi (t) = ξi (0) eλi t for i = 1 · · · nx (6.35)
where, ξi (0) is the initial value of ξi (t) at t = 0. From (6.31), we have
ξi (0) = Vil x(0) (6.36)
100

which is a scalar, and will henceforth be denoted by αi . Moreover, x(0) is the initial
values of all the original state variables x(t) at t = 0.
Next, by virtue of the expression for x in (6.31) and also equations (6.35)-(6.36), the
solution of the original dynamic system (6.18) can be expressed by
nx
X nx
 l  λi t X
x(t) = Vir Vi x(0) e = Vir αi eλi t (6.37)
i=1 i=1

or for the k-th state variable

nx
X
r
xk (t) = vki αi eλi t (6.38)
i=1

Equation (6.35) gives the each dynamic mode (corresponding to eλi t ) of the system
with magnitude αi . However, the dynamic response of each state variable is given by
equation (6.38) which is a linear combination of nx dynamic modes.
From equation (6.32) (and also (6.37)) we can see that the right eigenvector Vir de-
scribes how each dynamic mode is distributed among the system states x that is it
describes the mode shape. From equation (6.36) we can however see that the left
eigenvector Vil weighs the contribution of the initial condition x(0) to the i-th mode.
Thus, the k-th element of Vir measures the activity of the state variable xk in the i-th
mode, and the k-th element of Vil weighs the contribution of this activity to the i-th
mode [15].

Eigenvalue sensitivity

Let akj be the element of the state matrix A in the k-th row and j-th column. The
sensitivity of the eigenvalue λi to akj is then determined by differentiating (6.23) with
respect to akj which yields

∂A r ∂V r ∂λi r ∂V r
Vi + A i = V i + λi i (6.39)
∂akj ∂akj ∂akj ∂akj

Pre-multiplication of (6.39) by Vil gives

∂A r ∂V r ∂λi r ∂V r
Vil Vi + Vil A i = Vil Vi + Vil λi i
∂akj ∂akj ∂akj ∂akj
or
∂λi ∂A r ∂V r
Vil Vir = Vil Vi + (Vil A − Vil λi ) i
| {z } ∂akj ∂akj | {z } ∂akj
=1 =0

which results in
∂λi l r
= vik vji (6.40)
∂akj
 101

∂A
since all elements of ∂akj
are zero, except for the element in the k-th row and j-th
column which is equal to 1.
Information about the sensitivity of eigenvalues to system parameters such as excitation
system gain, generator inertia and line reactance is of importance in power system
stability analysis and control. Eigenvalue sensitivity can also be used to ascertain
which power system parameters have a major impact on damping of particular modes.

Participation factor

For small-signal stability and control, it is of importance to measure properly the

participation of state variables within a mode i. The right eigenvector Vir might be
a possible choice for this issue since its elements measure the activity of the state
variables in the mode i. However, these elements are dependent on the dimensions
and scaling of the state variables which are incommensurable. As a solution to this
problem, a related but dimensionless measure pki termed participation factor has been
presented in [15]. The participation factor pki is a measure of the relative participation
of the k-th state variable in the i-th mode, and vice versa. It is given by
l r r l
pki = vik vki = vki vik (6.41)
r
Since the vki measures the activity of the state variable xk in the i-th mode and the
l
vik weighs the contribution of this activity to the mode, the product pki measures the
net participation. Furthermore, a comparison between (6.41) and (6.40) shows that
∂λi
= pki (6.42)
∂akk

For the purpose of small-signal analysis, it is convenient to introduce a matrix P

containing all participation factors. This matrix is termed participation matrix, and
has the form

P = [P1 P2 ··· Pnx ]

   
r l r l r
p11 p12 ··· p1nx v11 v11 v12 v21 ··· v1n vl
x nx 1
r l r l r (6.43)
 p21
 p22 ··· p2nx  
  v21 v12 v22 v22 ··· v2n vl
x nx 2


= .. .. .. .. = .. .. .. .. 
 . . . .   . . . . 
pn x 1 pn x 2 · · · pn x n x vnr x 1 v1n
l
x
vnr x 2 v2n
l
x
· · · vnr x nx vnl x nx

Modal controllability and observability

Consider the LTI system (6.17). By applying the transformation (6.32) to (6.17), the
following is obtained
˙
ξ(t) = Λ ξ(t) + V L B U(t) (6.44)
R
Y(t) = C V ξ(t) (6.45)
102

Equation (6.44) can be written as nx uncoupled equations

r
X r
X
ξ˙i (t) = λi ξi (t) + Vil Bj uj (t) = λi ξi (t) + cij uj (t) for i = 1 · · · nx (6.46)
j=1 j=1

where, cij = Vil Bj , Vil is the i-th row of V L , Bj is the j-th column of B and uj (t) is
the j-th element of U(t).
Obviously, the input uj has no effect on the i-th mode if cij = 0. Thus, the i-th mode
is controllable if and only if there is any cij 6= 0.
The nx × r matrix V L B is termed as the mode controllability matrix whose element
in the i-th row and j-th column is cij . By inspecting this matrix, the controllable (or
uncontrollable) modes can be identified. For instance, the i-th mode is uncontrollable
if the i-th row of this matrix is zero.
Equation (6.45) can also be written as
nx
X nx
X
Yj (t) = Cj Vir ξi (t) = oji ξi (t) for j = 1 · · · m (6.47)
i=1 i=1

where, oji = Cj Vir , Cj is the j-th row of C and Vir is the i-th column of V R .
It is evident that the i-the mode cannot be observed in the j-th output variable (i.e.
Yj (t)) if oji = 0. Thus, the i-the mode is observable if and only if there is any oji 6= 0.
The m × nx matrix C V R is termed as the mode observability matrix whose element
in the j-th row and i-th column is oji . By inspecting this matrix, the observable (or
unobservable) modes can be identified. For instance, the i-th mode is unobservable if
the i-th column of this matrix is zero.

Residues

Consider again equations (6.44). Let m = r = 1, i.e. a Single Input Single Output
(SISO) system. Taking Laplace transform, the following is obtained
ξ(s) = (s1 − Λ)−1 V L B U(s) (6.48)

Taking Laplace transform of (6.45), and substituting (6.48) into it, we have then
Y(s) = C V R (s1 − Λ)−1 V L B U(s) (6.49)
where, s is the Laplace operator. Since Λ is a diagonal matrix, the transfer function of
(6.17) can be expressed in partial function as
x n
Y(s) X R̄i
G(s) = = (6.50)
U(s) i=1
s − λi

where, R̄i is the residue of G(s) at λi and is expressed as

R̄i = C Vir Vil B (6.51)
 103

Note that Ri = |R̄i | is indeed the product of the controllability and observability of
the i-th mode.
Having a feedback transfer function of the form H(s, k) = kH(s) (where, k is a constant
gain) between output and input, it can be shown that [16]

∂λi
= R̄i H(λi ) (6.52)
∂k
For small values of gain, equation (6.52) can be written as

∆λi
= R̄i H(λi ) (6.53)
∆k
Thus, if a feedback transfer function is added to the system, the i-th eigenvalue will
be changed as
∆λi = R̄i H(λi , k) (6.54)
Equations (6.53)-(6.54) will be used later on designing a PSS for improving damping
of electromechanical oscillations.

6.2.2 Small signal stability of power systems

Let the dynamic of a power system be described by

ẋ = f (x, y)
(6.55)
0 = g(x, y)

where, f (x, y) and g(x, y) are given by equations (5.7)-(5.11).

By linearizing the above non-linear system around an operating point (x0 , y0), the small
signal analysis can be applied to the linearized system. In Example 5.1, it has been
shown how (x0 , y0 ) can be calculated.
The linearized system is given by

∆ẋ = fx ∆x + fy ∆y (6.56)
0 = gx ∆x + gy ∆y (6.57)

where
   
∂f (x, y) ∂f (x, y)
fx = , fy =
∂x x=xo ; y=yo ∂y x=xo ; y=yo
   
∂g(x, y) ∂g(x, y)
gx = , gy =
∂x x=xo ; y=yo ∂y x=xo ; y=yo

fx , fy , gx and gy are Jacobian matrices which are explained in Section B.2 of Appendix
B.
 104

From equation (6.57), ∆y can be solved as

∆y = −(gy )−1 gx ∆x (6.58)

Assuming (gy )−1 is non-singular, and substituting ∆y into equation (6.56), the following
is obtained

∆ẋ = fx − fy (gy )−1 gx ∆x = As ∆x (6.59)
where, As is the overall system state matrix.
Equation (6.59) is the linearized system of the non-linear system (6.55). The eigenval-
ues of the linearized system are given by equation (6.19) where the state matrix A is
replaced by As . Note however that the linearized system (6.59) has (at least) two zero
eigenvalues which are not of interest in this analysis. The first zero eigenvalue is due to
use of absolute changes in rotor angles (i.e. ∆δ) as state variables. Since angles appear
as differences, an equal change in each of the rotor angles has no effect on power flow
equations. Therefore, the state matrix columns associated with the rotor angles are
linearly dependent, and make the state matrix As singular. The second zero eigenvalue
is due to zero damping constant, i.e. Di = 0. If there is any non-zero damping constant
in the system, the second zero eigenvalue vanishes. If the system contains an infinite
bus, these two eigenvalues vanish.

Example 6.1 Consider the system shown in Figure 6.2.

U1 U3 U2
jx13 jx23 Infinite
Gen Bus

SL
Figure 6.2. A single generator connected to an infinite bus.

The system data is given as follows (all values are expressed in pu):

Gen: One-axis model is used with H = 4, D = 0, x′d = 0.15, xd = 1, xt = 0.1, Tdo

′
=6
and Pm = 1

Network: Ū1 = 1∠θ1 , Ū2 = 1∠0 (infinite bus), x13 = 0.3, x23 = 0.5 the load model (5.2) is
used with PL0 = 0.7, QL0 = 0.01, mp = 1 and mq = 2.

a) Linearize the system around its equilibrium point, and calculate the eigenvalues.
Calculate also the damping ratio and frequency of the oscillatory mode.

b) Draw the block diagram of the linearized system and calculate the synchronizing
and damping torque coefficients at the oscillatory mode.
 105

c) Simulate the non-linear dynamic system for the following cases

Case 1: A three-phase fault occurs at bus 3. The fault is cleared after 100 (ms) (large
disturbance).
Case 2: 20% of the active load is disconnected during 100 (ms) (small disturbance).

a) The system dynamic is given by

ẋ = f (x, y)
0 = g(x, y)

From load flow, x0 and y0 are calculated.

x0 = [δ0 ω0 Eq′ 0 ]T = [40.2667 (deg.) 0 1.1001]T

y0 = [θ10 θ30 U10 U30 ]T = [27.1319 (deg.) 8.9674 (deg.) 1 0.9623]T

Linearizing the non-linear system around (x0 , y0 ), the following is obtained

∆ẋ = As ∆x + B U

or
      
∆δ̇ a11 a12 a13 ∆δ 0 0  
 ∆ω̇  = a21 a22 a23   ∆ω  +  M1 ∆Pm
0 
′ 0 1 ∆Ef
∆Ėq ′ a31 a32 a33 ∆E q ′
Tdo
    
0 1 0 ∆δ 0 0  
∆Pm
= −27.1507 0 −37.6825
   ∆ω + 39.2699
  0 
′ ∆Ef
−0.1094 0 −0.2912 ∆Eq 0 0.1667

where, ∆Pm = ∆Ef = 0.

The eigenvalues are

λ1 , λ2 = −0.0758 ± j 5.2081
λ3 = −0.1395

Therefore, the equilibrium point is asymptotically stable. The damping ratio and the
frequency of the oscillatory mode (λ1 or λ2 ) are
−σ1
ζ1 = = 0.0146
|λ1 |
ωp 1
fp1 = = 0.8289 (Hz)
2π
b) Figure 6.3 shows the block diagram of the linearized system which is redrawn as
shown in Figure 6.4 to have a representation similar to the well-known K representa-
tion.
106

− a31

∆Pm

1
M
- ∆Pe
∆E f ∆E′q + ∆ω ∆δ
1 + 1 + M 1 1
Tdo′
Σ s − a 33
− a23 Σ - Σ s s
+
Field
circuit
− a21

Figure 6.3. Block diagram representation with one-axis model.

K4

∆Pm = 0

∆E f = 0
- + ∆ω ∆δ
+ K3 ∆E′q + ∆Pe 1 1
Σ 1 + T3 s
K2 Σ - Σ Ms s
+
Field
circuit K1

Figure 6.4. The well-known K block diagram representation with one-axis model.

In Figure 6.4

K1 = − M a21 = 0.6914 , K2 = −M a23 = 0.9596

1 T3
T3 = − = 3.4339 , K3 = ′ = 0.5723 (6.60)
a33 Tdo
′
K4 = − Tdo a31 = 0.6563

In the steady-state, ∆Pe = ωωgs ∆Te = ∆Te (expressed in (pu)), since ωs = ωg . In the
transient-state based on the assumption ωωgs ≈ 1, we may also assume that ∆Pe ≈ ∆Te .
Therefore, the synchronizing and damping torque coefficients at the oscillatory mode
λ1 are calculated by setting s = λ1 = σ1 + j ωp1 as follows.
From the block diagram of Figure 6.4

1 1
∆δ = ∆ω = ∆ω ⇒
s σ1 + j ωp1
(6.61)
1 σ1
j∆δ = ∆ω − ∆δ
ωp 1 ωp 1
 107

Also,

∆Te = ∆Pe = K1 ∆δ + K2 ∆Eq′

= K1 ∆δ + K2 F (s = λ1 ) ∆δ
(6.62)
= K1 ∆δ + K2 (Fre + jFim ) ∆δ
= K1 ∆δ + K2 Fre ∆δ + K2 Fim j∆δ

where,
K3 K 4 K 3 K4
F (s = λ1 ) = − =− = Fre + jFim
1 + T3 s 1 + T3 λ1

Substituting j∆δ from (6.61) into (6.62), the following is obtained

    
σ1 K2 Fim
∆Te = K1 + K2 Fre − Fim ∆δ + ∆ω
ωp 1 ωp 1
= (K1 + KS |1axis ) ∆δ + KD |1axis ∆ω (6.63)
= KS ∆δ + KD ∆ω
= 0.6908 ∆δ + 0.0039 ∆ω

The synchronizing and damping torque coefficients can also be calculated as follows.
From the block diagram of Figure 6.4
1 1 σ1 − j ωp1
∆δ = ∆ω = ∆ω = ∆ω ⇒
s σ1 + j ωp1 |λ1 |2 (6.64)
jωp1 ∆ω = σ1 ∆ω − |λ1 |2 ∆δ

Also,

∆Te = ∆Pe = −M s ∆ω = −M [(σ1 + j ωp1 ) ∆ω]

= − M [σ1 ∆ω + j ωp1 ∆ω]
= − M [σ1 ∆ω + σ1 ∆ω − |λ1 |2 ∆δ]
(6.65)
= M |λ1 |2 ∆δ + (−2 M σ1 ) ∆ω
= KS ∆δ + KD ∆ω
= 0.6908 ∆δ + 0.0039 ∆ω

Comments:

• With the classical model ∆Eq′ = 0. Therefore, the synchronizing torque coefficient
would be KS = K1 = 0.6914. Moreover, since there is no inherent damping in
the system (i.e. D = 0 ), the damping torque coefficient would be KD = D = 0.

• With one-axis model, we see that a positive damping torque component exists, i.e.
KD = KD |1axis = 0.0039. This is due to variation of Eq′ since it is a state variable.
Having an inherent damping in the system, the damping torque coefficient is then
108

KD = D + KD |1axis . The effect of field flux variation (i.e. Eq′ ) on the system
damping can also be observed in (6.12) which is the time derivative of the energy
function (i.e. V̇) along the non-linear system. Since V̇ < 0, the equilibrium point
is asymptotically stable that is the system is dissipative (or there is a positive
damping in the system). If Eq′ is constant, then Ė ′ q = 0 which results in V̇ = 0
that is the system is conservative (or there is no damping in the system).
• The contributions of the state variable Eq′ to the synchronizing and damping
torques coefficients are KS |1axis = KS − K1 = −0.0006 and KD |1axis = 0.0039
that is a (small) negative contribution to the synchronizing torque, but a (small)
positive contribution to the damping torque.

c) Figure 6.5 shows variations of the rotor angle for each case. Obviously the system
is transiently unstable for case 1, however it is stable for case 2 (small-signal stability).
For case 2, we see that the amplitude of the oscillation decays with time that is there
is a positive damping in the system. Note that the damping ratio is about 1.5%. It
can also be observed that the oscillation frequency is about
1 1 1
= fp = = = 0.8264 (Hz)
T t2 − t1 3.27 − 2.06
ωp
Thus, the calculated frequency fp = = 0.8289 (Hz) (which is based on the lin-
2π
earized system) gives a good estimation of the oscillation frequency.
43
Case 2
Case 1
42

41
δ (deg.)

40

39

40 38
< T >
37
0 0.5 1.4 0 1 2 3 4 5
Time (s) Time (s)

Figure 6.5. Variations of rotor angle with one-axis model.

Example 6.2 To improve the transient stability of the above system, the AVR shown
in Figure 3.14 is added to the generator with Te = 0.01 and KA = 150 (Efmax and
Efmin are not considered in this example).

a) Simulate the non-linear dynamic system for the two cases in Example 6.1.
b) Based on the small-signal analysis, describe the effect of the AVR on the system
stability.
 109

a) Figure 6.6 shows variations of the rotor angle for each case. For case 1, it is
obvious that the AVR with its high gain (i.e. KA ) has improved the transient (or first
swing) stability. However, it has introduced a negative damping in electromechanical
oscillations which results in oscillatory instability. This phenomenon is a typical small-
signal stability problem in today’s power systems. In case 2, small-signal instability is
obvious.
110 110
Case 1 Case 2
δ (deg.)

40 40

−25 −25
0 1 2 3 4 5 0 10 19
Time (s) Time (s)

Figure 6.6. Variations of rotor angle with AVR in the system.

b) Because of the AVR, the differential equation (3.101) with UP SS = 0 is added to

the f (x, y). From load flow, x0 and y0 are calculated.
x0 = [δ0 ω0 Eq′ 0 Ef0 ]T = [40.2670 (deg.) 0 1.1001 1.5296]T
y0 = [θ10 θ30 U10 U30 ]T = [27.1320 (deg.) 8.9674 (deg.) 1 0.9623]T

Linearization of the non-linear system around (x0 , y0) gives

∆ẋ = As ∆x(t) + B U(t)
or
      
∆δ̇ a11 a12 a13 a14 ∆δ 0
 ∆ω̇  a21 a22 a23 a24   ∆ω   0 
   
∆U
 ∆Ėq′  =a31 +
  
a32 a33 a34   ∆Eq′   0  ref
KA
∆E˙f a41 a42 a43 a44 ∆Ef Te
     (6.66)
0 1 0 0 ∆δ 0
 −27.1507 0 −37.6825 0    ∆ω′  +  0 ∆Uref
   
=
 −0.1094 0 −0.2912 0.1667   ∆Eq   0 
2283.9790 0 −11436.1545 −100 ∆Ef 15000

The eigenvalues are

λ1 , λ2 = 0.1752 ± j 5.8155
λ3 = −74.1635
λ4 = −26.4780
110

The damping ratio and the frequency of the oscillatory mode are
−σ1
ζ1 = = −0.0301
|λ1 |
ωp 1
fp1 = = 0.9256 (Hz)
2π
Obviously, the equilibrium point is unstable.
Figure 6.7 shows the block diagram of the linearized system.

K4

∆Pm = 0

- + ∆ω ∆δ
∆U ref + KA ∆E f + K3 ∆E′q
+ ∆Pe 1 1
Σ 1 + Te s
Σ 1 + T3 s
K2 Σ - Σ Ms s
- +
∆ U1 Exciter Field
circuit K1

K6

+
+
Σ K5

Figure 6.7. Block diagram representation with AVR.

where,
KA KA
K5 = − a41 = −0.1523 and K6 = − a43 = 0.7624
Te Te

From the block diagram of Figure 6.7 (with ∆Uref = 0)

∆Te = K1 ∆δ + K2 ∆Eq′
= K1 ∆δ + K2 F (s = λ1 ) ∆δ
= (K1 + KS |AV R ) ∆δ + KD |AV R ∆ω (6.67)
= KS ∆δ + KD ∆ω
= 0.8620 ∆δ − 0.0089 ∆ω

where,
K3 K5 KA + K3 K4 (1 + Te s)
F (s = λ1 ) = − = Fre + jFim
(1 + Te s)(1 + T3 s) + K3 K6 KA

The contributions of the AVR to the synchronizing and damping torques coefficients
are KS |AV R = KS −K1 = 0.1706 and KD |AV R = −0.0089 that is a positive contribution
to the synchronizing torque, but a negative contribution to the damping torque.
 111

As discussed earlier, participation factors are useful in identifying those states which
have the most impact on a selected mode. Furthermore, pki gives the sensitivity of λi
to the diagonal element akk of the state matrix. The participation matrix (only the
magnitudes of the participation factors are of interest) of system (6.66) is
λ1 λ2 λ3 λ4
 
0.4933 0.4933 0.0007 0.0144 ∆δ
 ∆ω′
0.4933 0.4933 0.0007 0.0144
|P| = 
0.0295 0.0295 0.5386 1.5087 ∆Eq
0.0018 0.0018 1.5401 0.5375 ∆Ef
From the participation matrix we see that ∆δ and ∆ω are the state variables which
have the most influence on the unstable mode. Furthermore, the state variables ∆Ef
and ∆Eq′ are the state variables which have most influence on the third and forth
modes, respectively.
To stabilize the unstable mode, a power system stabilizer (PSS) is added in the ex-
citation system. To provide positive damping, the PSS must be tuned such that it
produces a component in phase with ∆ω.

Example 6.3 Use the PSS shown in Figure 3.17 without the washout block and with
only one lead-lag filter. Let ω be the input signal.

a) For the unstable mode (i.e. s = λ1 ), tune the lead-lag block of the PSS so that
the PSS produces a component in phase with ∆ω, and find a value for KP SS so
that the damping ratio of this mode is about 15%.
b) Simulate the non-linear dynamic system for the two cases in Example 6.1.
c) Linearize the system, and calculate the eigenvalues. Calculate also the damping
ratio and frequency of the oscillatory mode.

a) Figure 6.8 shows the block diagram representation with PSS. In the figure
1 + T1 s
∆UP SS =KP SS H(s) ∆ω = KP SS ∆ω
1 + T2 s
=KP SS |H(s)| ej arg(H(s)) ∆ω

From the block diagram of Figure 6.8 (with ∆Uref = 0)

∆Te = K1 ∆δ + K2 ∆Eq′
= K1 ∆δ + K2 (∆Eq′ |AV R + ∆Eq′ |P SS ) (6.68)
= (K1 + KS |AV R ) ∆δ + KD |AV R ∆ω + K2 ∆Eq′ |P SS

where
KA K 3
∆Eq′ |P SS =F (s) ∆UP SS = ∆UP SS
(1 + Te s)(1 + T3 s) + K3 K6 KA (6.69)
=KP SS |H(s)| |F (s)| ej ϕ ∆ω
112

PSS

K4

∆Pm = 0
∆U PSS
+ - + ∆δ
∆U ref + KA ∆E f + K3 ∆E′q
+ ∆Pe 1 1
Σ 1 + Te s
Σ 1 + T3 s
K2 Σ - Σ Ms ∆ω s
- +
∆ U1 Exciter Field
circuit K1

K6

+
+
Σ K5

Figure 6.8. Block diagram representation with PSS.

where,
ϕ = arg(H(s)) + arg(F (s))
From equation (6.69) we can see that ∆Eq′ |P SS will be in phase with ∆ω if ϕ = 0.
Let φ = −arg(F (s = λ1 )) = 16.9619 (deg.). Then, ∆Eq′ |P SS will be in phase with ∆ω
if    
ωp1 T1 ωp1 T2
arctan − arctan =φ (6.70)
1 + σ1 T1 1 + σ1 T2

Now, by setting a value for T1 (or T2 ), then T2 (or T1 ) can easily be determined.
T1 and T2 can also (approximately) be determined by setting s = λ1 ≈ jωp1 (i.e.
σ1 ≈ 0 in equation (6.70)), and

1
T1 = α T , T2 = T where, T = √ and α > 1 (6.71)
ωp 1 α
Then from equation (6.70) (with σ1 ≈ 0), α and T can easily be determined as follows

q 2 1 + sin(φ)
α = 1 + 2 tan2 (φ) + (1 + 2 tan2 (φ)) − 1 = = 1.8238
1 − sin(φ)
1
T = √ = 0.1273 (6.72)
ωp 1 α
T1 = 0.2322
T2 = 0.1273
 113

With the above T1 and T2 , the total damping torque coefficient is then
KD = KD |AV R + K2 KP SS |H(s)| |F (s)| (6.73)
Note that |H(s)| and |F (s)| are the magnitudes of the transfer functions H(s = λ1 )
and F (s = λ1 ) (with λ1 = σ1 + jωp1 ), respectively.
The desired damping ratio for the unstable mode is ζdes = 0.15. Based on equations
(6.22), (6.65) and (6.73), KP SS can be determined as follows so that the ζdes is obtained.
ζdes ωp1
σdes = − p 2
= −0.8823
1 − ζdes
KDdes = − 2 M σdes = 0.0449 (6.74)
KDdes − KD |AV R
KP SS = = 0.0333
K2 |H(s)| |F (s)|
Note that the aim is to shift only the real part of the unstable mode (i.e. σ1 = 0.1751) to
the new position σdes = −0.8823. Therefore, we let the imaginary part be unchanged,
i.e. ωpdes = ωp1 = 5.8155.
b) Because of the power system stabilizer, the differential equation (3.102) is added to
the f (x, y) of Example 6.2. Note that in differential equation Ėf , the auxiliary signal
UP SS is not longer zero since it is now a state variable.
From load flow, x0 and y0 are calculated.
x0 = [δ0 ω0 Eq′ 0 Ef0 UP SS 0 ]T = [40.2670 (deg.) 0 1.1001 1.5296 0]T
y0 = [θ10 θ30 U10 U30 ]T = [27.1320 (deg.) 8.9674 (deg.) 1 0.9623]T

Figure 6.9 shows variations of the rotor speed for each case. It is obvious that not only
the system is stabilized, but also it is well-damped for both cases.
70

Case 1
42 Case 2
δ (deg.)

40 40

39
25
0 2 4 6 8 0 2 4 6 8
Time (s) Time (s)

Figure 6.9. Variations of rotor angle with PSS in the system.

c) Linearization of the non-linear system around (x0 , y0) gives

∆ẋ = As ∆x(t) + B U(t)
114

or
     
∆δ̇ a11 a12 a13 a14 a15 ∆δ 0

 ∆ω̇  a21
  a22 a23 a24 a25   ∆ω   0 
   

  =a31
∆Ėq′ a32 a33 a34 a25   ∆Eq′  +  0 ∆Uref
      
 ∆E˙f a41 a42 a43 a44 a45   ∆Ef   KTeA 
∆U̇P SS a51 a52 a53 a54 a55 ∆UP SS 0
    
0 1 0 0 0 ∆δ 0
 −27.1507 0 −37.6825 0 0   ∆ω   0 
    
= −0.1094
 0 −0.2912 0.1667 0   ∆Eq′  +  0 ∆Uref
   
2283.9790 0 −11436.1545 −100 15000   ∆Ef  15000
−1.6512 0.2619 −2.2917 0 −7.8537 ∆UP SS 0

The eigenvalues are

λ1 , λ2 = −0.9136 ± j 5.6900
λ3 = −75.7613
λ4 = −20.3710
λ5 = −10.1854

The damping ratio and the frequency of the oscillatory mode are
−σ1
ζ1 = = 0.1585
|λ1 |
ωp 1
fp1 = = 0.9056 (Hz)
2π
We see that σ1 , ωp1 and ζ1 do not have exactly the values which were desired. However,
the differences are small.

Example 6.4 For the unstable mode (i.e. λ1 ), use equations (6.51)-(6.54) to tune the
PSS so that the damping ratio of this mode is about 15%. Let T1 and T2 be defined as
in equation (6.71).

We select ∆ω as output variable, to have the following system which is of form given
by (6.17)

ẋ(t) = A x(t) + B U(t)

(6.75)
Y(t) = C x(t)

where, x(t), A, B and U(t) are given in equation (6.66), and C = [0 1 0 0]. The
transfer function of the system without PSS is given by
4
Y(s) ∆ ω(s) X R̄i
G(s) = = = (6.76)
U(s) ∆ Uref (s) i=1
s − λi
 115

The residue of the unstable mode is given by

R̄1 = C V1r V1l B = −21.968 + j7.3534 = 23.166 ∠161.49 (deg.) (6.77)

Adding a PSS in the generator, we are then dealing with the following feedback control
system
∆ U ref ( s ) + ∆ω ( s )
Σ G( s)
+

H ( s, K PSS )

Figure 6.10. Feedback control system.

It is well-known that the residue of an eigenvalue gives the direction of the eigenvalue
departure for a small change of system parameters.
Im
R1

arg( R1 )
f
* *
Desired eigenvalue Actual eigenvalue

Re

Figure 6.11. Direction of the eigenvalue departure for small changes.

From equation (6.54) and Figure 6.11, it is obvious that to move the actual eigenvalue
to the desired position (only the real part is changed) the argument of H(s = λ1 )
(denoted by φ) must be φ = 180 − arg(R̄1).
In this case the residue argument of the interest mode is arg(R̄1 ) = 161.49 (deg.).
Substituting φ = 18.51 into equation (6.72), T1 = 0.2389 and T2 = 0.1238 are obtained.
From equation (6.54) we have
∆λ1 = σdes − σ1 = R̄1 H(s, k) = KP SS R1 |H(s)| ej π ⇒
|∆λ1 |
KP SS = = 0.0329
R1 |H(s)|
where, s = λ1 , R1 = 23.166 is the magnitude of the residue R̄1 and |H(s)| = 1.3895 is
the magnitude of the lead-lag block of the PSS.
We can see that the obtained values to tune the PSS with this method (also known as
the residue technique) are almost similar to the values obtained in Example 6.3.
116

Example 6.5 Consider again the lossless system shown in Figure 5.6. The one-axis
model is used for the generators, and the loads are considered as impedance loads. Fur-
thermore, all generators are equipped with AVR shown in Figure 3.14 where Ef max = 5
and Ef min = −5 (pu). By using the PSS shown in Figure 3.17, make an appropriate
damping in the system.
(See the MATLAB files for ”Project D2”)
We firstly analyse the system without PSS. The RNM is used to describe the dynamic of
the system. Figure 6.12 shows the simulation results for a fault at BUS 4 with clearing
time tc = 0.1 (s). The plotted rotor angles are defined in the COI reference frame.
As shown in the figure, the system is unstable. The variations of the field voltages of
the generators are shown in the figure. As the system approaches its separation (i.e.
losing its synchronism), the AVRs behave in a ”bang-bang” control manner between
their limits.
100
5
∆ δCOI (deg.)

Ef (pu)

0 0

−5
−100
0.5 5 9 0.5 5 9
Time (s) Time (s)

Figure 6.12. Variations of ∆δ in the COI reference frame and Ef of Gen1

(blue), Gen2 (red) and Gen3 (black).

Next, the system is linearized around its stable equilibrium point (see also Appendix
B). Table 6.1 shows some results from the linear analysis of the system.

Mode λ fp ζ
1 -80.3517 0 1
2 -86.7215 0 1
3 -89.7353 0 1
4 -19.9922 0 1
5 0.1918 + j 7.2317 1.1510 -0.0265
6 0.1918 - j 7.2317 1.1510 -0.0265
7 0.0136 + j 10.0342 1.5970 -0.0014
8 0.0136 - j 10.0342 1.5970 -0.0014
9 -13.8411 0 1
10 -10.6692 0 1
11 0.0022
12 -0.0022
Table 6.1. Some results from the linear analysis of the system.
 117

The last two eigenvalues can be considered as zero eigenvalues. As shown in the table,
since the system has 3 × 4 = 12 state variables, there are twelve eigenvalues two of
which are zeros (why?). The linearized system has two unstable oscillatory modes,
namely mode 5 and mode 7. Below, the participation matrix (with the magnitudes
of the participation factors) of these two unstable modes is given. In mode 5, the
participation factor of the rotor speed of the Gen 1 has the highest value. However in
mode 7, the participation factor of the rotor speed of the Gen 3 has the highest value.
λ5 λ7
 
0.2618 0.0001 ∆δ1
 ∆δ2
0.1367 0.2076

 ∆δ3
0.0919 0.2917

 ∆ω1
0.2618 0.0001

 ∆ω2
0.1367 0.2076

 ∆ω′3
0.0919 0.2917
|P| =  (6.78)
 ∆Eq1
0.0270 0.0000
 ′
 ∆Eq2
0.0010 0.0008

0.0004 ′
 0.0005
 ∆Eq3
0.0021 0.0000
  ∆Ef 1
0.0001 0.0001 ∆Ef 2
0.0000 0.0001 ∆Ef 3

Figure 6.13 shows the mode shapes of the unstable modes based on the right eigenvector
(the elements corresponding to the rotor angles) of each mode, Gen 1 (in blue), Gen 2
(in red) and Gen 3 (in black).
90 90
1 1
120 60 120 60
0.8 0.8

0.6 0.6
150 30 150 30
0.4 0.4

0.2 0.2

180 0 180 0

210 330 210 330

240 300 240 300

270 270

Figure 6.13. The mode shapes of the unstable modes.

From Table 6.1, it can be found that mode 5 has the lowest damping ratio and fre-
quency. From the participation matrix and the compass plots, it can be concluded that
Gen 1 is mostly participated in this mode. Therefore, a PSS is used in Gen 1 with ω1
as its input signal. Based on the linearized system (6.17), U = ∆U1ref and
KA1
B =[0 0 0 0 0 0 0 0 0 0 0]T i.e. of order 12 × 1
Te1 (6.79)
C =[0 0 0 1 0 0 0 0 0 0 0 0] i.e. of order 1 × 12
the feedback control system is shown in Figure 6.14 which will be studied.
118

D U1ref + x = Ax
A + Bu D w1
S
+ y = Cx
u pss
1 + T3 s 1 + T1s Tw s
K PSS
1 + T4 s 1 + T2 s 1 + Tw s

Figure 6.14. The closed-loop of the linearized system with a PSS in Gen1.

To find appropriate values for T1 -T4 , the residue of the selected unstable mode and the
angle φ must firstly be calculated, and they are

R̄5 = CV5r V5l B = 12.0799∠ 146.2486 ⇒ φ = 180 − arg(R̄5 ) = 33.7514 (6.80)

Let ”nf ” be the number of the lead-lag filters. Then, if

0 < |φ| ≤ 60 ⇒ nf = 1
60 < |φ| ≤ 120 ⇒ nf = 2 (6.81)
jπ
120 < |φ| ≤ 180 ⇒ nf = 1 set φ = −arg(R̄i ) , KP SS = KP SS e

Now, T1 -T4 can be obtained by

1 + sin( nφf ) 1
α= , T = √ , T1 = α T , T2 = T
1− sin( nφf ) ωp α
(6.82)
nf =1 ⇒ T3 = T4 to remove the second lead-lag filter.
nf =2 ⇒ T3 = T1 T4 = T2

The above equations are also valid for a negative arg(R̄i ), but then φ = −180−arg(R̄i ).
In this example for the PSS of Gen 1, nf = 1, T1 = 0.2587 and T2 = 0.0739 are obtained.
Setting T3 = T4 = 1, the contribution of the second lead-lag filter is removed. Having
obtained the parameters of the PSS, the root locus diagram of the feedback control
system as a function of KP SS can be drawn as shown in Figure 6.15.
Root Locus

50
0.28 0.2 0.14 0.095 0.06 0.03

40
40

0.42 30
30

20
20

0.7
10
10
Imaginary Axis

−10
10
0.7

−20
20

−30
0.42 30

−40
40

0.28 0.2 0.14 0.095 0.06 0.03

−50
−14 −12 −10 −8 −6 −4 −2 500 2
Real Axis

Figure 6.15. The root locus of the system modes and the selected unstable mode 5.
 119

The next step is to choose an appropriate gain KP SS to stabilize the unstable mod 5,
but not destabilize the other modes. To find an appropriate gain, the eigenvalues of
the closed-loop system will be determined for KP SS = [0 : 0.001 : 0.2]. For each KP SS
the least damping ratio of all the oscillatory modes is obtained which will be plotted
as shown in Figure 6.16. The appropriate gain is the one which gives the maximum of
the least damping ratios, i.e. KP SS = 0.0920.

0
−0.0013

ζmin

−0.0265
0 0.092 0.16
KPSS

Figure 6.16. The least damping ratios of all the oscillatory modes versus the gain KP SS .

Table 6.2 shows some linear analysis results of the closed-loop system with the obtained
T1 , T2 and KP SS for the first PSS which is installed in Gen 1. As shown in the table, the

Mode λ fp ζ
1 -89.7262 0 1
2 -88.1906 0 1
3 -82.1920 0 1
4 -12.2383 + j 13.2524 2.1092 0.6784
5 -12.2383 - j 13.2524 2.1092 0.6784
6 -15.5262 0 1
7 -10.6337 0 1
8 0.0129 + j 10.0337 1.5969 -0.0013
9 0.0129 - j 10.0337 1.5969 -0.0013
10 -1.4835 + j 6.2597 0.9963 0.2306
11 -1.4835 - j 6.2597 0.9963 0.2306
12 -0.8440 0 1
13 0.0007
14 -0.0007
Table 6.2. Some results from the linear analysis with a PSS in Gen1.

system is still unstable since it has an unstable mode, i.e. mode 8 whose participation
factors corresponding to each state variable are given below where S11 and S21 are the
state variables defined in (3.103).
120

|P| =
′ ′ ′
∆δ1 ∆δ2 ∆δ3 ∆ω1 ∆ω2 ∆ω3 ∆Eq1 ∆Eq2 ∆Eq3 ∆Ef 1 ∆Ef 2 ∆Ef 3 S11 S21
[0.00 0.2093 0.2900 0.00 0.2093 0.2900 0.00 0.0008 0.0005 0.00 0.0001 0.0001 0.0 0.0]T

Figure 6.17 shows the variations of ∆δCOI for the same fault, and the mode shape of
mode 8.
10

90
1
120 60
0.8

0.6
(deg.)

150 30
0.4
0
COI

0.2
∆δ

180 0

210 330

−11
0 2 4 6 8 10 240 300

Time (s) 270

Figure 6.17. Variations of ∆δ in the COI reference frame and the mode shape
of mode 8. Gen 1 (blue), Gen 2 (red) and Gen 3 (black).

From the participation factors and the mode shape in Figure 6.17, it can be concluded
that Gen 3 is mostly participated in this unstable mode, and the second PSS should be
used in Gen 3 whose rotor speed will be the input signal of the PSS. Then, we will have
a similar closed-loop system as shown in Figure 6.14, but with different A, B and C.
Note that the new open-loop system includes the dynamic of the first PSS, therefore
the new A is of order 14 × 14, and the new B and C must be defined.
In a similar manner as described above for the first PSS, we find that nf = 1, T1 =
0.2718 and T2 = 0.0365 for the second PSS. The appropriate gain (KP SS = 0.1320) is
obtained from Figure 6.18.
0.078

ζmin

−0.0013
0 0.132 0.25
K
PSS

Figure 6.18. The least damping ratios of all the oscillatory modes versus the gain KP SS .
 121

The data of each PSS and their impact on the system are summarized in Table 6.3. As
shown in the table, by using the second PSS the system becomes stable and its least
damping ratio is about 8%. The PSS parameters obtained from the linearized system,
are now used in the nonlinear system.
PSS Gen T1 T2 KP SS
ζmin fp
0 -0.0265 1.1510
1 1 0.2587 0.0739 0.0920 -0.0013 1.5969
2 3 0.2718 0.0365 0.1320 0.078 1.4819
Table 6.3. The obtained values of the parameters of each PSS and their impact
on the system.

Figure 6.19 shows the response of the nonlinear system for the same fault at BUS 4.
5
8
∆ δCOI (deg.)

Ef (pu)

0
0

−5
−8
0.5 5 10 0.5 5 10
Time (s) Time (s)

Figure 6.19. Variations of ∆δ in the COI reference frame and Ef of Gen1

(blue), Gen2 (red) and Gen3 (black).

Next, we will study how the other input signals will affect the system damping and
stability. Using ωsime (see equations (6.14) and (6.15)) as input signal, then based on
Figure 6.12 and Figure 6.13 the following ωsime is used for the first PSS,

H1 ω1 H2 ω2 + H3 ω3 H2 ω2 + H3 ω3
−
ωsime1 = = ω1 −
H1 H2 + H3 H2 + H3
which results in the following output vector C.
H2 H3
C = [0 0 0 1 −
− 0 0 0 0 0 0] (6.83)
H2 + H3 H2 + H3
For the second PSS, based on Figure 6.17, the following ωsime is used
H3 ω3 H2 ω2
− ωsime2 =
= ω3 − ω2
H3 H2
Then for the second PSS, we have the following B and C.
KA3
B =[0 0 0 0 0 0 0 0 0 0 0 0 0]T i.e. of order 14 × 1
Te3 (6.84)
C =[0 0 0 0 − 1 1 0 0 0 0 0 0 0 0] i.e. of order 1 × 14
122

Using Pei as input signal (see equation (B.12)), the output vector C is obtained by
equation (B.14).
The data of each PSS and their impact on the system are summarized in Table 6.4.

PSS Gen uin T1 T2 nf KP SS ζmin fp

0 -0.0265 1.1510
ω1 0.2587 0.0739 1 0.0920 -0.0013 1.5969
1 1 ωsime1 0.2480 0.0771 1 0.0520 -0.0013 1.5969
Pe1 0.0440 0.4350 1 -2.6710 -0.0013 1.5969
ω3 0.2718 0.0365 1 0.1320 0.0780 1.4819
2 3 ωsime2 0.2702 0.0368 1 0.1450 0.2942 1.0246
Pe3 0.0462 0.2148 1 -3.8910 0.0809 1.4828
Table 6.4. The obtained values of the parameters of each PSS and their impact
on the system.

Figure 6.20 also shows the response of the nonlinear system for the same fault at BUS
4, with different input signals.

10
uin = ωi (in blue)

uin = ωsime (in red)

uin = Pei (in black)

∆ δCOI (deg.)

−8
0.5 4
Time (s)

Figure 6.20. Variations of ∆δ1 in the COI reference frame with different input signals.

From Table 6.4 and Figure 6.20, it is obvious that ωsime as input signal implies a
better power oscillations damping. However, it should be noted that ωsime is based
on remote information. The geographical distances between the source of information
(i.e. the generators in this example) may be from 500 (km) to 2000 (km). Therefore,
the quality and reliability of the remote information are important factors for power
system stability and control.
Chapter 7
Voltage stability
It is well-known that the reactive power and the voltage are closely coupled. An
injection of reactive power into a bus will result in increasing of the bus voltage, and a
consumption of reactive power at a bus will result in decreasing of the bus voltage. In
this chapter it will be shown that reactive power has a fundamental impact on voltage
stability.
In recent decays, the term “voltage collapse” (or voltage instability) has been used
to describe the reason for the blackouts of December 19, 1978, in France; December
27, 1983, in Sweden; July 23, 1987, in Tokyo; September 23, 2003, in Sweden and
Denmark [17], and September 28, 2003, in Italy [18]. As introduced in Chapter 1,
voltage stability refers to the ability of a power system to maintain steady voltages
at all buses in the system after being subjected to a disturbance from a given initial
operating condition. Voltage instability normally occurs in heavily stressed systems
in the form of a progressive and uncontrollable fall in voltage. A main factor causing
voltage instability is inadequate reactive power supply which is normally a consequence
of load demand increase, line outages, as well as shortage of reactive power resources.
A criterion for voltage stability is that at a given operating point for every bus i

d Qi
>0 (7.1)
d Ui

where Qi is the injected reactive power at bus i. The physical interpretation of (7.1) is
that reactive power injection at a bus i will result in increasing of the voltage magnitude
of bus i. The system is voltage unstable if for any bus i, the condition (7.1) is not
satisfied [6].

7.1 Voltage stability analysis

Like transient stability analysis, voltage stability analysis of a power system is also
an extensive and complicated task. However, it turns out that many of the most
important phenomena and mechanism can be found in very simple systems. In large
and complicated systems it is often hard to distinguish the fundamental and decisive
phenomena from the more irrelevant ones. It is therefore of importance to study simple
systems to get an insight into and understanding of the basics, that can be used in the
analysis of more complex systems.
Consider the Single-Load-Infinite-Bus (SLIB) system shown in Figure 7.1.
This simple system may represent a generation area from which power is delivered to a
load area via a transmission system with long lines (this is indeed a good representation
of the Swedish power system). In the figure, the generation area is considered as a

123
124

UN∠0 parallel lines U L∠ − θ L

Infinite
Bus
SL = PL + jQL

Figure 7.1. SLIB system.

strong system (represented by an infinite bus) and the load area as an equivalent load
represented by S̄L . The transmission system is composed of η identical parallel long
lines. Each line is represented by a series reactance x (i.e. the transmission system is
assumed lossless). The short-circuit power at the load bus is given by

UN2 x
Ssc = (pu), where Xeq = (pu) (7.2)
Xeq η

The short-circuit power Ssc measures the system voltage strength. As seen from a point
in the system, the network can be considered as “weak” if the short-circuit power is
low at this point (normally due to high Xeq ), and strong or stiff if the short-circuit
power is high. At a point in the network with high short-circuit power, switching on a
load will not change the voltage magnitude very much.
At the load bus, the active and reactive load can be expressed as
UN UL
PL = sin(θL )
Xeq
 2  (7.3)
UL UN UL cos(θL )
QL = − −
Xeq Xeq

Since the angle θL is not of interest in this study, it can be eliminated from (7.3) which
results in [19]:
(UL2 )2 + (2QL Xeq − UN2 )UL2 + Xeq2
(PL2 + Q2L ) = 0
Solving for UL2 and using (7.2), the following is obtained:
 q 
2 2 2
UL = Xeq (0.5 Ssc − QL ) ± (0.5 Ssc) − (PL + Ssc QL ) (7.4)

A necessary condition for having a real solution for (7.4) is that

PL2 + Ssc QL ≤ (0.5 Ssc)2 (7.5)

From (7.5) it can be obtained that setting PL = 0, the maximum of purely reactive
load is one fourth of the short-circuit power Ssc that is QLmax = 0.25 Ssc. Thus, it may
be difficult to supply large amounts of reactive power load via transmission system
with long lines, especially via the weak one. Therefore, the required reactive power
load should be compensated locally. From (7.5) it can also be obtained that setting
QL = 0, the maximum of purely active load is one half of the short-circuit power that
is PLmax = 0.5 Ssc .
 125

Theoretically, any amount of active power load can be consumed as long as required
reactive power compensation is available at the load bus. Practically for a normal
operation, the voltage (in the steady-state) should be held within an acceptable range
at all load levels.
For the SLIB system shown in Figure 7.1, let UN = 1.044, x = 1.2, η = 4 and
sin(ϕ)
QL = PL = PL tan(ϕ), where cos(ϕ) is the load power factor, and assumed
cos(ϕ)
constant.
Figure 7.2 shows the variation of the voltage at the load bus versus the active load with
tan(ϕ) = 0 (i.e. purely active load). The diagram is known as U-P curve, P-U curve
and also nose curve. As shown in the figure, PLmax = 1.8166 (pu) which is indeed one
half of Ssc = 3.6331 (pu) since QL = 0. For PL > PLmax , it is obvious that the system is
unstable since there is no solution (or operating point). For PL = PLmax , the maximum
active power is transferred. This maximum power is termed as the theoretical transfer
limit, and the corresponding voltage is termed as the critical voltage (for this case
ULcr = 0.7382 (pu)). For PL < PLmax , any value of power can be transferred at two
different values of UL . For instance, PL = 1 (pu) is transferred at UL1 = 1 (pu) (on
the blue line) and UL2 = 0.3 (pu) (on the red line). Thus, to transfer this amount of
active power at UL2 , the current through the transmission system will be about 3.3
times larger than the current at UL1 . Since real transmission systems are not lossless,
with the current at UL2 the active and reactive losses in the transmission system will
significantly be higher. Therefore, the upper solution (i.e. UL1 ) corresponds to normal
operating condition, and it is practically considered as stable solution.

1.2

0.8
UL

0.6

0.4

0.2

0
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
PL

Figure 7.2. U-P curve (or nose curve).

Figure 7.3 shows the family of U-P curves for different power factors. The black curve
corresponds to tan(ϕ) = 0.25 that is lagging power factor (QL > 0), the blue curve
corresponds to tan(ϕ) = 0 that is unity power factor (QL = 0), and the red curve
corresponds to tan(ϕ) = −0.25 that is leading power factor (QL < 0). Leading power
126

factor is obtained by shunt compensation. As shown in the figure, the load power factor
has a significant influence on the U-P curve, and thereby on the voltage stability. As
compared with the unity power factor, with lagging power factor PLmax is lower, and
UL (the upper solution) declines faster. However, with leading power factor PLmax is
higher and the voltage profile of the upper solution is flatter. The conclusion is that
(due to the inductive nature of the transmission system) for being able to transfer
more active power to a load bus (i.e. the load center) and to keep the voltage of the
load bus close to its nominal (or rated) value, the reactive power demand should be
compensated at the load bus. Note that in the steady-state, typical limits are ±10%
of the rated voltage that is ∆UL = ±0.1 (pu) for a rated voltage of 1 (pu).

1.2

QL< 0
0.8
QL= 0
QL> 0
UL

0.6

0.4

0.2

0
0 0.5 1 1.5 2 2.5
PL

Figure 7.3. U-P curves with different power factors.

Another method to analyze voltage stability is based on the so called Q-U curves as
shown in Figure 7.4.
The curves are obtained by a series of power flow calculations as follows [20]:

• Let a fictitious synchronous condenser (i.e. a generator with Pg = 0) without

reactive power limits be placed at the load bus to make it as a PU-bus type. The
specified active power at this bus is PSk = 0 − PL .
• Run power flow calculation for a series of specified voltages from ULmax to ULmin .
• For each specified voltage UL calculate the generated reactive power Qg .
• Plot Qg versus the specified voltages as shown in Figure 7.4

Qg < 0 indicates that the generator absorbs (or consumes) reactive power, and Qg > 0
indicates that the generator injects (or produces) reactive power.
Figure 7.4 shows the Q-U curves for PL = 1 and PL = PLmax (pu) with tan(ϕ) = 0. The
intersection between the Q-U curve and Qg = 0 gives the operating point (or solution)
 127

PL=PLmax

PL=1

Qg 0

0.3 U =U 1
UL L Lcr

Figure 7.4. Q-U curves with different active power loading (tan(ϕ) = 0).

of the system (why?). For PL = 1, there are two solutions which are indeed the same
solutions in Figure 7.2. According to the voltage stability criterion (7.1), UL = 1 (pu)
d Qg
is the stable operating point since > 0 at this point. The critical operating point
d UL
d Qg
is reached when = 0. For PL = PLmax , there is only one operating point (i.e. the
d UL
d Qg
intersection between the Q-U curve and Qg = 0) at which = 0. The voltage at
d UL
this point is the critical voltage shown in Figure 7.2.
Since voltage security is strongly coupled to reactive power, the Q-U curve may be a
powerful tool to measure reactive power margin at a bus of interest.
Figure 7.5 shows the Q-U and P-U curves for PL = 0.8 (tan(ϕ) = 0.25) with different
Xeq . For η = 4 there are two solutions (or operating points), with the stable operating

Operating point
↓
0.95
P =0.8, η=2
L

PL=0.8, η=4

Q2 ↑ η=2 η=4
op
UL
Qg

0
Q1

U ULcr 0.8
Lcr
η=4 η=2
PL

Figure 7.5. Q-U and P-U curves with different Xeq (tan(ϕ) = 0.25).
128

d Qg
point to the right of = 0 (indicated by “op” in the figure). However, when two
d UL
lines are disconnected (i.e. η = 2) there is no operating point in the system that is the
system is unstable. It is assumed that PL is constant and QL = 0.25 PL .
Q1 and Q2 values shown in Figure 7.5 are reactive power margins with respect to the
corresponding transfer power limit PLmax , or ULcr . Since Q1 < 0, it corresponds to
the maximum amount of more reactive load consumption without losing an operating
point that is if QL > PL tan(ϕ) + |Q1 |, then there is no operating point in the system.
However, since Q2 > 0 in case η = 2, it corresponds to the minimum requirement of
reactive power injection (or compensation) at the load bus to have an operating point.
Note that Q2 corresponds to a constant reactive power injection independent of the
load bus voltage. If compensation at the load bus is provided by a shunt capacitor
Qc = B UL2 , the minimum requirement of reactive power compensation for having an
operating point is the distance between Qg = 0 and a point where Qc is tangent to the
Q-U curve.

Example 7.1 Consider the system shown in Figure 7.1. With η = 4, tan(ϕ) = 0.25,
PL = 0.8 (pu) and QL = PL tan(ϕ) (pu) the voltage magnitude of the load bus is
UL = 0.95 (pu). Due to a disturbance two lines in the transmission system have been
disconnected (i.e. η = 2). Assume that a shunt capacitor is available at the load bus.

a) Determine the minimum reactive power compensation Qcmin such that the post-
disturbance system has an operating point.

b) Determine the required reactive power reserve (i.e. Qcres = Qc − Qcmin ) such that
the voltage magnitude of the load bus is restored to the pre-disturbance value, i.e.
UL = 0.95 (pu).

Figure 7.6 shows the Q-U characteristic of the post-disturbance system.

0.35

0.3

0.25 op2
↑

0.2
Qcres
Qg

0.15
op1
0.1 Qc2
↑

0.05 Qcmin
Qc1
0

0.5 0.6 0.7 0.8 0.9 1

UL

Figure 7.6. Q-U curve of the Example 7.1 (η = 2).

 129

a) At the operating point “op1 ”, Qc1 = Bmin UL2 is tangent to the Q-U curve. The
distance from “op1 ” to Qg = 0 gives the minimum reactive power compensation
Qcmin = Qcop1 = Bmin UL2op1 ≈ 0.11 (pu) which can be read from the figure.

b) The operating point ”op2 ” corresponds to the desired voltage (i.e. UL = 0.95 (pu))
at the load bus. The intersection between Qc2 = (Bmin + ∆B) UL2 and ”op2 ” gives
the required reactive power compensation to obtain the desired voltage. From the
figure it is easy to find that Qcreq = Qcop2 = (Bmin + ∆B) UL2op2 ≈ 0.26 (pu). Thus,
Qcres = Qcop2 − Qcop1 .

7.2 Voltage instability mechanisms and prevention

Voltage instability occurs in a time interval ranging from a few seconds to tens of min-
utes, and is classified into slow and fast instabilities. The loads characteristics and the
voltage control actions are usually the main reason for this instability. A large amount
of the loads have voltage-dependent characteristics with a tendency to restore the power
demands, for instance induction motor slip adjustment, and thermostatically-controlled
heating loads. The power demand (or load) restoration may also be performed by Load
Tap Changers (LTC) which have also significant impact on voltage instability. Voltage
instability may also be experienced at the terminals of line commutated High Voltage
Direct Current (HVDC) links connected to weak power systems. The converters of this
kind of HVDC consume reactive power of 50-60% of the dc power.

7.2.1 Fast voltage instability

The fast voltage instability takes place just a few seconds or less after a disturbance.
The main reason for this instability is the fast load-restoring actions by components
such as induction motors, line commutated HVDC, and electronically-controlled loads.
After a large disturbance (such as short-circuit or tripping of a long transmission line),
induction motor responds rapidly to match the mechanical torque due to its dynamic
2H ṡ = Tm − Te (UL , s), where (all quantities are expressed in pu) s is the motor slip
(note that the motor speed is given by ω = 1 − s), and UL is the voltage at the bus
where the motor is located. If there is no intersection between the mechanical torque
(which is assumed constant) and the electrical torque after the disturbance, the system
loses a post-disturbance equilibrium point which results in stalling of the motor (i.e.
the motor decelerates to a complete stop, and s = 1). The motor stalling implies
a higher reactive power consumption which causes the voltage to collapse. Another
situation may be that the motor slip is greater than the unstable slip su when a fault
is cleared. At that point Te (which is a function of UL and s) is less than Tm . Thus
due to the motor dynamic, s increases and approaches s = 1 at which the motor stalls.
Assume that the load in Figure 7.1 is an induction motor, see Figure 5.2. In this study,
Ėq′ and Ėd′ are set to zero in equation (5.6). A shunt capacitor is installed at the load
bus to compensate the reactive power consumption of the motor such that the voltage
130

at the load bus is 1 (pu). A three-phase fault occurs at the load bus, and it is cleared
at the clearing time tc . Figure 7.7 shows the system response for tc = tc1 , and Figure
7.8 shows the system response for tc = tc2 > tc1 . In the figures, the post-fault electrical
torques are shown.
At the clearing time tc = tc1 , the initial value of the post-fault electrical torque (in-
dicated in the figure with “o”) is greater than Tm , and the motor slip is less than su .
Thus, s decreases and returns to the equilibrium point at which Te = Tm . Also, PL and
QL (which are functions of UL and s) are restored. The reactive power consumption is
also proportional to s and PL that is QL ∼ sPL .

P P
L L
1
UL

UL
Te

QL QL

T
m

Time Motor slip

Figure 7.7. Dynamic response of the system with tc = tc1 .

Q
L
T
m
1

UL

T
P e
L

Time Motor slip

Figure 7.8. Dynamic response of the system with tc = tc2 > tc1 .

At the clearing time tc = tc2 > tc1 , the electrical torque (indicated in the figure with
“o”) is less than Tm , and the motor slip is greater than su . Thus, s increases and so
does QL which causes the voltage to drop.
In a similar manner (especially due to a reduction in the voltage at the ac terminal),
the power control of HVDC responds rapidly to restore the demanded active power
which results in more reactive consumption in converters, and causing further voltage
reduction at the ac terminal. This process leads to progressive fall of voltage.
 131

7.2.2 Slow voltage instability

The time frame of this instability may extend from tens of seconds to several min-
utes. Equipments such as LTC, thermostatically-controlled loads and generator reac-
tive power limiters (which have slow dynamics) are mostly involved in the slow voltage
instability.
The mechanism of the slow voltage instability is similar to that for the fast voltage
instability, however within a longer time frame. This mechanism will be presented
below by using a Load Tap Changer (LTC).
Assume that the load in Figure 7.1 is connected to the transmission system via an LTC
as shown in Figure 7.9
UN∠0 parallel lines
UL U

SL = PL + jQL
n :1

Figure 7.9. Load tap changer in the SLIB system.

An LTC is a transformer with variable turns-ratio (or tap changer, i.e. n). The function
of the LTC is to (automatically) control the voltage at the load (i.e. U) by changing
n. Normally, the tap is on the high voltage side where it is easier to change the tap
since the current on this side is lower. The relation between UL and U is given by
UL
U= (7.6)
n

The dynamic of an LTC is usually described by a discrete tap changing logic. However,
in our analysis a continuous tap changing model is applied as follows
1 1 UL
ṅ = (U − U0 ) = ( − U0 ), nmin ≤ n ≤ nmax (7.7)
T T n
where T is a time constant which is typically some ten seconds, and U0 is the reference
voltage.
In Figure 7.9, we assume an ideal LTC (i.e. the transformer impedance is omitted).
Typical tap changer range is ±10%, but in this study we assume that the LTC does
not hit its limit. Also, the load is represented by an impedance with XL = QL = 0
that is  2
U2 2 UL
S̄L = PL = = GL U = GL (7.8)
RL n

Consider now the SLIB system shown in Figure 7.9. The pre-disturbance system data
are as follows
2
η = 4 , UN = 1.044 , PL = UL = U = n = GL = 1 , QLoss = Xeq ILine = 0.3 (pu)
132

Figure 7.10 shows the system response after disconnection of a line in the transmission
system (i.e. η = 3). The diagram on the left hand side shows variations of PL , UL , U
and n versus time.
1.01
1
η=4
1
U
η=3
0.99 P (n )
L 1

0.98
PL(n2)

UL
0.97
n

0.96
UL
0.95 P
L

0.94

0.93 1
0 100 200 300 400 500
Time (s) P
L

Figure 7.10. Dynamic response of the system with η = 3.

When the line is disconnected, PL , UL , and U decrease. Thus, the LTC starts to
restore the load voltage U which results in decreasing of n. When n decreases, the
load voltage U increases, and so does PL since it is voltage dependent. An increase
of PL implies higher current trough the transmission system which results in higher
reactive power loss in the transmission system (i.e. QLoss ), and causing further voltage
reduction (i.e. UL ). This situation continues until the load is restored that is PL =
PL0 = 1 and U = U0 = 1. The diagram on the right hand side shows the U-P
curves of the pre-disturbance and post-disturbance systems, as seen from the high
voltage side of the LTC. The red curves are the pre-disturbance and post-disturbance
systems characteristics. The vertical dashed line is the steady-state (or long-term) load
characteristic. In the steady-state, the load is considered as constant power, and it is
restored to its set value PL0 = 1. The intersection between the system characteristic
and the steady-state load characteristic gives the equilibrium point of the system. The
blue curve is the dynamic characteristic of the load (i.e. PL (n)) as a function of the
state variable n, also as a function of UL (see equation (7.8)). The equilibrium point
of the pre-disturbance system is indicated by “o” at which PL (n) = 1 with n = n1 = 1
and UL = 1. When the line is disconnected (i.e. η = 3) , the operating point moves
to the point “*” on the post-disturbance system characteristic. Due to the dynamic of
the LTC, PL (n) will be varied until it reaches the post-disturbance equilibrium point
indicated by “⊠” (i.e. the intersection point between PL (n2 ) and the U-P curve of
η = 3). Note that n2 < n1 .
Now assume that due to a protection action in the transmission system, another line
is disconnected (i.e. η = 2) just eight minutes after the first disconnection. The post-
disturbance system (i.e. η = 2) characteristic is shown in Figure 7.11. Since there is no
intersection between the post-disturbance system characteristic and the steady-state
load characteristic (i.e. the dashed line), this post-fault system has no equilibrium
point. When the second line is disconnected (i.e. η = 2) , the operating point moves
to the point “⋄” on the post-disturbance system characteristic. At that point U is less
 133

than U0 . Thus, the LTC starts to restore the load voltage U which results in decreasing
of n. When n decreases, the load voltage U increases, and so does PL since it is voltage
dependent. An increase of PL implies higher current trough the transmission system
which results in higher reactive power loss in the transmission system, and causing
further voltage reduction (i.e. UL ). Since there is no equilibrium point in the system,
this situation continues slowly due to the slow dynamic of the LTC. However, when
PL (n) crosses the critical voltage indicated by “⊛”, and enters into the lower side of
the U-P curve the load restoration by the LTC fails which results in decreasing of U
and PL (i.e. when n < n3 ). This situation leads to voltage instability (or collapse) in
the system as shown in Figure 7.11.

PL(n2)
1

1
U
0.8 P (n )
L 1
n UL

0.6 UL
η=2 η=3 η=4
PL(n3)
QLoss
0.4

P (n )
L 4
0.2

0 1
0 200 400 600 800 1000
Time (s) P
L

Figure 7.11. Dynamic response of the system with η = 2.

It should be noted that if the LTC had a limited range of action (i.e. nmin ≤ n ≤ nmax ),
the LTC could hit its limit before reaching the critical point “⊛” as shown in Figure
7.12. Since the LTC hits its limit there is no dynamic in the system (i.e. ṅ = 0), and
the load will not be restored. Thus, the point indicated by “⊞” (i.e. the intersection
point between PL (n = 0.9) and the U-P curve of η = 2) is the equilibrium point of the
post-disturbance system.

1 PL(n=1)
PL(n=0.9)
0.98 1

0.96

0.94 U
0.92
UL

n
0.9
η=2 η=3 η=4
0.88 PL

0.86
UL
0.84

0.82

0.8 1
0 500 1000
Time (s) P
L

Figure 7.12. Dynamic response of the system with nmin = 0.9.

134

In case of having a limited range of LTC action we should, however, consider the
thermostatically-controlled action of the load for restoring the load. The simplest
dynamic of the thermostatically-controlled load is given by
 2
1 n
ĠL = (PL0 − PL ) (7.9)
TG UL
where TG is a time constant which is usually in order of several minutes, PL0 is the
reference active power, and PL is given by equation (7.8).
Figure 7.13 shows the dynamic response of the system when the LTC has a limited
range of action, and the load self-restoration action (i.e. equation (7.9)) is considered.

P (G ,n )
1.2 L L1 1
G P (G ,n )
L
1 L L3 min

1
n P (G ,n )
L L4 min
0.8 U
UL

U
0.6 L
η=2 η=3 η=4
QLoss
0.4
PL(GL5,nmin)

0.2

0 1
0 500 1500
Time (s) P
L

Figure 7.13. Dynamic response of the system with the load self-restoration action.

At point “⊞” (i.e. the intersection point between PL (GL3 , nmin ) and the U-P curve
of η = 2), the LTC hits its limit, and the only dynamic in the system is given by
equation (7.9) which continues to restore the load by increasing GL which results in
higher reactive power loss in the transmission system (i.e. QLoss ), and causing further
voltage reduction (i.e. UL ). Since there is no equilibrium point in the system, this
situation continues slowly due to the slow dynamic of the thermostatically-controlled
action, and leads to voltage collapse.
Obviously, the dynamic response of the system shown in Figure 7.13 is similar to
that shown in Figure 7.11 where we took the unrealistic assumption that having an
unlimited range of LTC action. However, by that assumption the dynamic of the
thermostatically-controlled action of the load was indirectly included in equation (7.7).

7.3 Prevention of voltage instability

In [6]-[20], different preventive and corrective actions to counteract voltage instabil-
ity are comprehensively described. However, the two essential (or general) corrective
actions to prevent a voltage instability (or collapse) are the following:
 135

Reactive power support

Naturally, the reactive power support should take place close to the load which leads
to a decrease of reactive power losses in the transmission system, and an increase of
the maximum deliverable active power.
Consider the case shown in Figure 7.11, by applying reactive power support a new
stable equilibrium point is restored, and the system will be stable if this reactive power
support is performed fast enough as shown in Figure 7.14.

P 1
L
U

n
P (n )
L 1

UL
UL

0.65 1
0 500 750 2000
Time (s) PL

Figure 7.14. Dynamic response of the system with fast enough reactive power support.

The diagram on the right hand side shows the U-P curves of the post-disturbance
system with η = 2. The red curve is the same curve shown in Figure 7.11 (with η = 2)
for which the system has no equilibrium point. The green curve shows the system U-P
characteristic when a shunt capacitor is switched in the transmission system close to
the load. By this action two new equilibrium points (indicated by “o”) are restored
in the system. The upper one is a stable equilibrium point, and the lower one is
unstable. The point “*” is the initial point of the post-disturbance system when the
second line was tripped (i.e η = 2) at t = 500 (s). The intersection point (indicated
by “”) between PL (n1 ) and the dotted U-P curve corresponds to the operating point
just before the capacitor switching at t = 750 (s). When the capacitor is switched, the
operating point jumps to the new one indicated by “⊞” at which U > U0 since at this
point PL > 1 and PL = GU 2 = 1 U 2 . Note that in this example U0 = 1. Thus, based
on the LTC dynamic (see equation (7.7)) the tap n increases, and so does UL . Since
PL (see equation (7.8)) is a function of n and UL , it takes the direction indicated by
the arrow and tends to the stable equilibrium point.
Figure 7.15 shows the dynamic response of the system when the shunt capacitor is
switched at t = 800 (s). When the capacitor is switched, the operating point jumps
from the point indicated by “” to the one indicated by “⊞” which is the initial point
of the post-disturbance system with reactive power support. At this point U < U0
since PL < 1. Thus, based on the LTC dynamic the tap n decreases, and so does
UL . Therefore, PL takes the direction indicated by the arrow. This situation leads to
voltage collapse.
136

1 1
U

PL
n
UL

UL
P (n )
L 2

0 1
0 500 870 1100
Time (s) PL

Figure 7.15. Dynamic response of the system with too slow reactive power support.

In this example for any initial point (of the post-disturbance system with reactive power
support) under the unstable equilibrium point the system will be unstable. Thus, the
switching time, and also the size of the shunt capacitor should be carefully selected so
that the initial point does not lie under the unstable equilibrium point.
For the case shown in Figure 7.8, a fast-acting of reactive power support may prevent
the motor slip to reach the unstable slip su . Power electronic based devices (such as
SVCs and STATCOMs which will be presented in EG2120 FACTS and HVDC in power
systems), generators, and synchronous condensers can be mentioned as fast reactive
power support devices. A synchronous condenser is a synchronous generator that only
produces or consumes reactive power. Such a synchronous condenser is installed in the
Swedish island Gotland due to the (line commutated) HVDC link between the weak
network of the island and the power system of the mainland.
Reactive power support can also be performed by switching series reactances, auto-
matic line re-closing following the fault clearing, switching off shunt inductors in the
transmission system. However, these actions should be fast enough to lead the system
to the stable equilibrium point.
Load reduction
This action can be performed by

• blocking the LTC tap changing, see PL in Figure 7.12. Note that if other load
restoration exists, this action may only delay the process of a voltage collapse,
see Figure 7.13,

• load shedding (i.e. by reducing GL , see Figure 7.16, the figure on the left hand
side),

• reducing the LTC set point (i.e. U0 in equation (7.7), see Figure 7.16, the figure
on the right hand side).
 137

1.05
1.05

U 1
1
U
0.95 0.95
PL
0.9 U 0.9
L

n 0.85
0.85
P
L
0.8 0.8 n

0.75 0.75 UL

0.7 0.7
0 500 700 2000 0 500 700 2000
Time (s) Time (s)

Figure 7.16. Dynamic response of the system by reducing GL with 20% (on
the left) and by reducing U0 with 5% (on the right) at t = 700 (s).

Note however that UL in the figure on the right hand side is still very low. This low
voltage may lead to other protection actions in the transmission system.
Example 7.2 Consider again the three generator system in Figure 5.6.
In this example, each generator is represented based on its steady-state model shown in
Figure 3.8.
a) In a similar way as described in Example 4.3, as seen from BUS 4 find an equivalent
model of the system with ri = 0 and ri = 0.1 xi , respectively, for each line.
Next, normalize the equivalent model with respect to the base values UT h , Ssc and ZT h .
b) Plot the un -pn curve with QL4 = 0, and with Ri = 0 and Ri = 0.1 Xi , respectively.
Now, let Ri = 0.
c) Plot the un -pn curve with tan(ϕ) = −0.25, tan(ϕ) = 0 and tan(ϕ) = 0.25, respec-
tively.
d) Plot the qn -un curve with pn = pmax
n and pn = pmax
n /3, respectively.
e) Using the unnormalized values (i.e. PL4 , QL4 and UL4 ), plot the stability region in
which for any active and reactive loads there is a real solution for the voltage. Plot
also the security region in which for any active and reactive loads the voltage (UL4 ) is
greater than 0.9. In the figure, also indicate the point (PL40 , QL40 ) from the load flow
calculation.
(See the MATLAB files for ”Project D3” in Canvas.)
a) In a similar way as described in Example 4.3, the following can be obtained.
With Ri = 0
UT2 h
ŪT h = 1.1663∠ − 1.8254 , Z̄T h = 0.0000 + j 0.3684 , Ssc = = 3.6919
ZT h
With Ri = 0.1 Xi
UT2 h
ŪT h = 1.1700∠ − 3.3630 , Z̄T h = 0.0102 + j 0.3684 , Ssc = = 3.7141
ZT h
138

Figure 7.17 shows the equivalent model of the test system as seen from BUS 4.

U Th ÐqTh U 4 Ðq 4
ZTh
PL 4 + jQL 4

Figure 7.17. The equivalent model of the test system as seen from BUS 4.

b) Let,
RT h UT h
α = θ4 − θT h − arctan( ) , UT hn = =1
XT h UT h
PL4 QL4 U4
pn = , qn = , un =
Ssc Ssc UT h
RT h XT h p
rn = , xn = = 1 − rn2
ZT h ZT h

The mismatch equations at BUS 4 give the following normalized equations.

RT h 2 UT h
U + UL4 sin(α) + PL4 = 0 ⇒ rn u2n + un sin(α) + pn = 0
ZT2 h 4 ZT h
(7.10)
XT h 2 UT h
U − UL4 cos(α) + QL4 = 0 ⇒ xn u2n − un cos(α) + qn = 0
ZT2 h 4 ZT h

Furthermore,
√
p2n +qn2 ⇒ u2n = a ± ⇒ b
p (7.11)
u2n =(0.5 − (rn pn + xn qn )) ± (0.5 − (rn pn + xn qn ))2 − (p2n + qn2 )

A necessary condition to have a real un is that b ≥ 0 which implies that

p2n + qn2 ≤ (0.5 − (rn pn + xn qn ))2 (7.12)

Thus,
r
0.5 0.5
qn =0 ⇒ pmax
n = ⇒ uncr =
1 + rn 1 + rn
r (7.13)
0.5 0.5
pn =0 ⇒ qnmax = ⇒ uncr =
1 + xn 1 + xn

Figure 7.18 shows the un -pn curves with Ri = 0 (in blue) and Ri = 0.1 Xi (in red).
Note that qn = 0 in the both cases.
c-d) Figure 7.19 shows the un -pn curves with tan(ϕ) = 0.25 (in black), tan(ϕ) = 0 (in
blue) and tan(ϕ) = −0.25 (in red), and also the qn -un curves with pn = pmax
n (in red)
max
and pn = pn /3 (in red).
 139

0.8

0.6

un 0.4

0.2

0
0 0.1 0.2 0.3 0.4 0.5
pn

Figure 7.18. The un -pn curves with Ri = 0 (in blue) and Ri = 0.1 Xi (in red).

0.8125

0.7071
0.6344
un

qn

-0.8204
0
0 0.3904 0.5 0.6404 0.1691 0.7071 0.9856
pn un

Figure 7.19. The un -pn curves with different power factors, and the qn -un
curves with pn = pmax
n and pn = pmax
n /3.

e) Figure 7.20 shows the stability region (in cyan) and the security region (in yellow)
for PL and QL without reactive power compensation. The stability region indicates
for which PL and QL there is a real solution for UL . However, due to practical issues
the voltage in the steady state cannot be less than a specified minimum value (0.9 in
this example). The security region indicates for which PL and QL the voltage UL is
still over the minimum value. It also gives some information to the system operator,
for a given PL and QL how much reactive power should be compensated to have the
operating point in the security region.

Example 7.3 Consider the equivalent model shown in Figure 7.21. As seen from
BUS 4, the three-machine power system is represented by its an equivalent model (i.e.
ŪT h and Z̄T h1 ). In this example we also add an equivalent model of the distribution
system as seen from an arbitrary bus to which a load and a generating facility based
on wind and solar power are connected. In contrast to a transmission system, RT h in
140

0.9230

0.6505
QL

QL
0 0
0 1.8460 0 1.8116
PL PL

Figure 7.20. the stability region (in cyan) and the security region (in yellow)
for PL and QL . The point (PL40 , QL40 ) is indicated by ”*”.

a distribution system has a significant value. For the three-machine power system, let
Ri = 0. For the distribution network, let Z̄T h2 = 0.3986(0.8 + j). Normalize the system
components with respect to the base values UT h , Ssc and ZT h = |Z̄T h1 + Z̄T h2 |.

U Th U4 U res
ZTh1 ZTh 2 Pg

n :1
PL + jQL

Figure 7.21. The equivalent model of the system.

• Let the generating facility be disconnected but not the load with qLn = 0. Plot the
variations of the un of the arbitrary bus and BUS 4 versus pLn .

• Let the generating facility be connected to the arbitrary bus. Assume that this
facility provides the given local load, and its surplus Pg is transferred to the power
system which is represented by the Thévenin equivalent ŪT h and Z̄T h1 . Plot the
variations of the un of the arbitrary bus and BUS 4 versus pgn .

As discussed in Chapter 5 for power system study the composite load model is used,
i.e. as seen from BUS4, the entire distribution system and its loads are modeled as
a composite load. The aim of this example is to present one of the challenges with
the Distribution Generation (DG) in a distribution system. Figure 7.21 shows a very
simple representation of a distribution system. However, it gives us an insight into the
challenges with DG, especially to keep the voltage profile within the acceptable limits.
Figure 7.22 shows the un -pLn curves (on the left hand side) and the un -pgn curves (on
the right hand side). In the purely consumption case at the arbitrary bus, we get the
well-known U − P curves pattern, i.e. a reduction of the voltages at these two buses
when more power is consumed. The distribution systems have been constructed and
developed to transfer power from the transmission system to the end-users. The applied
 141

1.1 1.1
1 1
0.9 0.9
un

un
Consumption:
Normolized U in blue
res
Normolized U in red
4

Generation:
Normolized U in blue
res
Normolized U in red
4
0 0
0 0.3613 0 0.8115
pn pgn

Figure 7.22. The un -pLn curves (on the left hand side) and the un -pgn curves
(on the right hand side).

means, such as LTC and/or reactive power compensations, has been developed as
technical solutions for this construction to keep the voltage profile within the acceptable
limits. Since the voltages of these two buses decrease with more consumption, the
mentioned means can be applied to have the voltages within the acceptable limits.
However, in the generation case at the arbitrary bus, a different pattern is obtained.
The voltage at BUS 4 decreases, but the voltage at the arbitrary bus increases when
more power is generated and transferred to the power system. If a means is applied
to increase the voltage at BUS 4, it may result in pushing the voltage at the arbitrary
bus out of the acceptable limits, or a means is applied to decrease the voltage at the
arbitrary bus, it may result in pushing the voltage at BUS 4 out of the acceptable
limits. It should be noted that in a distribution system we may have thousands of
buses similar to the arbitrary bus, and today’s means may not be effective for keeping
the voltage profile within the acceptable limits when more DG will be integrated in
the distribution systems. This challenge requires more research to develop new and
innovative voltage control methods.
142
Chapter 8
Frequency control
(Part of this chapter follows in large the discussion in [21].)
Power system frequency is an appropriate measure on the active power balance in a
power system. The frequency is constant when the same amount of electrical power
is produced as consumed by the loads, including system losses. If this is not the case
frequency changes will occur. The frequency is reduced when a load increase or a loss
of production is not compensated by a corresponding increase of the turbine power
of the connected generators. The power deficit decelerates the generator rotors and
consequently the frequency is reduced. Too large reductions of the frequency can trigger
protection system which may result in system separation, loss of load and costumer
outages, since many equipments in a power system, e.g. power supply systems, do not
tolerate too low frequencies.
Since the stored kinetic energy in the system is relatively small (the inertia constant of
a typical generator is about 5 (s)), the electrical energy must be produced at the same
moment as it is consumed by the loads. As the power consumption varies and also due
to disturbances such as outage of generation and load, the active power production
must be regulated accordingly to keep the frequency within the acceptable limits. In
accordance with the European Network of Transmission System Operators for Elec-
tricity (ENSTO-E) terminology [22]-[23], this frequency control is performed in three
processes, namely

• Frequency Containment Reserves (FCR),

• Frequency Restoration Reserves (FRR),
• Replacement Reserves (RR).

In this compendium, FCR and FRR will be discussed.

The objective of FCR is to stabilize the system frequency after a disturbance at a
steady-state value within the permissible maximum Steady-State Frequency Deviation
(SSFD). At the maximum SSFD, the FCR are fully activated. In this process, the
turbine control (governor) is automatically activated to adjust the active power gen-
eration based on frequency deviation. In order to share the control between different
generators involved in FCR, a permanent frequency droop is used. In some power
systems, the term primary control is used for this process.
In the Northern Europe (NE) synchronous area (comprises Sweden, Norway, Finland
and Eastern Denmark), FCR are above all located in hydro units, and are divided up
into

• FCR for Normal operation (FCR-N) which is at least 600 (MW) at 50 (Hz),
and will be fully activated when the permissible maximum SSFD (± 0.1 (Hz))

143
144

is reached. The full activation time of FCR-N is within 180 (s), (about 63% is
activated within 60 (s) and a full activation within 180 (s)).
• FCR for Disturbances (FCR-D) which is about 1000 (MW), and is used in
case of contingencies (such as the trip of the largest generating unit) so that
post-disturbance steady-state frequency does not become less than 49.5 (Hz), i.e.
the permissible maximum SSFD is −0.5 (Hz). This action is supposed to start
at 49.9 (Hz) and will be fully activated at 49.5 (Hz). The full activation time of
FCR-D is within 30 (s), (about 50% is activated within 5 (s) and a full activation
within 30 (s)).

Note however that when the system frequency is changed the power demand of certain
loads is also changed, specially for motors, in such a way that a frequency increase
leads to increased power consumption and a frequency decrease gives lower power
consumption. This load frequency dependency thereby stabilizes the frequency.
The objective of FRR is to restore the system frequency to its nominal (or scheduled)
value, and to replace the activated FCR. This process may be performed manually
and/or automatically by applying a supplementary control loop (an integrator). Since
year 2013, an automatic FRR has been introduced in the NE synchronous area to
improve frequency quality performance. However, manual regulation is predominant
FRR [22]. For the both cases the frequency is used as the input signal. From the
actual frequency deviation and the knowledge about the system frequency droop it
is straightforward for the system operators to compute how much additional power
is needed to restore the system frequency and the used FCR. In the NE synchronous
area, the full activation time of the automatic FRR is about 2 minutes, and the full
activation time of the manual FRR is about 15 minutes. In some power systems, the
term secondary control is used for this process.
Another important parameter for frequency control is Instantaneous Frequency Devi-
ation (IFD) on which the system inertia has a significant impact. The system inertia
should be sufficiently large in such a way that the instantaneous frequency does not
exceed a permissable maximum IFD. For instance in the NE synchronous area, a fre-
quency between 49 and 47 (Hz) will trigger the protection systems to under-frequency
load-shedding and system separation, and also a frequency less than 47.5 (Hz) will
trigger the protection systems of large steam turbines to be disconnected.

8.1 System model

Consider the multi-machine power system shown in Figure 5.3. The swing equation of
the k-th generator is given by (all variables including ω are expressed in (pu))
Mk′ ω̇k = (Pmk − Pek )

where,
Sngk
Mk′ = 2 Hk
Sbase
 145

Transforming the swing equations to the COI reference frame, the following is obtained.
ng ng
1 X ′ 1 X
ω̇COI = ′ M ω̇k = ′ (Pmk − Pek ) (8.1)
MT k=1 k MT k=1

where ng
X
MT′ = Mk′
k=1

By considering a small deviation (denoted by ∆) from initial values (denoted by 0), we

may write
ng ng ng
X X X
Pm = Pmk = Pm0k + ∆Pmk = Pm0 + ∆Pm
k=1 k=1 k=1
ng ng ng
X X X
Pe = Pek = Pe0k + ∆Pek = Pe0 + ∆Pe
k=1 k=1 k=1

Thus, equation (8.1) may be rewritten as

MT′ ∆ω̇COI = (Pm0 − Pe0 ) + (∆Pm − ∆Pe ) = ∆Pm − ∆Pe (8.2)

| {z }
=0

Assume that the overall load is modeled as a composite load (see equation (5.4)) which
has both frequency-dependent and non-frequency-dependent characteristics. Then,
∆Pe may be expressed as

∆Pe = ∆PL + DCOI ∆ωCOI (8.3)

where, ∆PL denotes the non-frequency-sensitive load change, DCOI ∆ωCOI denotes the
frequency-sensitive load change, and DCOI is a small positive damping constant.
If the system losses are included in Pe , the term ∆Ploss is then added to (8.3). However,
since ∆Ploss is much smaller than the sum of the two other terms, it can be neglected
that is we may set ∆Ploss ≈ 0.
Let ∆ωCOI , DCOI and MT′ be henceforth denoted by ∆ω, D and M, respectively,
where M is referred to as the system inertia and ∆ω is termed the system (or average)
frequency deviation, since ∆ω = ∆f in (pu).
Equation (8.2) can now be rewritten as

M ∆ω̇ = ∆Pm − ∆PL − D ∆ω (8.4)

whose block diagram representation is shown in Figure 8.1. From equation (8.4), we
see that the steady-state frequency deviation (SSFD) for a load change is given by
∆PL ∆PL
∆f = ∆ω = − (pu) ⇒ ∆f = − fs (Hz) (8.5)
D D
146

∆Pm
+
1 ∆ω
¦
- Ms + D
∆PL

Figure 8.1. Block diagram representation of system (8.4).

8.2 Turbine and turbine governor

The objective of a turbine governing system installed in a generating unit is to produce
a desired power which is partly determined by the set value for the produced power
and partly by a contribution originating from the frequency control. In this context,
the latter is of interest. Figure 8.2 shows a schematic diagram of the mechanical part
of the generating unit k, where ∆Pmk is the contribution of the k-th generator to ∆Pm
in Figure 8.1, ∆ω is the system frequency deviation, and ∆Y represents the change
of the gate position of a hydro turbine, or the change of the valve position of a steam
turbine. The block indicated with “turbine” represents the dynamic of the turbine, and
the block indicated with “governor” represents the dynamic of the governing system.

∆ω ∆Yk ∆Pmk
Governor Turbine

Figure 8.2. Schematic diagram of the mechanical part of a generating unit.

In this compendium, only the hydro turbine and turbine governor are discussed. For
steam turbines and governing systems, the readers are referred to [6] and associated
references therein.

8.2.1 Hydro turbine model

Figure 8.3 depicts a hydro turbine with penstock and hydro reservoir, and defines the
notation that will be used from now on. Bernoulli’s equation for a trajectory between
the points P1 and P2 can be written as
Z P2 Z P2
∂v 1 2 2 1
· dr + (v2 − v1 ) + Ω2 − Ω1 + dp = 0 (8.6)
P1 ∂t 2 P1 ρ

The following assumptions are usually made:

• v1 = 0, since the reservoir is large and the water level does not change during the
time scale that is of interest here.

• The water velocity is non–zero only in the penstock.

 147

• The water is incompressible, i.e. ρ does not change with water pressure.
• The water pressure is the same at P1 and P2 , i.e. p1 = p2 .

Further,
Ω2 − Ω1 = −gh (8.7)

P1

Length of
Penstock = L
h = Head

Area = A
Velocity = v

P2

Effective Area= a
Output Velocity = v out

Figure 8.3. Schematic drawing of hydro turbine with water paths.

The above assumptions together with equation (8.7) make it possible to rewrite equa-
tion (8.6), with vout = v2 and the length of the penstock L, as
dv 1 2
L + v − gh = 0 (8.8)
dt 2 out
where, v is the water velocity in the penstock.
Let a denote the effective opening of the penstock (determined by the opening of the
control gate of the turbine) and A denote the area of the penstock. Since the water is
assumed incompressible, the rate of the water flow is the same at P1 and P2 , i.e.
A
a vout = A v ⇒ vout = v (8.9)
a
Thus, equation (8.8) can be written as
 2
dv 1 1 A
= gh − v (8.10)
dt L 2L a
The maximum available power at the turbine is
1 3 1 A3 v 3
Pm = ρ a vout = ρ 2 (8.11)
2 2 a

To get the system into standard form,

a
x=v , u= , y = Pm (8.12)
A
148

are introduced. The system can now be written as

gh 1
ẋ = − x2
L 2Lu2 (8.13)
3
x
y =ρA 2
2u
whose block diagram is shown in Figure 8.4.
u ÷ x
x /u x3
y = Pm = ρ A
2u 2
Π
2
x /u 2 1 - x 1 x y
¦ Π
2L + s
gh
L

ρA
2

Figure 8.4. Block diagram representation of hydro turbine.

Equation (8.13) describes the dynamic of a nonlinear system of the form

ẋ =f (x, u)
(8.14)
y =ℏ(x, u)

To get an idea of the properties of this system, the nonlinear system is linearized around
its equilibrium point. In the steady-state ẋ = 0. For a given u0 , the equilibrium point
x0 and the initial value of y, i.e. y0 are obtained as follows
p
x0 =u0 2gh
ρAx30 (8.15)
y0 =
2u20

Linearization of (8.14) around its equilibrium gives

∆ẋ =fx ∆x + fu ∆u
(8.16)
∆y =ℏx ∆x + ℏu ∆u
or
1 2x20
∆ẋ = − 2x0 ∆x + ∆u
2Lu20 2Lu30
(8.17)
Ax2 Ax3
∆y = 3ρ 20 ∆x − 2ρ 30 ∆u
2u0 2u0
 149

which, using equations (8.15), can be rewritten as

√
2gh 2gh
∆ẋ = − ∆x + ∆u
u0 L u0 L
(8.18)
3y0 2y0
∆y = √ ∆x − ∆u
u0 2gh u0
√
Note that L/ 2gh has dimension of time, and from the above equations it is apparent
that this is the time it takes the water to flow through the penstock if a = A. That
time is denoted by T : p
T = L/ 2gh (8.19)

The Laplace transformation of (8.18) leads to

L/T
∆x = ∆u (8.20)
1 + su0 T

Based on (8.20), the relationship between the input ∆u and the output ∆y is then
given by
y0 1 − 2u0T s
∆y = · ∆u (8.21)
u0 1 + u0 T s
Note that u0 T = a0 T /A also has dimension of time and is denoted by Tw . Thus,
equation (8.21) can be rewritten as

y0 1 − 2Tw s
∆y = · ∆u (8.22)
u0 1 + Tw s

It is obvious that the transfer function in equation (8.22) is of non–minimum phase,

i.e. not all zeros are in the left half plane. In this case, one zero is in the right half plane.
That is evident from the step response to equation (8.22), depicted in Figure 8.5.

∆u

∆u 1
t(s)

0 5 10 15 20

∆y
y0
----- ∆u 1
u0 t(s)

y0 Tw = 5 s
- 2 ----- ∆u 1
u0

Figure 8.5. Variation of produced power, ∆y, after a step change in control gate.
150

The system has the (unpleasant) property to give a lower power just after the open-
ing of the control gate is increased before the desired increased power generation is
reached. The physical explanation is the lower pressure appearing after the control
gate is opened, so that the water in the penstock can be accelerated. When the wa-
ter has been accelerated, the generated power is increased as a consequence of the
increased flow. That property of water turbines places certain demands on the design
of the control system for the turbines.
Normalizing the input and output variables of equation (8.22), this equation can be
rewritten as
1 − 2Tw s
∆Pm = ∆Y (8.23)
1 + Tw s
where,
∆y ∆u
∆Pm = and ∆Y =
y0 u0

In this compendium, the transfer function (8.23) is used to represent the dynamic of a
hydro turbine. Thus, for the k-th generator we have (see Figure 8.2)

1 − 2Tw s
∆Pmk = ∆Yk (8.24)
1 + Tw s

8.2.2 Hydro turbine governor model

It is the task of the turbine governor to control the control gate such that the desired
power is produced by the generator in question. That power is partly determined by
the set value for the produced power and partly by a contribution originating from the
frequency control. In this context, the latter is of interest. A model of this controller is
given in Figure 8.6. The control servo is here represented simply by a time constant Tp .
The main servo is represented by an integrator with the time constant TG . The change
of the set value is indicated by ∆Pmset which is set to zero prior to a disturbance.
uopen
∆Ymax
∆ω - 1 1 1 ∆Y
¦
- 1 + Tp s TG s
∆Ymin
uclose

+ gtrTR s
¦
+ 1 + TR s

+
∆Pmset
g st ¦ -

Figure 8.6. Model of governors for hydro turbines.

Typical values for these parameters are given in Table 8.1.

 151

Parameter Typical values

TR 2.5 – 7.5 s
TG 0.2 – 0.4 s
Tp 0.03 – 0.06 s
gtr 0.2 – 1
gst 0.03 – 0.06
Table 8.1. Typical values for some parameters of the turbine controller for hydro power.

The controller has two feedback loops, a transient feedback loop and a static feedback
loop. The transient feedback loop has the gain gtr for high frequencies, and the static
feedback loop has the gain gst . Thus, the total feedback gain after a frequency change
is −(gtr + gst ). The transient feedback is needed since the water turbine is a non–
minimum phase system. If the transient feedback is left out or made too small, the
system can become unstable. The transient feedback causes the system to be slower;
the transient frequency deviations become considerably larger since the initial total
feedback can be about ten times larger than the static feedback.
Setting Tp = 0, and neglecting the limiters, the dynamic of the hydro governor of the
k-th generator may be (approximately) described by the following transfer function [5]

1 1 + TR s
∆Yk = − ∆ω (8.25)
gstk (1 + Tg1 s)(1 + Tg2 s)

where,
TR TG TG + TR (gstk + gtr )
Tg1 ≈ and Tg2 = (8.26)
TG + TR (gstk + gtr ) gstk
Usually Tg2 ≫ Tg1 , therefore it has been assumed Tg2 + Tg1 ≈ Tg2 .
Figure 8.7 shows the block diagram of the mechanical part of a hydraulic unit which
contributes to the control of the system frequency.

∆Pmkmax
∆ω + 1 + TR s 1 − 2Tw s ∆Pmk
¦ − Rk
(1 + Tg1s )(1 + Tg 2 s) 1 + Tw s
-
g stk
∆Pmkset

Figure 8.7. Block diagram of the mechanical part of a hydraulic unit.

In the steady-state (see equation (8.25)), the ratio between the frequency deviation
and the change in the control gate is given by

1
∆Yk = − ∆ω (8.27)
gstk
152

Using equation (8.24), the stationary change of power is obtained as

1
∆Pmk = − ∆ω = −Rk ∆ω (8.28)
gstk

Summing equation (8.28) for all generators, we have

ng ng ng
X X 1 X
∆Pm = ∆Pmk =− ∆ω = − Rk ∆ω
gstk
k=1 k=1 k=1 (8.29)
1
=− ∆ω = −R ∆ω
gst

where,
ng ng
1 X 1 X 1
= , R= Rk and R = (8.30)
gst k=1
gstk k=1
gst

• gst is termed the speed-droop (or the droop), and can be interpreted as the
percentage change in frequency required to move the gate from fully closed to
fully opened. The droop expressed in the nominal value has the unit ( MHzW ).

• R ( MHzW ) is termed the total gain of the governing systems.

Moreover by virtue of equations (8.4) and (8.29), the steady-state frequency deviation
(SSFD) for a load change in a system with governing systems is given by

∆PL
0 = ∆Pm − ∆PL − D ∆ω ⇒ ∆ω = − (8.31)
R+D
where, R + D is referred to as the stiffness.

Example 8.1 Consider the system shown in Figure 8.8 where a hydraulic unit and an
infinite bus feed a load. Let the active load be modeled as PL = PL0 + D ∆ ω.

Line 1 Line 2 Infinite

Gen
Bus

SL
Figure 8.8. An SMIB system.

The system data are given as follows

• Sbase = 1000 (MVA), fs = 50 (Hz)

• Pm = 900 (MW), Pmmax = 1100 (MW) , H = 5 (s), Sng = 1200 (MVA)

 153

• PL0 = 1000 (MW), D = 20 (MW/Hz)

Assume that due to a disturbance line 2 is disconnected at time t = 5 (s). Assume also
that there is no frequency control in this isolated system, i.e. ∆Pm = 0. Calculate the
steady-state frequency deviation (SSFD) of the isolated system in (Hz).

For frequency control study, the system may be represented by the block diagram
shown in Figure 8.9 where all values are expressed in (pu).

∆Pm
+
1 ∆ω
¦
- Ms + D
∆PL

Figure 8.9. Block diagram representation .

We have
Sng 100 fs
∆Pm = 0 , M = 2H = 12 , ∆PL = = 0.1 and D = 20 =1
Sbase Sbase Sbase
where ∆PL represents the lost power from the infinite bus to the load. Note also that
∆ω = ∆f since ∆ω is expressed in (pu). The SSFD of the isolated system is given by
∆Pm − ∆PL 0 − 0.1
∆f = fs = 50 = −5 (Hz)
D 1
Figure 8.10 shows the system frequency response to the disturbance.
50.5
50
f (Hz)

47.5

45
0 50 100
Time (s)

Figure 8.10. System frequency response to the disturbance.

Example 8.2 Assume that the isolated system has a frequency control function, and
the available FCR are used fully at f = 49.5 (Hz). For the same disturbance, calculate
the SSFD of the isolated system in (Hz), and analyze the response of the FCR process.
154

∆Pmmax
+¦ 1 + TR s 1 − 2Tw s ∆Pm
−R (1 + Tg1s )(1 + Tg 2 s)
- 1 + Tw s
g st

∆Pmset
∆ω 1 +
¦
Ms + D -
∆PL

Figure 8.11. Block diagram representation.

Now, we are dealing with the system shown in Figure 8.11.

The SSFD is given by (8.31). The available FCR are given by Pmmax − Pm = 200
(MW). To fulfill the frequency deviation requirement, the total system gain R is then
200 fs 1
R= = 400 (MW/Hz) or R = 400 = 20 (pu/pu) and gst = = 0.05 (pu/pu)
0.5 Sbase R

The stiffness is given by R + D = 21, and the SSFD is given by

−∆PL −0.1
∆f = fs = 50 = −0.2381 (Hz)
R+D 21

Figure 8.12 shows the system frequency response to the disturbance, with TG = 0.2,
TR = 7, gtr = 0.8 and TW = 2. From equation (8.26), Tg1 and Tg2 can be determined.

50.5

50

SSFD: −0.2381 (Hz)

49
f (Hz)

48

Maximum IFD: −2.6527 (Hz)

47
0 40 80 120
Time (s)

Figure 8.12. The response of FCR to the disturbance.

During transient the frequency drops to f = 47.3473 (Hz), i.e. the maximum instanta-
neous frequency deviation (IFD) is ∆f = −2.6527 (Hz) for this disturbance, however
it eventually settles to f = 49.7619 (Hz), i.e. the SSFD is ∆f = −0.2381 (Hz) which
was determined by the stiffness R + D.
 155

Figure 8.13 shows the generation and load characteristics of the isolated power system.
In the figure, the characteristic of the load is given by PL = PL0 + D∆f whose slope is
1
determined by , and the characteristic of the generation of the isolated system is given
D
by Pm = Pm0 − R∆f whose slope is determined by − gst . Prior to the disturbance,
the system settles at the equilibrium point indicated with “∗”, where ∆f = 0, and
Pm +100 = PL = 1000 (MW). Note that the Pm characteristic shown in the figure is for
the isolated system, i.e. the post-disturbance system. (How does the Pm characteristic
look like in the pre-disturbance system?)
When the disturbance occurs the system moves from point “∗” to point p1 at which
Pm = 900. However, the frequency initially reminds unchanged (why?). At point p1 ,
the power difference is Pm − PL = −100 (MW) (which corresponds the lost power from
the infinite bus), therefore the frequency starts to drop. When the frequency drops,
the governing system should increase the generator output. However, as shown in the
figure, the power output first decreases before starting to compensate the lost power.
The reason to that is the unpleasant dynamical property of the hydro turbine (i.e.
non-minimum phase property), and that is the reason (compared to the steam units)
why the system with frequency control based on hydraulic units has larger transient
frequency deviation, and needs longer time to reach the new equilibrium point.

P→
L
p
1
50
p4
↑
P
f (Hz)

p2
f
↓
p3

850 900 950 1000 1050

P (MW)

Figure 8.13. Generation and load characteristics of the isolated power system.

When the trajectory intersects the PL characteristic at point p3 , the output power Pm is
equal to PL (which has decreased due to its frequency dependency), and the frequency
reaches a local minimum of f = 47.3473 (Hz) (see Figure 8.12). Since p3 is not an e.p,
the trajectory does not stay at this point, but passes it. Then Pm > PL , and therefore
the system frequency starts to increase. When the trajectory intersects again the PL
characteristic, the frequency reaches a local maximum. When the trajectory passes the
intersection point, Pm becomes less than PL , and therefore the system frequency starts
to decrease. Eventually, the trajectory reaches its new equilibrium point p4 .
Figure 8.14 shows the effects of the governing system and the frequency dependency of
the load on the system frequency change. The disturbance ∆PL results in a generation
156

increase of ∆Pm , and a load reduction of D∆f .

∆P
L
50

f (Hz)
49.7619

∆P
m

D∆f

49.5
900 1000
P (MW)

Figure 8.14. The impact of the frequency dependency of the load.

Example 8.3 Apply the FRR process to restore the system frequency and the load.
Also explain the impact of the inertia constant H on the system frequency response.
In order to restore the system frequency and the load, the generation characteristic must
be shifted to a point close to the nominal frequency. Such a shift can be performed in
different ways. Since the frequency deviation and the system gain after the disturbance
are known, ∆Pm = −R∆f can be calculated. Having known ∆Pm , the generation
characteristic can be shifted either by changing the set value of the generator (i.e.
∆Pmset ) to ∆Pm , or starting a new generator whose power output should be equal
to ∆Pm . Another way is to add a supplementary control loop (an integrator) to the
governing system with ∆f as the input signal as shown in Figure 8.15.

DPmmax
+å 1 + TR s 1 - 2Tw s DPm
-R (1 + Tg1s )(1 + Tg 2 s )
- 1 + Tw s
g st
TFRR
s +
1
å
Dw Ms + D -
DPL

Figure 8.15. Block diagram representation with automatic FRR.

Figure 8.16 shows the system frequency response to FCR process (in blue) and the
FRR process (in red). In the figure on the left hand side, the FRR was performed at
t = 120 (s) by a step change of the set point to ∆Pmset = −R∆f , where R and ∆f
were calculated in Example 8.2. However, in the figure on the right hand side, the
FRR was performed at t = 120 (s) by activating the supplementary control loop shown
in Figure 8.15.
 157

50 50
f (Hz)

f (Hz)
47.3473 47.3473

0 120 200 0 120 200

Time (s) Time (s)

Figure 8.16. System frequency response to the FCR process and the FRR process.

In the first few seconds following the loss of a power plant (in this example the infinite
bus), the system frequency starts to drop. Initially, the dynamical behavior of the
system frequency is dominated by the inertial response of the generators that are
synchronously connected to the system (in this example there is only one generator).
This response is known as Inertial Frequency Response (IFR) which is an inertia-
based inherent response due to the total stored kinetic energy in the rotating masses
(mainly from synchronous generators). The IFR provides a counter response to resist
a fast frequency deviation after a disturbance. A low system inertia results in a faster
and greater frequency deviation. A large and fast frequency deviation can trigger
protection system which may result in system separation, loss of load and costumer
outages. Therefore, keeping a minimum level of system inertia is a necessity for having
a secure system operation.
Figure 8.17 shows the variations of the rate of change of frequency (RoCoF) after the
loss of the infinite bus. In the figure on the right, the region around the minimum of
RoCoF is enlarged.
−3 −3
x 10 x 10
5 −7.0863

0
df/dt (pu/s)

df/dt (pu/s)

−5

−8.6363
−10
0 50 100 150 200 6.415 9
Time (s) Time (s)

Figure 8.17. Variations of the RoCoF after the loss of the infinite bus.

In a real power system, after loss of a power plant it normally takes some time before the
governor control action (i.e. the FCR process) will be activated. Therefore, during the
first second(s) following the disturbance, the stored kinetic energy in the rotor and the
158

frequency dependency of the load are the only counter response to resist the frequency
drop (i.e. the IFR). Thus, from Example 8.2 and Example 8.3 we may conclude that
as shown in Figure 8.18 following a disturbance the system frequency response may
be classified into three different categories, namely the inertial frequency response (in
green), the response due to the FCR process (in blue) and the response due to the
FRR process (in red).

50

49
f (Hz)

48

47
0 120 200
Time (s)

Figure 8.18. Different categories of frequency response.

Figure 8.19 shows the system frequency response for H = 3 (in red), H = 5 (in green)
and H = 7 (in blue).

−0.0061
50

−0.0086
df/dt (pu/s)
f (Hz)

47.7475
47.3473

46.4814 −0.0147
0 80 6.41 8
Time (s) Time (s)

Figure 8.19. System frequency response for H = 3, H = 5 and H = 7.

As shown in the figure, the system inertia has a significant impact on the IFD and
the RoCoF. A lower H results in a greater IFD and RoCoF. The settings of some
protection systems are based on the frequency deviation and the RoCoF. For exam-
ple, the steam turbines are very sensitive for a too low frequency, and they must be
disconnected by their protection systems when the frequency drop is ∆f = α > 0
(Hz) to avoid damaging of these generators. One of the preventive actions to avoid
the frequency reaches f = fs − α (Hz) is that to disconnect some pre-defined loads
(i.e. under-frequency load-shedding) when the frequency drop is ∆f = β < α (Hz).
There are also protection systems which will be triggered to disconnect some power
 159

system components if |df /dt| > γ > 0 (Hz/s). This action may result in jeopardizing
the system stability. Therefore, the system must be operated in such a way to fulfill
N − 1 criterion with the frequency nadir greater than f = fs − β (Hz) and the RoCoF
less than γ.

Example 8.4 Consider again the system in Example 5.1. The generators are rep-
resented by the classical model with ωk in (pu), see equation (3.18). The loads are
represented by the exponential model expressed in equation (5.2). Let the disturbance
be a step increase of the load at BUS 4 with 10%. Let also the load frequency dependency
be represented by the damping constants of the generators, i.e. Dk .
a) Let mp = mq = 0 and assume that no generator has governor control, i.e. Pmk is
constant. Find a value for each Dk such that the frequency drop is ∆f = 5 (Hz) for
this disturbance. Plot also ∆f (Hz) in the COI reference frame.
b) Re-do a), but with

mp = 0 and mq = 2 , mp = 2 and mq = 0 , mp = 2 and mq = 2

c) Let mp = mq = 0, and Dk be the same as in task a). Let also Gen 1 and Gen 2 be
equipped with the governor control shown in Figure 8.15, but without the supplementary
control loop, i.e. TF RR = 0. Assume that the available FCR used fully at f = 49.5
(Hz). Analyze the system frequency response for this disturbance.
d) Re-do c), but with the supplementary control loop, i.e. TF RR 6= 0, from t ≥ 120 (s).

a-b) Based on the required frequency drop, we firstly find DCOI which is given by

∆PL fs
DCOI = − = 1.5000 (pu/pu)
∆f Sbase

Next, from equations (3.18) and (8.1)-(8.4) we find that D1 = 0.0020, D2 = 0.0016 and
D3 = 0.0012.
To run the simulation, the system dynamic described in task a) of Example 5.1 with the
generator dynamic expressed in equation (3.18) are used. Figure 8.20 shows the system
frequency response (in the COI reference frame) to the disturbance with different mp
and mq. Why does the system have different settling frequency?
c-d) Based on the given data and the frequency drop requirements, we find that R1 =
20 and R2 = 6 (pu/pu). Next, based on equation (8.31) the SSFD can be obtained
which is ∆f = 49.7273 − 50 = −0.2727 (Hz).
To run the simulation, we need to derive a set of differential equation to describe
the dynamics of the governor and turbine shown in Figure 8.15. To do that, we use
the equivalent model shown in Figure 8.21 from which the following set of differential
160

0 0

−1 mp=0 , mq=0 −1 mp=0 , mq=2 (in blue)

mp=2 , mq=0 (in red)
−2 −2 mp=2 , mq=2 (in green)
∆f (Hz)

∆f (Hz)
−3 −3

−4 −4

−5 −5
10 100 200 10 100 200
Time (s) Time (s)

Figure 8.20. System frequency response to the disturbance with different values
for mp and mq.

equations can be obtained.

−1
Ṡ1k = (c1k Rk + S1k )
Tg1k
1
Ṡ2k = (c2k − S2k )
Tg2k (8.32)
1
Ṡ3k = (c4k − S3k )
TW k
∆Ṗ msetk = ωk TF RRk

where,

c1k =ωk − gstk ∆P msetk

TRk
c2k =S1k (1 − )
Tg2k
(8.33)
S1k TRk
c3k = + S2k
Tg2k
c4k =3 c3k

and finally

∆Pmk = −2 c3k + S3k (8.34)

TFRR
DPmset
g st
s

Dw - c1 -R s1 TR + c2 1 s2 c3 + c4 s3 DPm
- - 1
å S S -2 S S
+ 1 + Tg 1s Tg 2 1 + Tg 2 s + 1 + Tw s +
+ +

Figure 8.21. The equivalent model of the system shown in Figure 8.15.
 161

Figure 8.22 shows the system frequency response (on the left hand side) and the varia-
tions of the mechanical and electrical powers (on the right hand side) to the disturbance.

0
0.15

SSFD: −0.2727 (Hz)

∆Pm & ∆Pe (pu)

∆P in blue
m
∆f (Hz)

∆P in red
e

Maximum IFD: −1.8098 (Hz)

0

−2
10 100 200 10 100 200
Time (s) Time (s)

Figure 8.22. System frequency response and the variations of the mechanical
and electrical powers with FCR.

As shown in the figure on the left hand side, the frequency nadir is f = 48.1902 (Hz),
and the settling frequency is f = 49.7273 (Hz) (as was expected). When the load is
suddenly increased with ∆PL = 0.15, initially the frequency is unchanged and the load
frequency dependency will therefore be unaffected. According to Kirchhoff’s current
(power) law, ∆Pe = ∆Pe1 + ∆Pe2 + ∆Pe3 = ∆PL at any instant of time. As shown
in the figure on the right hand side, immediately after the disturbance ∆Pe = 0.15
(pu), but ∆Pm = 0. Where is this additional power coming from? Moreover, as
shown in the figure, the total increased of the load is not entirely compensated, i.e.
∆Pm = ∆Pe 6= 0.15 (pu). This is due to the load frequency dependency, since ∆f 6= 0.
By applying the FRR process, i.e. by activating the supplementary control loop with
TF RR = −0.5 for Gen 1 and Gen 2, both the frequency and the total new load are
restored as shown in Figure 8.23.

0
0.15
∆Pm & ∆Pe (pu)

∆Pm in blue
∆f (Hz)

∆P in red
e

Maximum IFD: −1.8098 (Hz)

0

−2
10 120 200 300 10 120 200 300
Time (s) Time (s)

Figure 8.23. System frequency response and the variations of the mechanical
and electrical powers with FCR and FRR.
162

Example 8.5 Let synchronous generator Gen 3 be replaced by a wind power plant with
identical Pg and Qg to obtain the same load flow results. Re-do task c) of Example 8.4.
From the load flow calculations we find that Pg3 = 0.2000 and Qg3 = 0.1579 which
will be used at BUS 3 as power generation at that bus. Then BUS 3 is considered as
a pq-bus and the simulation is performed. Figure 8.24 shows some of the simulation
results. Due to the fast power electronic based controller in the modern wind power
5
0

∆δ1 (deg.)
∆f (Hz)

1
Maximum IFD:
−1.8098 (Hz) without wind power 0
Maximum IFD: −2.0186 (Hz) with wind power
−2
−1
10 100 10 15 20
Time (s) Time (s)

Figure 8.24. System frequency response and the variations of ∆δ1 with (red)
and without(blue) wind power.

plants, the power system ”sees” constant power generations from these wind power
plants (which are also termed as non-synchronous generation). This implies that these
wind power plants (or in general non-synchronous generations) are not able inherently
to be involved in IFR. Today’s power systems have been designed and developed based
on synchronously connected power plants. To operate a power system in a reliable and
cost-effective manner, a mixture of three types of resources has been applied, namely
energy, capacity (i.e. power) and flexibility (i.e. controllability). All of conventional
power plants (i.e. those which are synchronously connected to the system) imply an
important contribution to ”capacity”. Moreover, they provide synchronizing power (or
torque) and inertia which have crucial roles on power system dynamical response, on
setting of rotor angle, voltage and frequency stability limits, and also on setting of
protection systems. The non-synchronous generation does not however contribute in
providing synchronizing power and inertia. Replacing some of the conventional power
plants with the non-synchronous generation (due to environmental and economical
concerns), it even results in less system inertia and synchronizing power. The high
penetration of non-synchronous generation (mostly wind power, but it can also include
solar power and other renewable energy) can therefore result in new challenges to
operate the system in a secure and cost-effective manner.
As shown in Figure 8.24, the IFD of the system becomes greater when the conventional
Gen 3 is replaced by a wind power plant. This means that the system has less inertia
compared to the system with three conventional generators. The lower inertia and
synchronizing power result also in the system oscillations with higher amplitude as
shown in the figure on the right hand side. A low inertia and synchronizing power can
jeopardize the system stability in case of large disturbances.
Appendix A
Load flow calculations based on
Newton-Raphson method
A.1 Theory
The Newton-Raphson method may be applied to solve for y1 , y2 , · · · , yn of the following
non-linear equations,
g1 (y1 , y2 , · · · , yn ) = Υ1 (y1 , y2 , · · · , yn ) − b1 = 0
g2 (y1 , y2 , · · · , yn ) = Υ2 (y1 , y2 , · · · , yn ) − b2 = 0
.. (A.1)
.
gn (y1 , y2 , · · · , yn ) = Υn (y1 , y2, · · · , yn ) − bn = 0
or in the vector form
g(y) = Υ(y) − b = 0 (A.2)

where
       
y1 g1 (y) Υ1 (y) b1
 y2   g2 (y)  Υ2 (y)   b2 
       
y =  ..  , g(y) =  ..  , Υ(y) =  ..  , b =  .. 
.  .   .  .
yn gn (y) Υn (y) bn

y is an n × 1 vector which contains variables, b is an n × 1 vector which contains

constants, and f (y) is an n × 1 vector-valued function.
Taylor’s series expansion of (A.2) is the basis for the Newton-Raphson method of
solving (A.2) in an iterative manner. From an initial estimate (or guess) y (0) , a sequence
of gradually better estimates y (1) , y (2) , y (3) , · · · will be made that hopefully will converge
to the solution y ∗ .
Let y ∗ be the solution of (A.2), i.e. g(y ∗) = 0, and y (i) be an estimate of y ∗. Let also
∆y (i) = y ∗ − y (i) . Equation (A.2) can now be written as
g(y ∗) = g(y (i) + ∆y (i) ) = 0 (A.3)
Taylor’s series expansion of (A.3) gives
(i) )
g(y (i) + ∆y (i) ) = g(y (i) ) + JAC (y ∆y (i) = 0 (A.4)
where  ∂g1 (y) ∂g1 (y) 
  ∂y1
··· ∂yn
(i) ) ∂g(y)  .. .. .. 
JAC (y = = . . .  (A.5)
∂y y=y (i) ∂gn (y) ∂gn (y)
∂y1
··· ∂yn y=y (i)

163
164

where, JAC ia called the jacobian of g.

From (A.4), ∆y (i) can be calculated as follows
(i) )
JAC (y ∆y (i) = 0 − g(y (i)) = ∆g(y (i) ) ⇒ (A.6)
h (i)
i−1
∆y (i) = JAC (y ) ∆g(y (i) ) (A.7)

Since g(y (i)) = Υ(y (i) ) − b, ∆g(y (i) ) is given by

∆g(y (i) ) = b − Υ(y (i) ) = −g(y (i) ) (A.8)

(i) )
Furthermore, since b is constant, JAC (y is given by
 ∂Υ1 (y) ∂Υ1 (y) 
    ∂y1
··· ∂yn
(i) ) ∂g(y) ∂Υ(y)  .. .. .. 
JAC (y = = = . . .  (A.9)
∂y y=y (i) ∂y y=y (i) ∂Υn (y) ∂Υn (y)
∂y1
··· ∂yn y=y (i)

Therefore, ∆y (i) can be calculated as follows

 (i)   ∂Υ1 (y) ∂Υ1 (y) −1  (i) (i)

∆y1 ∂y1
··· ∂yn b1 − Υ1 (y1 , · · · , yn )
 ..   .. .. .. ..
∆y (i)
  
= . = . . .   .  (A.10)
(i) ∂Υn (y) ∂Υn (y) (i) (i)
∆yn ∂y1
··· ∂yn bn − Υn (y1 , · · · , yn )
y=y (i)

Finally, the following is obtained

i = i+1
(i)
y = y (i−1) + ∆y (i−1)

The intention is that y (1) will estimate the solution y ∗ better than what y (0) does. In the
same manner, y (2) , y (3) , · · · can be determined until a specified condition is satisfied.
Thus, we obtain an iterative method according to the flowchart in Figure A.1.

Example A.1 Using the Newton-Raphson method, solve for y of the equation

g(y) = k1 y + k2 cos(y − k3 ) − k4 = 0

Let k1 = −0.2, k2 = 1.2, k3 = −0.07, k4 = 0.4 and ǫ = 10−4 .

This equation is of the form given by (A.2), with Υ(y) = k1 y + k2 cos(y − k3 ) and
b = k4 .
Step 1
Set i = 0 and y (i) = y (0) = 0.0524 (radians), i.e. 3 (degrees).
 165

Set i = 0 Step 1
Give y (i )

Calculate D g (y(i ) ) Step 2

Step Final
Is the magnitude of the all entries of D g (y(i ) ) Yes
y = y (i )
less than a small positve number ε?
No

Calculate JAC (i ) Step 3

Calculate D y (i ) Step 4

i = i +1
Step 5
y (i ) = y (i -1) + D y (i -1)

Figure A.1. Flowchart for the Newton-Raphson method.

Step 2
∆g(y (i) ) = b − Υ(y (i) ) = 0.4 − [(−0.2 ∗ 0.0524) + 1.2 cos(0.0524 + 0.07)] = −0.7806
Go to Step 3 since |∆g(y (i) )| > ǫ
Step 3
h i
(i) ) ∂Υ
JAC (y = ∂y
= −0.2 − 1.2 sin(0.0524 + 0.07) = −0.3465
y=y (i)

Step 4
h i−1
(i)
∆y (i) = JAC (y ) ∆g(y (i) ) = −0.7806
−0.3465
= 2.2529

Step 5
i=i+1=0+1=1
y (i) = y (i−1) + ∆y (i−1) = 0.0524 + 2.2529 = 2.3053. Go to Step 2
After 5 iterations, i.e. i = 5, it was found that |∆g(y (i))| < ǫ for y (5) = 0.9809 (rad.).
Therefore, the solution becomes y = 0.9809 (rad.) or y = 56.2000 (deg.).
Analysis of Example A.1
Figure A.2 shows variations of g(y) versus y. The figure shows that the system (or
equation) has only three solutions, i.e. the points at which g(y) = 0. Due to practical
issues, y ∗ indicted with (O) in the figure is the interesting solution.
166

2.5

1.5

1
g(y) 0.5 y*
0

−0.5

−1

−1.5

−2

−2.5
−400 −300 −200 −100 0 100 200 300 400
y

Figure A.2. Variations of g(y) vs. y.

Figure A.3 shows how the equation is solved by the Newton-Raphson method.

g(y(0))
g(y) (0)
JAC(y )

y(1)
0
(2)
y(0) y

(1)
JAC(y )

g(y(1))

Figure A.3. Variations of g(y) vs. y.

We first guess the initial estimate y (0) . In this case y (0) = 0.0524 (rad.),
h i.e i3 (deg.).


The tangent to g(y) through the point y , g(y ) , i.e. g (y ) = dg(y)
(0) (0) ′ (0)
dy (0)
=
y=y
(y (0) ) (1)
JAC , intersects the x–axis at point y . The equation for this tangent is given by
G − g(y (0) ) = g ′ (y (0) ) ∗ (y − y (0) )
The intersection point y (1) is obtained by setting G = 0, i.e.
g(y (0) ) ′ (0) −1


y (1) = y (0) − = y (0)
− g (y ) g(y (0))
g ′ (y (0) )

−1 h (0)
i−1
∆y (0) = y (1) − y (0) = − g ′(y (0) ) g(y (0) ) = JAC (y ) ∆g(y (0) )
 167

In a similar manner, y (2) can be obtained which is hopefully a better estimate than
y (1) . As shown in the figure, from y (2) we obtain y (3) which is a better estimate of y ∗
than what y (2) does. This iterative method will be continued until |∆g(y)| < ǫ.

Example A.2 Solve for x in Example A.1, but let y (0) = 0.0174 (rad.), i.e. 1 (deg.).

D.I.Y, (i.e., Do It Yourself)

A.2 Application to power systems

Consider a symmetrical power system with N buses. The aim is to determine the
voltage at all buses in the system by applying the Newton-Raphson method. All
variables are expressed in (pu).
Figure A.4 schematically shows connection of the system components to bus k.

I k1
Uk
IG k
~ Ik 2

I Lk
I kN

Figure A.4. Notation of bus k in a network.

The generator generates the current I¯Gk , the load at the bus draws the current I¯Lk ,
and I¯kj is the currents from bus k to the neighboring buses. According to Kirchhoff’s
current law, the sum of all currents injected into bus k must be zero, i.e.
N
X
I¯Gk − I¯Lk = I¯kj (A.11)
j=1

By taking the conjugate of equation (A.11) and multiply the equation with the bus
voltage, the following holds
N
X
Ūk I¯Gk
∗
− Ūk I¯Lk
∗
= Ūk I¯kj
∗
(A.12)
j=1

This can be rewritten as an expression for complex power in the per-unit system as
N
X
S̄Gk − S̄Lk = S̄kj (A.13)
j=1
where
168

S̄Gk = PGk + jQGk is the generated complex power at bus k,

S̄Lk = PLk + jQLk is the consumed complex power (the load) at bus k,

S̄kj = Pkj + jQkj is the complex power flow from bus k to bus j.

The power balance at the bus according to equation (A.13) must hold both for the
active and for the reactive part of the expression. By using PSk and QSk as notation
for the specified active and reactive power (or net generation of active and reactive
power) at bus k, respectively, the following expression holds
N
X
PSk = PGk − PLk = Pkj (A.14)
j=1
N
X
QSk = QGk − QLk = Qkj (A.15)
j=1

i.e. for any bus k in the system, the power balance must hold for both active and
reactive power.
Now, consider the π-equivalent model of a line shown in Figure A.5, where all variables
expressed in per-unit.

Uk Z kj Uj
S kj I
I sh
jbsh − kj jbsh − kj

Figure A.5. π-equivalent model of a line.

Let
Ūk = Uk ejθk , Ūj = Uj ejθj
√
Z̄ = R + jX , Z = R2 + Y 2 (A.16)
θkj = θk − θj

The power S̄kj in the sending end k is given by

!


∗ Ūk∗ − Ūj∗
S̄kj = Ūk I¯sh
∗
+ I¯∗ = Ūk j bsh−kj Ūk + ∗
=
Z̄kj
Uk2 Uk Uj j(θk −θj ) (A.17)
= −j bsh−kj Uk2 + − e =
R − jX R − jX
U2 Uk Uj
= −j bsh−kj Uk2 + k2 (R + jX) − (R + jX) (cos θkj + j sin θkj )
Y Y2
 169

By dividing equation (A.17) into a real and an imaginary part, expressions for the
active and reactive power can be obtained, respectively, as
R 2 Uk Uj
Pkj = U + (X sin θkj − R cos θkj )
Z2 k Z2    (A.18)
R 2 Uk Uj R
= 2 Uk + sin θkj − arctan
Z Z X

X Uk Uj
Qkj = −bsh−kj Uk2 + 2 Uk2 − (R sin θkj + X cos θkj )
Z Z2
    (A.19)
X 2 Uk Uj R
= −bsh−kj + 2 Uk − cos θkj − arctan
Z Z X

From equations (A.18) and (A.19), it can be concluded that if the phasor voltages (i.e.
the voltage magnitude and phase angle) at both ends of the line are known, the power
flow can be uniquely determined. This implies that if the phasor voltages of all buses
in a system are known, the power flows in the whole system are known, i.e the phasor
voltages define the system state.
Consider again Figure A.5. Let
1 1 R −X
gkj + j bkj = = = 2 +j 2 ⇒
Z̄kj R+jX Z Z
R (A.20)
gkj = 2
Z
X
bkj = − 2
Z
Based on (A.20), we rewrite (A.18) and (A.19) as follows
Pkj = gkj Uk2 − Uk Uj [gkj cos(θkj ) + bkj sin(θkj )] (A.21)
Qkj = Uk2 (−bsh−kj − bkj ) − Uk Uj [gkj sin(θkj ) − bkj cos(θkj )] (A.22)
The current through the line, and the active power losses in the line can be calculated
by
Pkj − j Qkj
I¯kj = (A.23)
Ūk∗
Plosskj = Pkj + Pjk (A.24)

Consider again Figure A.4. Let Y = G+jB denote the admittance matrix of the system
(or Y-matrix), where Y is an N × N matrix, i.e. the system has N buses. The relation
between the injected currents into the buses and the voltages at the P
buses is given by
¯
I = Y U. Therefore, the injected current into bus k is given by Ik = N j=1 Ȳkj Ūj .

The injected complex power into bus k can now be calculated by

N
X N
X
S̄k = Ūk I¯k∗ = Ūk Ȳkj∗ Ūj∗ = Uk (Gkj − jBkj ) Uj (cos(θkj ) + j sin(θkj ))
j=1 j=1
170

N
! N
!
X X
= Uk Uj [Gkj cos(θkj ) + Bkj sin(θkj )] +j Uk Uj [Gkj sin(θkj ) − Bkj cos(θkj )]
j=1 j=1

Let Pk denote the real part of S̄k , i.e. the injected active power, and Qk denote the
imaginary part of S̄k , i.e. the injected reactive power, as follows:
N
X
Pk = Uk Uj [Gkj cos(θkj ) + Bkj sin(θkj )]
j=1
N
(A.25)
X
Qk = Uk Uj [Gkj sin(θkj ) − Bkj cos(θkj )]
j=1

Note that Gkj = −gkj and Bkj = −bkj for k 6= j. Furthermore,

N
X
Pk = Pkj
j=1
N
X
Qk = Qkj
j=1

Equations (A.14) and (A.15) can now be rewritten as

Pk − PSk = 0
(A.26)
Qk − QSk = 0

which are of the form given in equation (A.2), where


    
θ1 P1 PS1
 ..   ..   .. 

    .   
  .     . 
 
θ  θN  ΥP (θ, U)  PN  bP  PSN 
  =   
y =   =   , Υ(θ, U) =   , b=  =   (A.27)
     
U  U1  ΥQ (θ, U)  Q1  bQ  QS1 
 .   .   . 
 ..   ..   .. 
UN QN QSN

The aim is to determine y = [θ U]T by applying the Newton-Raphson method.

Assume that there are 1 slack bus and M PU-buses in the system. Therefore, θ becomes
an (N − 1) × 1 vector and U becomes an (N − 1 − M) × 1 vector, why?
Based on (A.8), we define the following:

∆Pk = PSk − Pk k 6= slack bus

(A.28)
∆Qk = QSk − Qk k =6 slack bus and PU-bus
 171

Based on (A.9), the jacobian matrix is given by

 ∂ΥP (θ,U ) ∂ΥP (θ,U )   
∂θ ∂U H N′
JAC =  =  (A.29)
∂ΥQ (θ,U ) ∂ΥQ (θ,U ) ′
J L
∂θ ∂U
where,
H is an (N − 1) × (N − 1) matrix
N ′ is an (N − 1) × (N − M − 1) matrix
J is an (N − M − 1) × (N − 1) matrix
L′ is an (N − M − 1) × (N − M − 1) matrix
The entries of these matrices are given by:
∂Pk
Hkj = ∂θj
k 6= slack bus j 6= slack bus

′ ∂Pk
Nkj = ∂Uj
k 6= slack bus j 6= slack bus and PU-bus

∂Qk
Jkj = ∂θj
k 6= slack bus and PU-bus j 6= slack bus

∂Qk
L′kj = ∂Uj
k 6= slack bus and PU-bus j 6= slack bus and PU-bus
Based on (A.6), (A.28) and (A.29), the following is obtained
    
H N ′ ∆θ ∆P
= (A.30)
J L′ ∆U ∆Q
To simplify the entries of the matrices N ′ and L′ , these matrices are multiplied with
U. Then, (A.30) can be rewritten as
    
H N ∆θ ∆P
= (A.31)
J L ∆U U
∆Q
where,
for k 6= j
∂Pk
Hkj = ∂θj
= Uk Uj [Gkj sin(θkj ) − Bkj cos(θkj )]

′
Nkj = Uj Nkj = Uj ∂P
∂Uj
k
= Uk Uj [Gkj cos(θkj ) + Bkj sin(θkj )]
(A.32)
∂Qk
Jkj = ∂θj
= −Uk Uj [Gkj cos(θkj ) + Bkj sin(θkj )]

Lkj = Uj L′kj = Uj ∂Q
∂Uj
k
= Uk Uj [Gkj sin(θkj ) − Bkj cos(θkj )]
and for k = j
∂Pk
Hkk = ∂θk
= −Qk − Bkk Uk2

∂Pk
Nkk = Uk ∂U k
= Pk + Gkk Uk2
(A.33)
∂Qk
Jkk = ∂θk
= Pk − Gkk Uk2

Lkj = Uk ∂Q
∂Uk
k
= Qk − Bkk Uk2
172

Now based on (A.10), the following is obtained:

   −1  
∆θ H N ∆P
∆U = (A.34)
U J L ∆Q

Finally, U and θ will be updated as follows:

θk = θk +∆θk  k 6= slack bus

∆Uk (A.35)
Uk = Uk 1 + Uk
k 6= slack bus and PU-bus

A.2.1 Newton-Raphson method for solving power flow

equations

Newton-Raphson method can be applied to non-linear power flow equations as follows:

• Step 1

1a) Read bus and line data. Identify slack bus, PU-buses and PQ-buses.
1b) Develop the Y-matrix and calculate the specified powers, i.e. PS = PG − PL
and QS = QG − QL .
1c) Give the initial estimate of the unknown variables, i.e. U for PQ-buses and
θ for PU- and PQ-buses. It is very common to set U = Uslack and θ = θslack .
However, the flat initial estimate may also be applied, i.e. U = 1 and θ = 0.
1d) Go to Step 2.

• Step 2

2a) Calculate the injected power into each bus by equation (A.25).
2b) Calculate the difference between the net production and the injected power
for each bus, i.e. ∆P and ∆Q by equation (A.28).
2c) Is the magnitude of all entries of [∆P ∆Q]T less than a specified small
positive constant ǫ ?
∗ If yes, go to Step Final.
∗ if no, go to Step 3.

• Step 3

3a) Calculate the jacobian by equations (A.32) and (A.33).

3b) Go to Step 4.

• Step 4
 
∆U T
4a) Calculate ∆θ U
by equation (A.34).
4b) Go to Step 5.
 173

• Step 5

5a) Update U and θ by equation (A.35).

5b) Go till Step 2.

• Step Final

– Calculate the generated powers, i.e. PG (MW) and QG (MVAr) in the slack
bus, and QG (MVAr) in the PU-buses by using equation (A.26).
– Calculate the power flows (MW, MVAr) by using equations (A.21) and
(A.22).
– Calculate active power losses (MW) by using equation (A.24).
– Give all the voltage magnitudes (kV) and the voltage phase angles (degrees).
– Print out the results.

A.3 The system data of the 3-generator system

NS=0;
Pg2=0.3;
Pg3=0.2;
PL01=0.0; QL01=PL01*0.1;
PL02=0.0; QL02=PL02*0.1;;
PL03=0.0; QL03=PL03*0.1;
PL04=1.5; QL04=PL04*0.1;

%%%%%%%%%%%
% Bus Data
%%%%%%%%%%%

% NS means "not specified"

% Type=1 means slack-bus, Type=2 means PU-bus, Type=3 means PQ-bus

BUSDATA=[
% 1 2 3 4 5 6 7 8 9 10
% BUS Type Pgen Qgen Pload Qload YL Ysh V Angle
1 1 NS NS PL01 QL01 0 0 1.00 0
2 2 Pg2 NS PL02 QL02 0 0 1.00 0
3 2 Pg3 NS PL03 QL03 0 0 1.00 0
4 3 0 0 PL04 QL04 0 0 1.00 0];

%%%%%%%%%%%%
% Line Data
%%%%%%%%%%%%
174

LINEDATA = [
% Line from to R X B
1 1 4 0 0.5 0
2 2 4 0 0.2 0
3 3 4 0 0.3 0];

%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%% Generator data
%%%%%%%%%%%%%%%%%%%%%%%%%%%%

% GENDATA will be used in the examples of Chapter 4 and Chapter 6

% 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
% Gen xt xdp xd H D Tdop Te KA T1 T2 T3 T4 Tw KPSS
GENDATA = [ 1 0.1 0.10 0.8 5 0 6 0.01 100 1 1 1 1 10 0
2 0.1 0.10 0.7 4 0 6 0.01 100 1 1 1 1 10 0
3 0.1 0.10 0.6 3 0 6 0.01 100 1 1 1 1 10 0];

% GOVDATA will be used in the examples of Chapter 8

% 1 2 3 4 5 6 7 8
% Gen TWAT TG G_tr TR dPm_max dPm_min GOV
GOVDATA = [ 1 1.0 0.4 1.0 5 0.2*PG(1) -0.5*PG(1) 1
2 1.0 0.4 1.0 5 0.2*PG(2) -0.5*PG(2) 1
3 1.0 0.4 1.0 5 0.2*PG(3) -0.5*PG(3) 0];

% TWAT is the water time constant TW

GOV=1 means the generator is equipped with a governor system
Appendix B
B.1 Estimation of the stability region
The estimation of the stability region of this system can be obtained by solving the
following optimization problem

min V(x) subject to V̇(x) = 0

First we define Lagrangian function by

L(x, γ) = V(x) − γ V̇(x)

where, γ is called a Lagrange multiplier. The necessary conditions for the optimization
problem are
∂ L(x, γ)
= 0
∂x
V̇(x) = 0

Based on the above conditions, it can easily be obtained that x21 = 7.5 and x22 = 5
which imply that V(x) = c = 12.5.

B.2 The Jacobian matrices of the linearized system

Let

• N be the number of network buses

• ng be the number of generators
• ns be the number of state variables for each generator.
• nx = ns × ng be the number of all system state variables
• ny = 2 × N be the number of all system algebraic variables
• the loads be modeled as
 mp  mq
U U
PL (U) = PL0 and QL (U) = QL0
U0 U0

Assume that the generators are represented by one-axis model. i.e. ns = 3. Then, the
dynamic of the system is given by

ẋ = f (x, y)
 
gP (x, y)
0 = g(x, y) =
gQ (x, y)

175
176

The structure of f (x, y) and g(x, y) is given in Example 5.1.

The Jacobian matrix fx is of order nx × nx , and has the form
 E′

f δ fδ̇ω fδ̇ q
 δ̇ 
E′ 
fx = fx1axis =  fω̇δ fω̇ω

fω̇ q  (B.1)
 E′

fEδ˙ ′ fEω˙′ fE˙ ′q
q q q

All submatrices in (B.1) are of order ng × ng with the following nonzero elements
for k = 1 · · · ng

fδ̇ω (k, k) = 1
1
fω̇δ (k, k) = − bdk Eqk′
Uk0 cos(δk0 − θk0 )
Mk 0

E′ 1
fω̇ q (k, k) = − bdk Uk0 sin(δk0 − θk0 )
Mk
xdk − x′dk
fEδ˙ ′ (k, k) = − ′ Uk0 sin(δk0 − θk0 )
q Tdok x′dk
E′ xdk
fE˙ ′q (k, k) = − ′
q Tdok x′dk

The Jacobian matrix fy is of order nx × ny , and has the form

 θ 
fδ̇ fδ̇U
 θ U
fy = fy1axis =  fω̇ fω̇  (B.2)
fEθ˙ ′ fEU˙′
q q

All submatrices in (B.2) are of order ng × N with the following nonzero elements
for k = 1 · · · ng
1
fω̇θ (k, k) = ′
bdk Eqk Uk0 cos(δk0 − θk0 )
Mk 0

1
fω̇U (k, k) = − bdk Eqk′
sin(δk0 − θk0 )
Mk 0

xdk − x′dk
fEθ˙′ (k, k) = ′
Uk0 sin(δk0 − θk0 )
q Tdok x′dk
xdk − x′dk
fEU˙′ (k, k) = ′
cos(δk0 − θk0 )
q Tdok x′dk

The Jacobian matrix gx is of order ny × nx , and has the form

" Eq′
#
δ ω
g P g P g P
gx = gx1axis = E′
(B.3)
δ ω
gQ gQ gQ q
 177

All submatrices in (B.3) are of order N × ng with the following nonzero elements (see
equations (5.8) and (5.10))
for k = 1 · · · ng

gPδ (k, k) = −bdk Eqk

′
0
Uk0 cos(θk0 − δk0 )
E′
gP q (k, k) = bdk Uk0 sin(θk0 − δk0 )
δ ′
gQ (k, k) = −bdk Eqk 0
Uk0 sin(θk0 − δk0 )
E′
gQq (k, k) = −bdk Uk0 cos(θk0 − δk0 )

The Jacobian matrix gy is of order ny × ny , and has the form

" #
gPθ gPU
gy = θ U
(B.4)
gQ gQ

All submatrices in (B.4) are of order N × N, and given by

 ∂Pg   ∂Pg 
01 01 ∂PL
gPθ = H+ ∂θ , gPU =N + ′ ∂U +
02 03 02 03 ∂U
 ∂Qg   ∂Qg 
θ 01 U 01 ∂QL
gQ = J + ∂θ , gQ = L′ + ∂U +
02 03 02 03 ∂U

where,

• H, N ′ , J and L′ are the submatrices of the Jacobian matrix defined in (A.29).

Note that all these submatrices are of order N × N that is all elements associated
with slack and P U-buses are included.

• 01 is a zero matrix of order ng × (N − ng ), 02 = (01 )T and 03 is a zero matrix of

order (N − ng ) × (N − ng ).

∂Pg ∂Pg ∂Qg ∂Qg

• , , and are diagonal matrices of order ng × ng whose diagonal
∂θ ∂U ∂θ ∂U
elements are
for k = 1 · · · ng

∂Pg ′
(k, k) = bdk Eqk Uk0 cos(θk0 − δk0 )
∂θ 0

∂Pg ′
(k, k) = bdk Eqk sin(θk0 − δk0 )
∂U 0

∂Qg ′
(k, k) = bdk Eqk Uk0 sin(θk0 − δk0 )
∂θ 0

∂Qg ′


(k, k) = bdk 2 Uk0 − Eqk cos(θk0 − δk0 )
∂U 0
178

∂PL ∂QL
• and are diagonal matrices of order N × N whose diagonal elements
∂U ∂U
are for k = 1 · · · N
∂PL mpk PL0k
(k, k) =
∂U Uk0
∂QL mqk QL0k
(k, k) =
∂U Uk0

Finally,

∆ẋ = fx − fy (gy )−1 gx ∆x = As ∆x

AVR:
If generators are also equipped with AVR whose dynamic is given by (3.101), then
ns = 4. Therefore, fx , fy and gx are modified as follows
  
E′ E E′ 
fδ̇δ fδ̇ω fδ̇ q fδ̇ f fδ̇δ fδ̇ω fδ̇ q 0
   
E′ Eq′
 fω̇ fω̇ω fω̇ q fω̇Ef  
 δ  δ ω 
 f ω̇ f ω̇ fω̇ 0 
fx =fxavr =  δ
 
E ′
E  =  E ′
Ef 

f ˙ ′ f ω˙ ′ f ˙ ′q f ˙ ′f   f δ
f ω
 E˙q′ E˙q′ E˙q′ f q
f
 Eq Eq Eq Eq 
 E˙q′ 

 Eq′  Ef
δ ω Ef
fE˙ fE˙ fE˙ fE˙ 0 0 0 fE˙f
f f f f
 θ   
fδ̇ fδ̇U fδ̇θ fδ̇U
 θ    (B.5)
 fω̇ fω̇U   fω̇θ fω̇U 
fy =fyavr =  θ
   
U  =  θ U 
fE˙q′ fE˙q′  fE˙q′ fE˙q′ 
   
fEθ˙ fEU˙ 0 fEU˙
 f f
′
 f 
δ ω E q E f δ ω Eq′
g P g P g P g P  g P g P g P 0
gx =gxavr =  E ′
E
= Eq′

δ ω q f δ ω
gQ gQ gQ gQ gQ gQ gQ 0

where, the additional nonzero elements are

for k = 1 · · · ng

E 1
fE˙ ′f (k, k) = ′
q Tdok
E 1
fE˙ f (k, k) = −
f Tek
KAk
fEU˙f (k, k) = −
Tek

PSS:
 179

If only a PSS with nf = 2 is used in the system (assume that it is installed in the
generator k), then nx = navr
x + 3.

Therefore, fx , fy and gx are modified as follows (see equation (3.112))

 
Eq′ Ef
fδ̇δ fδ̇ω fδ̇ fδ̇ fδ̇S1k fδ̇S2k fδ̇S3k
 
 δ ω Eq′ Ef S1k S2k S3k 
 fω̇ fω̇ fω̇ fω̇ fω̇ fω̇ fω̇ 
 
 δ ω Eq′ Ef S1k S2k S3k 
f ′ f ′ f ′ f ′ f ′ f ′ f ′ 
 Ė q Ė q Ė q Ė q Ė q Ė q Ė q 
 Eq′
Ef

1pss fδ ω S1k S2k S3k 
fx =fx =  Ėf f Ėf
fĖf
fĖf
fĖf
fĖf
fĖf 
 
 δ Eq′
f ω Ef S1k S2k S3k 
 Ṡ1k fṠ1k fṠ1k fṠ1k fṠ1k fṠ1k fṠ1k 

 Eq′ Ef

f δ ω S1k S2k S3k 
 Ṡ2k fṠ2k fṠ2k fṠ2k fṠ2k fṠ2k fṠ2k 
 
δ ω Eq′ Ef S1k S2k S3k
fṠ fṠ fṠ fṠ fṠ fṠ fṠ
3k 3k 3k 3k 3k 3k 3k
 θ U 
fδ̇ fδ̇ (B.6)
 θ
 fω̇ fω̇U 

 θ 
 f ′ f U′ 
 Eq ˙ ˙
Eq 
 
θ U 
1pss 
fy =fy =  E˙f f f E˙f 
 
f θ U 
 Ṡ1k fṠ1k 
 
f θ f U 
 Ṡ2k Ṡ2k 
fṠθ fṠU
3k 3k
 
δ ω Eq′ Ef S1k S2k S3k
g P g P g P g P g P g P g P 
gx =gx1pss =  E ′
E
δ ω S1k S2k S3k
gQ gQ gQ q gQ f gQ gQ gQ

E′ E
Note that fĖδ , fĖω , fĖ q , fĖ f , fĖθ and fĖU are indicated in blue (not red) since the
f f f f f f
selection of the input signal uink may affect these submatrices.
Comments:

• If one of the generator terminal buses is considered as an infinite bus, then all
rows and columns of fx , fy , gx and gy associated both to the state variables of
the corresponding generator, and the algebraic variables of the generator terminal
bus are removed.

• In general, the dynamic of a multi-machine power system may be described by

ẋ =f (x, y, u)
(B.7)
0 =g(x, y, u)
180

Linearizing the above nonlinear system around its e.p, the following is obtained

∆ẋ = fx ∆x + fy ∆y + fu ∆u (B.8)
0 = gx ∆x + gy ∆y + gu ∆u (B.9)

From (B.9),
∆y = −gy−1 [gx ∆x + gu ∆u] (B.10)

Using (B.10) in (B.8), the following is obtained

   
∆ẋ = fx − fy gy−1 gx ∆x + fu − fy gy−1 gu ∆u
(B.11)
=A ∆x + B ∆u

Let the output function be given by

Y = ℏ(x, y, u) (B.12)

Linearization of this function around the e.p yields

∆Y = ℏx ∆x + ℏy ∆y + ℏu ∆u (B.13)

Using (B.10) in (B.13), the following is obtained

   
∆Y = ℏx − ℏy gy−1 gx ∆x + ℏu − ℏy gy−1 gu ∆u
(B.14)
=C ∆x + D ∆u

Thus, we are dealing with this Linear Time-Invariant system

∆ẋ =A ∆x + B ∆u
(B.15)
∆Y =C ∆x + D ∆u
 181

References
[1] IEEE/CIGRE Joint Task Force on Stability Terms and Definitions, Definition
and Classification of Power System Stability, Power Systems, IEEE Transactions
on ,Volume: 19 , Issue: 3 , Aug. 2004.

[2] H. K. Khalil, Nonlinear Systems (second edition), Prentice-Hall, Inc., 1996.

[3] P. M. Andersson and A. A. Fouad, Power System Control and Stability, The Iowa
State University Press, 1977.

[4] M. Pavella and P.G. Murthy, Transient Stability of Power Systems, Theory and
Practice, John Wiley & Sons, 1994.

[5] J. Machowski, J. W. Bialek and J. R. Bumby Power System Dynamics and Sta-
bility, John Wiley & Sons, 1997.

[6] P. Kundur, Power System Stability and Control, McGraw-Hill Inc., 1994.

[7] M. A. Pai, Energy Function Analysis for Power System Stability, Kluwer Academic
Publishers, 1989.

[8] IEEE Task Force on Load Representation for Dynamic performance, Load Rep-
resentation for Dynamic Performance Analysis, IEEE Transactions on Power
Systems, Volume: 8 , Issue: 2 , May 1993.

[9] IEEE Task Force on Load Representation for Dynamic performance, Standard
Load Models for Power Flow and Dynamic Performance Simulation, IEEE Trans-
actions on Power Systems, Volume: 10 , Issue: 3 , August 1995.

[10] IEEE Task Force on Load Representation for Dynamic performance, Bobliography
on Load Models for Power Flow and Dynamic Performance Simulation, IEEE
Transactions on Power Systems, Volume: 10 , Issue: 1 , February 1995.

[11] P. C. Krause, Analysis of Electric Machinery, McGraw-Hill, 1986.

[12] M. Pavella, D. Ernst and D. Ruiz-Vega Power System Transient Stability Analysis
and Control, Kluwer Academic Publishers, 2000.

[13] D. Ruiz-Vega and M. Pavella, A Comprehensive Approach to Transient Stability

Control: (PART I and PART II), IEEE Transactions on Power Systems ,Volume:
18 , No: 4 , Nov. 2003.

[14] IEEE Power System Engineering Committee, Eigenanalysis and frequency domain
methods for system dynamic performance, IEEE Publication 90TH0292-3-PWR,
1989.

[15] I. J. Perez-Arriaga, G. C. Verghese and F. C. Schweppe, Selective modal analysis

with application to electric power systems, PART I: Heuristic introduction, IEEE
Transactions on Power Apparatus and Systems ,Volume: PAS-101 , No: 9 , Sep.
1982.
182

[16] F. L. Pagola, I. J. Perez-Arriaga , G. C. Verghese, On Sensitivities, Residues

and participation. Application to Oscillatory Stability Analysis and Control, IEEE
Transactions on Power Systems ,Volume: 4 , No: 1 , Feb. 1989.

[17] The black-out in southern Sweden and eastern Denmark, September 23, 2003,
http://www.svk.se/upload/3813/IEEE engelskpresentation jul2003 2230.pdf

[18] Interim Report of the Investigation Committee on the 28 September 2003 Blackout
in Italy, http://www.ucte.org/pdf/News/20040427 UCTE IC Final report.pdf

[19] T. Van Cutsem and C. Vournas, Voltage Stability of Electric Power Systems,
Kluwer Academic Publishers, 1998.

[20] C. W. Taylor, Power System Voltage Stability, McGraw-Hill Inc., 1994.

[21] G. Andersson, Dynamic Phenomena in Electric Power Systems, KTH, March

2000.

[22] Supporting Document for the Network Code on Load-Frequency Control and Re-
serves, https://www.entsoe.eu/major-projects/network-code-development/load-
frequency-control-reserves/

[23] OPERATIONAL RESERVE AD HOC TEAM REPORT FINAL VER-

SION, https://www.entsoe.eu/major-projects/network-code-development/load-
frequency-control-reserves/