Chapter 12 Small-Signal Stability

small-Signal Stability
Chapter 2 provided a general introduction to the power system stability

problem, including a discussion of the basic concepts, classification, and definitions
of related terms. We will now consider in detail the various categories of system
stability, beginning with this chapter on small-signal stability. Knowledge of the
characteristics and modelling of individual system components as presented in
Chapters 3 to 11 should be helpful in this regard.
Small-signal stability, as defined in Chapter 2, is the ability of the power
system to maintain synchronism when subjected to small disturbances. In this context,
a disturbance is considered to be small if the equations that describe the resulting
response of the system may be linearized for the purpose of analysis. Instability that
may result can be of two forms: (i) steady increase in generator rotor angle due to
lack of synchronizing torque, or (ii) rotor oscillations of increasing amplitude due to
lack of sufficient damping torque. In today's practical power systems, the small-signal
stability problem is usually one of insufficient damping of system oscillations. Smallsignal analysis using linear techniques provides valuable information about the
inherent dynamic characteristics of the power system and assists in its design.
This chapter reviews fundamental aspects of stability of dynamic systems,
presents analytical techniques useful in the study of small-signal stability, illustrates
the characteristics of small-signal stability problems, and identifies factors influencing
them.
700
Small-Signal Stability
Chap. 1
12.1 FUNDAMENTAL CONCEPTS OF STABILITY

OF DYNAMIC SYSTEMS
12.1. I State-Space Representation
The behaviour of a dynamic system, such as a power system, may be described

by a set of n first order nonlinear ordinary differential equations of the following
form:
where n is the order of the system and r is the number of inputs. This can be written
in the following form by using vector-matrix notation:
where
The column vector x is referred to as the state vector, and its entries xi as state
variables. The column vector u is the vector of inputs to the system. These are the
external signals that influence the performance of the system. Time is denoted by t,
and the derivative of a state variable x with respect to time is denoted by 1.If the
derivatives of the state variables are not explicit functions of time, the system is said
to be autonomous. In this case, Equation 12.2 simplifies to
We are often interested in output variables which can be observed on the

system. These may be expressed in terms of the state variables and the input variables
in the following form:
where
s c
Set. 12.1
Fundamental Concepts of Stability of Dynamic Systems
701
The column vector y is the vector of outputs, and g is a vector of nonlinear functions
,,lating state and input variables to output variables.
he concept of state
The concept of state is fundamental to the state-space approach. The state of
a system represents the minimum amount of information about the system at any
instant in time to that is necessary so that its future behaviour can be determined
reference to the input before to.
Any set of n linearly independent system variables may be used to describe the
state of the system. These are referred to as the state variables; they form a minimal
set of dynamic variables that, along with the inputs to the system, provide a complete
description of the system behaviour. Any other system variables may be determined
from a knowledge of the state.
The state variables may be physical quantities in a system such as angle, speed,
voltage, or they may be abstract mathematical variables associated with the differential
equations describing the dynamics of the system. The choice of the state variables is
not unique. This does not mean that the state of the system at any time is not unique;
only that the means of representing the state information is not unique. Any set of
state variables we may choose will provide the same information about the system.
If we overspecify the system by defining too many state variables, not all of them will
be independent.
The system state may be represented in an n-dimensional Euclidean space
called the state space. When we select a different set of state variables to describe the
system, we are in effect choosing a different coordinate system.
Whenever the system is not in equilibrium or whenever the input is non-zero,
the system state will change with time. The set of points traced by the system state
in the state space as the system moves is called the state trajectory,
Equilibrium (or singular) points
The equilibrium points are those points where all the derivatives 21,d2, ... ,in
are simultaneously zero; they define the points on the trajectory with zero velocity.
The system is accordingly at rest since all the variables are constant and unvarying
with time.
The equilibrium or singular point must therefore satisfy the equation
where xo is the state vector x at the equilibrium point.

If the functions f;(i=1,2, ..., n) in Equation 12.3 are linear, then the system is
linear. A linear system has only one equilibrium state (if the system matrix is nonsingular). For a nonlinear system there may be more than one equilibrium point.
Chap.
The singular points are truly characteristic of the behaviour of the dynamic
system, and therefore we can draw conclusions about stability from their nature
12.1.2 Stability of a Dynamic System
The stability of a linear system is entirely independent of the input, and

state of a stable system with zero input will always return to the origin of the state
space, independent of the finite initial state.
In contrast, the stability of a nonlinear system depends on the type and
magnitude of input, and the initial state. These factors have to be taken into account
in defining the stability of a nonlinear system.
In control system theory, it is common practice to classify the stability of a
nonlinear system into the following categories, depending on the region of state space
in which the state vector ranges:
Local stability or stability in the small
Finite stability
Global stability or stability in the large
Local stability
The system is said to be locally stable about an equilibrium point if, when
subjected to small perturbation, it remains within a small region surrounding the
equilibrium point.
If, as t increases, the system returns to the original state, it is said to be
asymptotically stable in the small.
It should be noted that the general definition of local stability does not require
that the state return to the original state and, therefore, includes small limit cycles. In
practice, we are normally interested in asymptotic stability.
Local stability (i.e., stability under small disturbance) conditions can be studied
by linearizing the nonlinear system equations about the 'equilibrium point in question.
This is illustrated in the next section.
Finite stability
If the state of a system remains within a finite region R, it is said to be stable
within R. If, further, the state of the system returns to the original equilibrium point
from any point within R, it is asymptotically stable within the finite region R.
Global stability
The system is said to be globally stable if R includes the entire finite space.
Set- 12.1
Fundamental Concepts of Stability of Dynamic Systems
703
,2,1.3Linearization
We now describe the procedure for linearizing Equation 12.3. Let xo be the
initial state vector and uo the input vector corresponding to the equilibrium point
about which the small-signal performance is to be investigated. Since xo and u, satisfy
Equation 12.3, we have
Let us perturb the system from the above state, by letting
where the prefix A denotes a small deviation.

The new state must satisfy Equation 12.3. Hence,
the perturbations are assumed to be small, the nonlinear functions f(x,u) can be
expressed in terms of Taylor's series expansion. With terms involving second and
higher order powers of Ax and Au neglected, we may write
AS
Since iio= f; (xo,u,) , we obtain
with i=1,2,..., n. In a like manner, from Equation 12.4, we have
Chap. 1
12.1.4 Analysis of Stability
Lyapunov 's first method [I]

The stability in the smaN of a nonlinear system is given by the roots of the
characteristic equation of the system of first approximations, i.e., by the eigenvalues
of A:
(i)
(ii)
(iii)
When the eigenvalues have negative real parts, the original system is
asymptotically stable.
When at least one of the eigenvalues has a positive real part, the original
system is unstable.
When the eigenvalues have real parts equal to zero, it is not possible on the
basis of the first approximation to say anything in the general.
The stability in the large may be studied by explicit solution of the nonlinear
differential equations using digital or analog computers.
A method that does not require explicit solution of system differential
equations is the direct method of Lyapunov.
Lyapunov 's second method, or the direct method

The second method attempts to determine stability directly by using suitable
functions which are defined in the state space. The sign of the Lyapunov function and
the sign of its time derivative with respect to the system state equations are
considered.
The equilibrium of Equation 12.3 is stable if there exists a positive definite
function V(x,,,: ... ,xn) such that its total derivative v with respect to Equation 12.3
is not positive.
The equilibrium of Equation 12.3 is asymptotically stable if there is a positive
definite function V(xl,x,, ...,xn) such that its total derivative with respect to Equation
12.3 is negative definite.
The system is stable in that region in which v is negative semidefinite, and'
asymptotically stable if v is negative definite.2
The stability in the large of power systems is the subject of the next chapter.
This chapter is concerned with the stability in the small of power systems, and this
is given by the eigenvalues of A. As illustrated in the following section, the natural
A function is called deflnite in a domain D of state space if it has the same sign for all
x within D and vanishes for x=O. For example, V(x,, x,, x3) =x: +$ +x: is positive definite.
A function is called semidefinite in a domain D of the state space if it has the same sign
or is zero for all x within D. For example, V(x, ,x, ,x3) = (x, -x212 +x: is positive semi-definite
since it is zero for x, =x, , x, =O .
Set 12.2
707
Eigenproperties of .the State Matrix
of system response are related to the eigenvalues. Analysis of the

,igenproperties of A provides valuable information regarding the stability
characteristics of the system.
It is worth recalling that the matrix A is the Jacobian matrix whose elements
are given by the partial derivatives afilx, evaluated at the equilibrium point about
Ghich the small disturbance is being analyzed. This matrix is commonly referred to
the state matrix or the plant matrix. The term "plant" originates from the area of
process control and is entrenched in control engineering vocabulary. It represents that
Part of the system which is to be controlled.
,,
12.2 EIGENPROPERTIES OF THE STATE MATRIX

12.2.1 Eigenvalues
The eigenvalues of a matrix are given by the values of the scalar parameter h
for which there exist non-trivial solutions (i.e., other than = 0 ) to the equation
A+
A+
(12.16)
where
A is an nxn matrix (real for a physical system such as a power system)
is an nx 1 vector
To find the eigenvalues, Equation 12.16 may be written in the form
For a non-trivial solution
det(A-AI)
Expansion of the determinant gives the characteristic equation. The n solutions of

h =A,,h,
..., h,, are eigenvalues of A.
The eigenvalues may be real or complex. If A is real, complex eigenvalues
always occur in conjugate pairs.
Similar matrices have identical eigenvalues. It can also be readily shown that
a matrix and its transpose have the same eigenvalues.
12.2.2 Eigenvectors
For any eigenvalue hi, the n-column vector 4, which satisfies Equation 12.16
is called the right eigenvector of A associated with the eigenvalue hi.Therefore, we
have
A+i
The eigenvector
api
Chap, 12
i=ly2,...yn
(12.19)
bi has the form
Since Equation 12.17 is homogeneous, k9i (where k is a scalar) is also a solution.

Thus, the eigenvectors are determined only to within a scalar multiplier.
Similarly, the n-row vector q iwhich satisfies
is called the lefr eigenvector associated with the eigenvalue hi.

The left and right eigenvectors corresponding to different eigenvalues are
orthogonal. In other words, if hi is not equal to A,
However, in the case of eigenvectors corresponding to the same eigenvalue,
$ 4
Ci
(12.22)
where Ciis a non-zero constant.

Since, as noted above, the eigenvectors are determined only to within a scalar
multiplier, it is common practice to normalize these vectors so that
Vibi = 1
(12.23)
I 2.2.3 Modal Matrices

In order to express the eigenproperties of A succinctly, it is convenient to
introduce the following matrices:
diagonal matrix, with the eigenvalues Al, A2, ..., An

as diagonal elements
Set 12.2
Eigenproperties of th,e State Matrix
709
~~~h of the above matrices is nxn. In terms of these matrices, Equations 12.19 and
12.23 may be expanded as follows.
A@
~tfollows
@A
(12.27)
from Equation 12.27

@-l~@
= A
12.2.4 Free Motion of a Dynamic System

Referring to the state equation 12.9, we see that the free motion (with zero
input) is given by
Ax
AAx
(12.30)
A set of equations of the above form, derived porn physical considerations,

is often not the best means of analytical studies of motion. The problem is that the
rate of change of each state variable is a linear combination of all the state variables.
As the result of cross-coupling between the states, it is difficult to isolate those
parameters that influence the motion in a significant way.
In order to eliminate the cross-coupling between the state variables, consider
a new state vector z related to the original state vector Ax by the transformation
where
is the modal matrix of A defined by Equation 12.24. Substituting the above
expression for Ax in the state equation (12.30), we have
The new state equation can be written as
In view of Equation 12.29, the above equation becomes
The important difference between Equations 12.34 and 12.30 is that A is a diagonal
matrix whereas A, in general, is non-diagonal.
set. 12.2
Eigenproperties of the State Matrix
In other words, the time response of the ith state variable is given by
The above equation gives the expression for the free motion time response of the
system in terms of the eigenvalues, and left and right eigenvectors.
Thus, the free (or initial condition) response is given by a linear combination
o , n dynamic modes corresponding to the n eigenvalues of the state matrix.
The scalar product ci= qiAx(0) represents the magnitude of the excitation of
the ith mode resulting from the initial conditions.
If the initial conditions lie along the jth eigenvector, the scalar products
y.~x(O)
for all i#j are identically zero. Therefore, only the jth mode is excited.
1
If the vector representing the initial condition is not an eigenvector, it can be
represented by a linear combination of the n eigenvectors. The response of the system
will be the sum of the n responses. If a component along an eigenvector of the initial
conditions is zero, the corresponding mode will not be excited (see Example 12.1 for
an illustration).
Eigenvalue and stability
The time dependent characteristic of a mode corresponding to an eigenvalue

hi is given by e"". Therefore, the stability of the system is determined by the
eigenvalues as follows:
(a)
A real eigenvalue corresponds to a non-oscillatory mode. A negative real

eige~ivaluerepresents a decaying mode. The larger its magnitude, the faster the
decay. A positive real eigenvalue represents aperiodic instability.
The values of c's and the eigenvectors associated with real eigenvalues are
also real.
(b)
Complex eigenvalues occur in conjugate pairs, and each pair corresponds to an

oscillatory mode.
The associated c's and eigenvectors will have appropriate complex values so
as to make the entries of x(t) real at every instant of time. For example,
has the form
714
Chap.
Cases (I), (3) and (5) ensure local stability, with (1) and (3) being asymptotically
stable.
12.2.5 Mode Shape, Sensitivity, and Participation Factor
(a) Mode shape and eigenvectors

In the previous section, we discussed the system response in terms of the state
vectors Ax and z, which are related to each other as follows:
and
The variables Axl, Ax,, ..., Axn are the original state variables chosen to represent the
dynamic performance of the system. The variables z,,z,, ...,z, are the transformed state
variables such that each variable is associated with only one mode. In other words,
the transformed variables z are directly related to the modes.
From Equation 12.47A we see that the right eigenvector gives the mode shape,
i.e., the relative activity of the state variables when a particular mode is excited. For
example, the degree of activity of the state variable xk in the ith mode is given by the
element 4ki of the right eigenvector gi.
The magnitudes of the elements of @igive the extents of the activities of the
n state variables in the ith mode, and the angles of the elements give phase
displacements of the state variables with regard to the mode.
As seen from Equation 12.47B, the left eigenvector q, identifies which
combination of the original state variables displays only the ith mode. Thus the kth
element of the right eigenvector Oimeasures the activity of the variable xk in the ith
mode, and the Ath element of the left eigenvector q iweighs the contribution of this
activity to the ith mode.
(b) Eigenvalue sensitivity
Let us now examine the sensitivity of eigenvalues to the elements of the state
matrix. Consider Equation 12.19 which defines the eigenvalues and eigenvectors:
see. 12.2
71 5
Differentiating with respect to a&(the element of A in kth row and jth column) yields
prernultiplying by @ i , and noting that qi+i

above equation simplifies to
=1
and
I& (A-IiI) = 0, we see that the
~ lelements
l
of dAlaa&are zero, except for the element in the kth row and jth column
which is equal to 1. Hence,
Thus the sensitivity of the eigenvalue hi to the element a,, of the state matrix is equal
to the product of the left eigenvector element vikand the right eigenvector element
4ji.
(c) Participation factor

One problem in using right and left eigenvectors individually for identifying
the relationship between the states and the modes is that the elements of the
eigenvectors are dependent on units and scaling associated with the state variables. As
a solution to this problem, a - matrix called the participation matrix (P), which
combines the right and left eigenvectors as follows is proposed in reference 2 as a
measure of the association between the state variables and the modes.
with
71 6
Chap. 1
where
the element on the kth row and ith column of the modal matrix g
= kth entry of the right eigenvector Oi
yik= the element on the ith row and kth column of the modal matrix rp
= kth entry of the left eigenvector qi
+ki =
The element pki=$kiyik

is termed the participation factor [2].It is a measure
of the relative participation of the kth state variable in the ith mode, and vice versa.
Since +ki measures the activity of xk in the ith mode and yik weighs the
contribution of this activity to the mode, the product pki measures the net
participation. The effect of multiplying the elements of the left and right eigenvectors
is also to make pkidimensionless (i.e., independent of the choice of units).
In view of the eigenvector normalization, the sum of the participation factors
n
i=l
k= 1
associated with any mode ( xp,) or with any state variable ( C p,) is equal to 1.
From Equation 12.48, we see that the participation factor pkiis actually equal
to the sensitivity of the eigenvalue hi to the diagonal element akkof the state matrix
A
As we will see in a number of examples in this chapter, the participation

factors are generally indicative of the relative participations of the respective states
in the corresponding modes.
12.2.6 Controllability and Observability
In Section 12.1.3 the system response in the presence of input was given as
Equations 12.8 and 12.9 and is repeated here for reference.
Expressing them in terms of the transformed variables z defined by Equation 12.31

yields
set. 12.2
71 7
The state equations in the "normal form" (decoupled) may therefore be written as
Ay
C'z + DAu
~ ~ f e r r i ntog Equation 12.51, if the ith row of matrix B' is zero, the inputs have no
effect on the ith mode. In such a case, the ith mode is said to be uncontrollable.
From Equation 12.52, we see that the ith column of the matrix C' determines
or not the variable zi contributes to the formation of the outputs. If the
column is zero, then the corresponding mode is unobservable. This explains why
some poorly damped modes are sometimes not detected by observing the transient
,response of a few monitored quantities.
The nxr matrix B' = 8 - I B is referred to as the mode controllability matrix, and
the mxn matrix C' =C@ as the mode observability matrix.
By inspecting B' and C' we can classify modes into controllable and
observable; controllable and unobservable; uncontrollable and observable;
uncontrollable and unobservable.
'
12.2.7 The Concept of Complex Frequency
Consider the damped sinusoid
The unit of a is radians per second and that of 8 is radians. The dimensionless unit
neper (Np) is commonly used for a t in honour of the mathematician John Napier
(1550-16 17) who invented logarithms. Thus the unit of a is neper per second (Npls).
For circuits in which the excitations and forced functions are damped
sinusoids, such as that given by Equation 12.55, we can use phasor representations
of damped sinusoids. This will work as well as the phasors of (undamped) sinusoids
normally used in ac circuit analysis because the properties of sinusoids that make the
phasors possible are shared by damped sinusoids. That is, the sum or difference of
two or more damped sinusoids is a damped sinusoid and the derivative or indefinite
,
This is referred to as Kalman's canonical structure theorem, since it was first proposed
by R.E. Kalman in 1960.
71 8
Chap. 1 2
integral of a damped sinusoid is also a damped sinusoid. In all these cases, vm and
8 may change; o and o are fixed.
Analogous to the form of phasor notation used for sinusoids, in the case of
damped sinusoids, we may write
Vmeutcos(ot+B)
Re[vme Ot e
- R~ [v,eje
""I
(0
+ja)t]
With s =o+jo, we have
is the phasor (V'L8) and is the same for both the undamped and damped
where
sinusoids. Obviously, we may treat the damped sinusoids the same way we do
undamped sinusoids by using s instead of j o .
Since s is a complex number, it is referred to as complex frequency, and V(')
is called a generalized phasor.
All concepts such as impedance, admittance, Thevenin's and Norton's
theorems, superposition, etc., carry over to the damped sinusoidal case.
It follows that, in the s-domain, the phasor current I(s) and voltage V(s),
associated with a two-terminal network are related by
where Z(s) is the generalized impedance.

Similarly, input and output relations of dynamic devices can be expressed as
In the factored form,
The numbers z, ,z2, ..., z, are called the zeros because they are values of s for which
G(s) becomes zero. The numbers p, ,p2, ... ,p, are called the poles of G(s). The values
set. 12.2
71 9
and zeros, along with a, and b,, uniquely determine the system transfer
function G(s). Poles and zeros are useful in considering frequency domain properties
dynamic systems.
of
12.2.8 Relationship between Eigenproperties and Transfer Functions
The state-space representation is concerned not only with input and output
properties of the system but also with its complete internal behaviour. In contrast, the
rnsfer function representation specifies only the inputloutput behaviour. Hence, one
can make an arbitrary selection of state variables when a plant is specified only by
a transfer function. On the other hand, if a state-space representation of a system is
known, the transfer function is uniquely defined. In this sense, the state-space
is a more complete description of the system; it is ideally suited for the
analysis of multi-variable multi-input and multi-output systems.
For small-signal stability analysis of power systems, we primarily depend on
the eigenvalue analysis of the system state matrix. However, for control design we are
interested in an open-loop transfer function between specific variables. To see how
this is related to the state matrix and to the eigenproperties, let us consider the transfer
function between the variables y and u. From Equations 12.8 and 12.9, we may write
where A is the state matrix, Ax is the state vector, Au is a single input, Ay is a single
output, c is a row vector and b is a column vector. We assume that y is not a direct
function of u (i.e., D=O).
The required transfer function is
This has the general form
If D(s) and N(s) can be factored, we may write
720
Chap. 12
As discussed in Section 12.2.7, the n values of s, namely, p1,p2,...,P,, which make the
denominator polynomial D(s) zero are the poles of G(s). The 1 values of s, namely,
z,, z,, ..., zl, are the zeros of G(s).
Now, G(s) can be expanded in partial fractions as
and Riis known as the residue of G(s) at pole pi.

To express the transfer function in terms of the eigenvalues and eigenvectors,
we express the state variables Ax in terms of the transformed variables z defined by
Equation 12.31. Following the procedure used in Section 12.2.4, Equations 12.57 and
12.58 may be written in terms of the transformed variables as
and
Hence,
Since A is a diagonal matrix, we may write
where
Ri
c@,qib
We see that the poles of G(s) are given by the eigenvalues of A. Equation 12.67 gives
the residues in terms of the eigenvectors. The zeros of G(s) are given by the solution
of
Set,
12.2
In this example we will study a second-order linear system. Such a system is easy to
analyze and is helpful in understanding the behaviour of higher-order systems. The
performance of high-order systems is often viewed in terms of a dominant set of
second-order poles or eigenvalues. Therefore, a thorough understanding of the
characteristics of a second-order system is essential before we study complex systems.
Figure E12.1 shows the familiar RLC circuit, which represents a second-order system.
Study the eigenproperties of the state matrix of the system and examine its modal
characteristics.
Figure E12.1
Solution
The differential equation relating v, to vi is
This may be written in the standard form
where
on = I/@
undamped natural frequency
( = (~/2)/m
= damping ratio
Chap.
In order to develop the state-space representation, we define the following state, input
and output variables:
Using the above quantities, Equation E12.2 can be expressed in t a m s of two firstorder equations:
In matrix form,
The output variable is given by
These have the standard state-space form:
Ax+bu
cx+du
The eigenvalues of A are given by
Hence,
12.2
Solving for the eigenvalues, we have
The right eigenvectors are given by

(A-AI)@
Therefore,
This may be rewritten as
If we attempt to solve the above equations for $ l i and $2i, we realize that they are not
independent. As discussed earlier, this is true in general; for an nth order system, the
equation (A- AI) @ = 0 gives only n - 1 independent equations for the n components of
eigenvectors. One component of the eigenvector may be fixed arbitrarily and then the
other components can be determined from the n-1 independent equations. It shodd,
however, be noted that the eigenvectors themselves are linearly independent if
eigenvalues are distinct.
For the second-order system, we can fix $li=land determine $2i, from one of the two
relationships in Equation E12.10, for each eigenvalue.
The eigenvector corresponding to 3L1 is
and the eigenvector corresponding to 3L2 is
724
Chap. 1
The nature of the system response depends almost entirely on the damping rati
O C;.
The value of onhas the effect of simply adjusting the time scale.
<
If is greater than 1, both eigenvalues are real and negative; if 5 is equal to 1, both
eigenvalues are equal to -a,; and if 6 is less than 1, eigenvalues are complex
conjugates, given as
The location of the eigenvalues in the complex plane with respect to

indicated in Figure E12.2.
< and anis
Damping angle 0
= cos-lr
Figure E12.2
We will first examine the singularities of the second-order system and discuss the
shape of the state trajectories near the singularity. We will then discuss in detail the
case where both eigenvalues are real and negative, with h, greater than A], but with
A, and A2 not far different.
The state equations in the normal form are given by
Hence,
Chap.
transformation x = O z .
If the input vi is zero, and if the initial conditions are such that (xl, x2) is on one of
the eigenvectors, the state vector will remain in the same direction but will vary in
magnitude by the factor e"l' or eh* as the case may be.
49
If the vector representing the initial condition is not an eigenvector, it can be

represented by a linear combination of the two eigenvectors. The response of the
circuit will be the sum of the two responses. As time increases, the component in the
direction of the eigenvector 42 becomes less significant because eht decays faster
than eal'. Thus the trajectories always approach the origin along the 4, direction
unless the component of this eigenvector was initially zero. If the eigenvector~are not
real, such a simple physical interpretation of eigenvectors is not possible.
A, and A, real and negative;

1121>
k
\\,
22
11
higenvector 4,
(slow decay)
~ i ~ e n v e c t o4,r
(fast decay)
Figure E12.4
12.2.9 Computation of Eigenvalues
In the above example, we computed the eigenvalues by solving the

characteristic equation of the system. This was possible because we were analyzing
a simple second-order system. For higher-order systems with eigenvalues of widely
differing magnitudes, this approach fails. The method that has been widely used for
the computation of eigenvalues of real non-symmetrical matrices is the QR
transformation method originally developed by J.G.F. Francis [3]. The method is
numerically stable, robust, and converges rapidly. It is used in a number of very good
general purpose commercial codes and has been successfully used for analyzing smallsignal stability of power systems with several hundred states. The right eigenvectors
may be computed by using the inverse iteration technique. A good description of the
QR transformation and inverse iteration methods may be found in reference 4.
Set. 12.3
727
Single-Machine Infinite Bus System
For large systems involving several thousand states, the QR method cannot be
for computing the eigenvalues. The reason for this and a description of special
for eigenvalue analysis of very large systems are presented in Section 12.8.
12.3 SMALL-SIGNAL STABILITY OF A SINGLE-MACHINE

INFINITE BUS SYSTEM
In this section we will study the small-signal performance of a single machine
connected to a large system through transmission lines. A general system
configuration is shown in Figure 12.3(a). Analysis of systems having such simple
configurationsis extremely useful in understanding basic effects and concepts. After
we develop an appreciation for the physical aspects of the phenomena and gain
experience with the analytical techniques, using simple low-order systems, we will be
in a better position to deal with large complex systems.
Large I
system Ii
(a) General configuration
1-
Infinite bus
Zeq= RE+jXE
(b) Equivalent system
Figure 12.3 Single machine connected to a large system
through transmission lines
For the purpose of analysis, the system of Figure 12.3(a) may be reduced to
the form of Figure 12.3(b) by using Thtvenin's equivalent of the transmission
network external to the machine and the adjacent transmission. Because of the relative
size of the system to which the machine is supplying power, dynamics associated with
the machine will cause virtually no change in the voltage and frequency of Thevenin's
Chap. 1 , ~
voltage EB. Such a voltage source of constant voltage and constant frequency is
referred to as an infinite bus.
For any given system condition, the magnitude of the infinite bus voltage E
B
remains constant when the machine is perturbed. However, as the steady-state system
conditions change, the magnitude of EB may change, representing a changed operating
condition of the external network.
@
In what follows we will analyze the small-signal stability of the system of
Figure 12.3(b) with the synchronous machine represented by models of varying
degrees of detail. We will begin with the classical model and gradually increase the
model detail by accounting for the effects of the *dynamics of the field circuit,
excitation system, and amortisseurs. In each case, we will develop the expressions for
the elements of the state matrix as explicit functions of system parameters. This will
help make clear the effects of various factors associated with a synchronous machine
on system stability. In addition to the state-space representation and modal analysis,
we will use the block diagram representation and torque-angle relationships to analyze
the system-stability characteristics. The block diagram approach was first used by
Heffron and Phillips [5] and later by deMello and Concordia [6] to analyze the smallsignal stability of synchronous machines. While this approach is not suited for a
detailed study of large systems, it is useful in gaining a physical insight into the
effects of field circuit dynamics and in establishing the basis for methods of
enhancing stability through excitation control.
12.3.1 Generator Represented by the Classical Model
With the generator represented by the classical model (see Section 5.3.1) and
all resistances neglected, the system representation is as shown in Figure 12.4.
Here E' is the voltage behind Xj.Its magnitude is assumed to remain constant
at the pre-disturbance value. Let 6 be the angle by which E' leads the infinite -bus
voltage EB. As the rotor oscillates during a disturbance, 6 changes.
With E' as reference phasor,
Figure 12.4
set. 12.3
The complex power behind Xj is given by
With stator resistance neglected, the air-gap power (P,) is equal to the terminal power
(p). In per unit, the air-gap torque is equal to the air-gap power (see Section 5.1.2).
Hence,
Linearizing about an initial operating condition represented by 6=a0yields
The equations of motion (Equations.3.209 and 3.2 10 of Chapter 3) in per unit

are
where Am, is the per unit speed deviation, 6 is rotor angle1 in electrical radians, w,
is the base rotor electrical speed in radians per second, and p is the differential
operator dldt with time t in seconds.
Linearizing Equation 12.73 and substituting for AT, given by Equation 12.72,
we obtain
where
K' is the synchronizing torque coefficient given by
As discussed in Section 5.3.1, for a classical generator model, the angle of E' with
respect to a synchronously rotating reference phasor can be used as a measure of the rotor
angle. Here we have chosen EB as the reference, and the rotor angle 6 is measured as the angle
by which E' leads EB.
Chap, 12
Linearizing Equation 12.74, we have

pA6
w0Aq
Writing Equations 12.75 and 12.77 in the vector-matrix form, we obtain
This is of the form x =Ax+bu. The elements of the state matrix A are seen to be
dependent on the system parameters KD, H, X ,and the initial operating condition
represented by the values of E' and Fo. The block diagram representation shown in
Figure 12.5 can be used to describe the small-signal performance.
From the block diagram of Figure 12.5, we have
Rearranging, we get
OO
s2(A6) + -KD
s(A~)+- 4 o , ( A ~ ) = AT,
2H
2H
2H
Therefore, the characteristic equation is given by
This is of the general form
Set 12.3
Synchronizing torque
component
Ks
ATe
AT,,
w,
Amr
2Hs
Damping torque
component
Ks
KD
H
Am,
A6
"0
A8
KD
synchronizing torque coefficient in pu torquehad

= damping torque coefficient in pu torque/pu speed deviation
= inertia constant in MW-s/MVA
= speed deviation in pu = (m, -mo)/wo
= rotor angle deviation in elec. rad
= Laplace operator
= rated speed in elec. rad/s = 2nfo
= 377 for a 60 Hz system
=
Figure 12.5 Block diagram of a single-machine infinite

bus system with classical generator model
Therefore, the undamped natural frequency is
and the damping ratio is
732
Chap. 12
As the synchronizing torque coefficient Ks increases, the natural frequency increases

and the damping ratio decreases. An increase in damping torque coefficient K
increases the damping ratio, whereas an increase in inertia constant decreases both D
and 6.
Example 12.2
Figure E12.5 shows the system representation applicable to a thermal generating
station consisting of four 555 MVA, 24 kV, 60 Hz units.
HT
/
/
j0.5
CCT 1
5 Infinite
bus
/
/
j0.93
CCT2
5
/
Figure E12.5
The network reactances shown in the figure are in per unit on 2220 MVA, 24 kV base
(referred to the LT side of the step-up transformer). Resistances are assumed to be
negligible.
The objective of this example is to an'alyze the small-signal stability characteristics
of the system about the steady-state operating condition following the loss of circuit
2. The postfault system condition in per unit on the 2220 MVA, 24 kV base is as
follows:
0.9
0.3 (overexcited)
E,
1.OL36"
EB = 0.99510"
The generators are to be modelled as a single equivalent generator represented by the

classical model with the following parameters expressed in per unit on 2220 MVA,
24 kV base:
(a)
Write the linearized state equations of the system. Determine the eigenvalues,
damped frequency of oscillation in Hz, damping ratio and undamped natural
frequency for each of the following values of damping coefficient (in pu
torquelpu speed):
!
(b)
For the case with KD=lO.O, find the left and right eigenvectors, and
participation matrix. Determine the time response if at t=O, A6=5" and A@=O.
12.3
Solution
(a) Figure E12.6 shows the circuit model representing the postfault steady-state
operating condition with all parameters expressed in per unit on 2220 MVA base.
Figure E12.6
With Et as reference phasor, the generator stator current is given by
The voltage behind the transient reactance is
The angle by which E' leads EB is
The total system reactance is
The corresponding synchronizing torque coefficient, from Equation 12.76, is
Linearized system equations are
The eigenvalues of the state matrix are given by
or
12+0 . 1 4 3 +~40.79
~ ~
This is of the form

2
h 2 + 2 [ o n h + o n= 0
with
on = d
m = 6.387 radls = 1.0165 Hz
= 0.143Kd(2~6.387)= 0.01 12KD
The eigenvalues are
The damped frequency is
The following are the required results for different values of KD.
Chap, 1
Set. 12.3
735
KD
10
- 10
Eigenvalues h
Damped frequency cod
Damping ratio 6
Undamped natural frequency con
096.39
1.0165 Hz
0
1.0165 Hz
-0.71496.35
1.0101 Hz
0.112
1.0165 Hz
0.714g6.36
1.0101 Hz
-0.1 12
1.0165 Hz
(b) The right eigenvectors are given by
For the given system, with KD=lO,the above equation becomes
For h = -0.714+j6.35, the corresponding equations are
The above equations are not linearly independent. As discussed in Example 12.1,one
of the eigenvectors corresponding to an eigenvalue has to be set arbitrarily. Therefore,
let
412, =
1'0
then
@,,= -0.0019+j0.0168
Similarly, eigenvectors corresponding to
412,
1.0
& = -0.714 -j6.35 are
@,,= -0.0019-j0.0168
The right eigenvector modal matrix is
The left eigenvectors normalized so that
@i@i = 1.0
The participation matrix is
The time response is given by
With A6 =5" =0.0873 rad and Ao,=O at t=O, we have
are given by
Chap,
Set . 12.3
The time response of speed deviation is
Similarly, the time response of rotor angle deviation is

A6(t) = 0.088e -0.714' cos(6.35 t- 0.112)
rad
This is a second-order system with an oscillatory mode of response having a damped

frequency of 6.35 radls or 1.0101 Hz. The oscillations decay with a time constant of
110.714 s. This corresponds to a damping ratio 6 of 0.1 12. As this is a rotor angle
mode, Am, and A6 participate in it equally.
12.3.2 Effects of Synchronous Machine Field Circuit Dynamics
We now consider the system performance including the effect of field flux
variations. The arnortisseur effects will be neglected and the field voltage will be
assumed constant (manual excitation control).
In what follows, we will develop the state-space model of the system by first
reducing the synchronous machine equations to an appropriate form and then
combining them with the network equations. We will express time in seconds, angles
in electrical radians, and all other variables in per unit.
Synchronous machine equations

As in the case of the classical generator model, the acceleration equations are
where oo=2xf0 elec. radls. In this case, the rotor angle 6 is the angle (in elec. rad) by
which the q-axis leads the reference EB. As shown in Figure 12.6, the rotor angle 6
is the sum of the internal angle 6i(see Section 3.6.3) and the angle by which E, leads
4.We need a convenient means of identifying the rotor position with respect to an
appropriate reference and keeping track of it as the rotor oscillates. As discussed in
q-axis
Chap.
12
'\my
//
d-axis
Figure 12.6
Chapter 3 (Section 3.6), the q-axis offers this convenience when the dynamics of rotor
circuits are represented in the machine model. The choice of EB as the reference for
measuring rotor angle is convenient from the viewpoint of solution of network
equations.
The per unit synchronous machine equations were summarized in Section 3.4.9
and the simplifications essential for large-scale stability studies were discussed in
Section 5.1. From Equation 5.10, with time t in seconds instead of per unit, the field
circuit dynamic equation is
where Efd is the exciter output voltage defined in Section 8.6.1. Equations 12.83 to
12.85 describe the dynamics of the synchronous machine with Am,, 6, and yp as the
state variables. However, the derivatives of these state variables appear in these
equations as functions of T, and $d, which are neither state variables nor input
variables. In order to develop the complete system equations in the state-space form,
we need to express ifd and T, in terms of the state variables as determined by the
machine flux linkage equations and network equations.
With amortisseurs neglected, the equivalent circuits relating the machine flux
linkages and currents are as shown in Figure 12.7.
12.3
Figure 12.7
The stator and rotor flux linkages are given by
i
= @ad +Lfdfd
In the above equations yadand vaqare the air-gap (mutual) flux linkages, and
and La,, are the saturated values of the mutual inductances.
From Equation 12.88, the field current may be expressed as
The d-axis mutual flux linkage can be written in terms of yfdand id as follows:
740
Chap, 1
where
Since there are no rotor circuits considered in the q-axis, the mutual flux linkage is
given by
qaq =
The air-gap torque is
With pyl terms and speed variations neglected as discussed in Section 5.1, the stator
voltage equations are
As a first step, we have expressed i and T, in terms of yl-, id, i, vadand yaq.
Sd
In addition, ed and e, have been expressed in terms of these variables and will be used
in conjunction with the network equations to provide expressions for id and iq in terms
-.
of the state variables.
The advantages of using w,, and yl, as intermediate variables in the
elimination process will be more apparent when we account for the effects of
amortisseur circuits in Section 12.6.
Network equations
Since there is only one machine, the machine as well as network equations can
be expressed in terms of one reference frame, i.e., the d-q reference frame of the
machine. Referring to Figure 12.6, the machine terminal and infinite bus voltages in
terms of the d and q components are
Set- 12.3
The network constraint equation for the system of Figure 12.3(b) is
E
(ed+jeq)
E,+(R,+~x,)~,
(EBd+jEBq) (RE+jXE)(id+jiq
+
Resolving into d and q components gives

ed
REid-xEiq+EBd
eq
REiq+XEid+EBq
where
Using Equations 12.94 and 12.95 to eliminate ed, eq in Equations 12.99 and 12.100,
and using the expressions for vadand yr,g given by Equations 12.90 and 12.92, we
obtain the following expressions for id and iq in terms of the state variables yfd and
6:
742
Chap.
where
The reactances X, and XAs are saturated values. In per unit they are equal to the
corresponding inductances.
Equations 12.103 and 12.104, together with Equations 12.89, 12.90 and 12.92
can be used to eliminate qdand T, from the differential equations 12.83 to 12.85 and
express them in terms of the state variables. These equations are nonlinear and have
to be linearized for small-signal analysis.
Linearized system equations
Expressing Equations 12.1 03 and 12.1 04 in terms of perturbed values, we may

write
where
By linearizing Equations 12.90 and 12.92, and substituting in them the above
expressions for Aid and Ai,, we get
set. 12.3
Linearizing Equation 12.89 and substituting for Ayrad from Equation 12.109 gives
The linearized form of Equation 12.93 is
Substituting for Aid, Ai,, d y a d , and Ay,, from Equations 12.106 to 12.110, we obtain
ATe
KlA8
(12.1 12)
where
By linearizing Equations 12.83 to 12.85 and substituting the expressions for A$ and
AT, given by Equations 12.111 and 12.112, we obtain the system equations in the
desired final form:
Chap.
where
and AT, and AEfd depend on prime-mover and excitation controls. With constant
mechanical input torque, AT, =O; with constant exciter output voltage, AEfd=O.
It is interesting to compare the above state-space equations with those derived
in Section 12.3.1 by assuming the classical generator model (which is equivalent to
assuming Rfd=0, Ra =O and X,=Xi).
The mutual inductances Lads and Laps in the above equations are saturated
values. The method of accounting for saturation for small-signal analysis is described
below.
Representation of saturation in small-signal studies
Since we are expressing small-signal performance in terms of perturbed values

of flux linkages and currents, a distinction has to be made between total saturation
~ e c12.3
Bnd incremental saturation.

Total saturation is associated with total values of flux linkages and currents.
The method of accounting for total saturation was discussed in Section 3.8.
Incremental saturation is associated with perturbed values of flux linkages and
,urent~. Therefore, the incremental slope of the saturation curve is used in computing
the incremental saturation as shown in Figure 12.8.
Denoting the incremental saturation factor
we have
Lads (incr)
Based on the definitions of A , B,
)
-
Ksd(incr)
and yr,
LL?m(
in Section 3.8.2, we can show that

1
1+ Bsat Asat
Bmt(dm-*~~)
A similar treatment applies to q-axis saturation.

For computing the initial values of system variables (denoted by subscript o),
total saturation is used. For relating the perturbed values, i.e., in Equations 12.105,
12.108, 12.113, 12.114, and 12.116, the incremental saturation factor is used.
The method of computing the initial steady-state values of machine parameters
was described in Section 3.6.5.
Slope represents saturated value of

Lad relating total values of v and I
\ value
Slope represents saturated
of Lad relating incremental
values of
I
I
I
I
Av and AI
Ijd or mmf
Figure 12.8 Distinction between incremental and total saturation
Chap. l 2
Summary of procedure for formulating the state matrix

(a)
The following steady-state operating conditions, machine parameters and

network parameters are given:
+&
pt
Qt
Et
Ld
Lq
RE
XE
Ra
Lfd
Rfd
Asat
'sat
WTI
Alternatively EB may be specified instead of Qt or El.

(b)
The first step is to compute the initial steady-state values of system variables:
It,
power factor angle @
Total saturation factors Ksd and Ksq (see Section 3.8)
xds
= Lds = KsdLadu+L~
Xqs = '9s = Ksq Lagu +Lz
ItXqscosQ- I,Ra sin@
6, = tan-'
= tan-'
Et+ ItRaCOSQ+ ZtXqssin@
(-1
EBdO
EBq0
(c)
The next step is to compute incremental saturation factors and the

corresponding saturated values of La,,, Lags, LLds, and then
R ,
XQ7X T D
~
m l , m 2 , "1,
4 , K2
n2
from Equation 12.105

from Equation 12.108
fiom Equations 12.113 and 12.114
single-~achineInfinite Bus System
set. 12.3
(4
Finally, compute the elements of matrix A fiom Equation 12.116.
~ ~ o diagram
ck
representation
Figure 12.9 shows the block diagram representation of the small-signal
of the system. In this representation, the dynamic characteristics of the
Performance
are expressed in terms of the so-called K constants [5]. The basis for the block
diagram and the expressions for the associated constants are developed below.
Field circuit
Figure 12.9 Block diagram representation with constant Efd
From Equation 12.112, we may express the change in air-gap torque as a

function of A6 and Ayfd as follows:
AT,
KlA6 + K2A$fd
where
K,
K2
AT,/AF with constant yfd

= ATe/Ayfd with constant rotor angle 6
=
The expressions for Kl and K2 are given by Equations 12.113 and 12.114.
The component of torque given by K,A6 is in phase with A6 and hence
represents a synchronizing torque component.
The component of torque resulting fiom variations in field flux linkage is
given by K2AyThe variation of vfdis determined by the field circuit dynamic equation:
748
Chap. 1
By grouping terms involving Ayfd and rearranging, we get
where
Equation 12.1 19, with s replacing p, accounts for the field circuit block in Figure
12.9.
Expression for the K constants in the expanded form
We have expressed the K constants in terms of the elements of matrix A. In

the literature [5,6], they are usually expressed explicitly in terms of the various system
parameters, as summarized below.
The constant Kl was expressed in Equation 12.1 13 as
= nl($ado+Laqsido) - m l ( $ a q o + L ~ i q O )
From Equation 12.95, the first term in parentheses in the above expression for Kl may
be written as
where Eqo is the predisturbance value of the voltage behind R,+jX,. The second term
in parentheses in the expression for K , may be written as
@aqo +
-L aqs iq0 +L' iq0
Substituting for nl, ml fi-om Equation 12.108 and for the terms given by Equations
12.121 and 12.122 in the expression for K,, yields
Set. 12.3
similarly, the expanded form of the expression for the constant K, is
From Equations 12.91, 12.108, and 12.116, we may write
+-
a33 = -o0-
Ld
'fd
'ah
(La&
Lfd) ('ah
'fd)
Substitution of the above in the expression for K3 and Tj given by Equation 12.120
yields
where T&, is the saturated value of T&, Similarly, from ~ ~ u a t i o n12.91,

s'
12.108 and
Chap. 12,
12.116, we may write
Substitution of the above in the expression for K4 given by Equation 12.120 yields
If the effect of saturation is neglected, this simplifies to
If the elements of matrix A are available, the K constants may be computed directly
from them. The expanded forms are derived here to illustrate the form of expressions
used in the literature. An advantage of these expanded forms is that the dependence
of the K constants on the various system parameters is more readily apparent. A
disadvantage, however, is that some inconsistencies appear in representing saturation
effects.
In the literature, Ei=(LadILld)~fd is often used as a state variable instead of vfl
(see Section 5.2). The effect of this is to remove the Lad/(Lad+Lfd)term from the
expressions for K2 and K3. The product K2K3 would, pqwever,
\
remain the same.
Effect of field Jlux linkage variation on system stability

We see from the block diagram of Figure 12.9 that, with constant field voltage
(Mfd=O), the field flux variations are caused only by feedback of A6 through the
coefficient K4. This represents the demagnetizing effect of the armature reaction.
The change in air-gap torque due to field flux variations caused by rotor angle
changes is given by
A s 1 due to A%d
1+ST,
The constants K2, K3, and K4 are usually positive. The contribution of Avfd to
synchronizing and damping torque components depends on the oscillating frequency
as discussed below.
(a)
In the steady state and at very low oscillating frequencies (s=jo -+ 0):
'
set. 12.3
ATe due to Aqjd
-KZK3K4A6
The field flux variation due to A6 feedback (i.e., due to armature reaction)
introduces a negative synchronizing torque component. The system becomes
monotonically unstable when this exceeds K1A6.The steady-state stability limit
is reached when
(b)
At oscillating frequencies much higher than 1/T3:
ATe = -
K K K
4 ~ 6
j0T3
-
K K K
4 i ~ 6
Thus, the component of air-gap torque due to Ayfd is 90' ahead of A6 or in

phase with do. Hence, Avfd results in a positive damping torque component.
(c)
At typical machine oscillating frequencies of about 1 Hz (27-c radls), Avfd

results in a positive damping torque component and a negative synchronizing
torque component. The net effect is to reduce slightly the synchronizing torque
component and increase the damping torque component.
Figure 12.10 Positive damping torque and negative

synchronizing torque due to KzAvfd
Chap. 1
Special situations with -K4 negative:

The coefficient K4 is normally positive. As long as it is positive, the effect of
field flux variation due to armature reaction
with,oonstant E') is to introduce
Ef
a positive damping torque component. However, there Can be situations where K~ is
negative. From the expression given by Equation 12.128, K4 is negative when
(XE+Xq)sin60-(R,+RE)cos60
is negative. This is the situation when a hydraulic
generator without damper windings is operating at light load and is connected bv
line of relatively high resistance to reactance ratio to a large system. This typiof
situation was reported in reference 7.
Also K4 can be negative when a machine is connected to a large local load,
supplied partly by the generator and partly by the remote large system [8]. Under such
conditions, the torques produced by induced currents in the field due to armature
reaction have components out of phase with Am, and produce negative damping.
(Aw
--a
Example 12.3
In this example we analyze the small-signal stability of the system of Figure E12.5
(considered in Example 12.2) including the effects of the generator field circuit
dynamics. The parameters of each of the four generators of the plant in per unit on
its rating are as follows:
The above parameters are unsaturated values. The effect of saturation is to be

represented by assuming that d and q axes have similar saturation characteristics with
A,,
B,, = 6.93
0.03 1
wTI = 0.8
The effects of the amortisseurs may be neglected. The excitation system is on manual
control (constant E:/d and transmission circuit 2 is out of service.
(a)
If the plant output in per unit on 2220 MVA, 24 kV base is
0.9
0.3 (overexcited)
E,
1.0
compute the following:

(i)
The elements of the state matrix A representing the small-signal

performance of the system.
(ii)
The constants K, to K4 and T3 associated with the block diagram

representation of Figure 12.9.
Single-Machine Infinite Bus S y s t e m
12.3
(b)
753
(iii)
Eigenvalues of A and the corresponding eigenvectors and participation

matrix; frequency and damping ratio of the oscillatory mode.
(iv)
Steady-state synchronizing torque coefficient; damping and

synchronizing torque coefficients at the rotor oscillating frequency.
Determine the limiting value of P (within k0.025 pu) and the corresponding
value of the rotor angle 6 beyond which the system is unstable, with
(0
Saturation effects neglected
(ii)
Saturation effects included
Assume that Q=P/3 as P varies and El=l.O. Comment on the mode of

instability and the effect of representing saturation.
Solution
The four units of the plant may be represented by a single generator whose parameters
on 2220 MVA base are the same as those of each unit on its rating. The circuit model
of the system in per unit on 2220 MVA base is shown in Figure E12.7.
Figure E12.7
The generators of this example have the same characteristics as the generator
considered in examples of Chapters 3 and 4, except for LI.
The per unit fundamental parameters (elements of the d- and q-axis equivalent
circuits) of the equivalent generator following the procedure used in Example 4.1 are
Lad, = 1.65
R, = 0.003
L, = 1.60
Rfd = 0.0006
L, = 0.16
Lfd = 0.153
(a) (i) The initial steady-state values of the system variables are computed by using
the procedure summarized earlier in this section.
754
Chap.
From Equation 12.108,
(ii) From Equations 12.113, 12.114, and 12.120, the constants of the block diagram
of Figure 12.9 are
(iii) Eigenvalues computed by using a standard routine based on the QR

transformation method are
A,,
h, =
-0.1 1kj6.41 (ad=1.02 Hz, <=0.017)

A3 = -0.204
The right eigenvector matrix is
The left eigenvector matrix normalized so that
The participation matrix is
'Pa =I is
Set. 12.3
Single-Machine Infinite Bus S y s t e m
755
From the participation matrix, we see that Am, and A6 have a high participation in the
oscillatory mode (corresponding to eigenvalues hl and k2);the field flux linkage has
a high participation in the non-oscillatory mode, represented by the eigenvalue h3.
(iv) The steady-state synchronizing torque coefficient due to Ayfd is
The total steady-state synchronizing torque coefficient is
Ks
K, - K2K3K4
0.7643-0.3963
0.3679
pu torquelrad
From the block diagram of Figure 12.9,
AS(s) I due to A$fd
1+ST,
1 -s2T:
Therefore, AT, due to Ayfd is
From the eigenvalues, the complex frequency of rotor oscillation is -0.1 1 +j6.41. Since
the real component is much smaller than the imaginary component, we can compute
Ks and KD at the oscillation frequency by setting s=j6.41 without loss of much
accuracy.
KS(A$fdfd)=
-K~K3K4
1-s2T:
-0.00172
1.53
-0.3963
1-~6.41~2.365)~
pu torquelrad
pu torquelpu speed change
Chap. 1
The effect of field flux variation (i.e., armature reaction) is thus to reduce the
synchronizing torque slightly and to add a damping torque component.
The net synchronizing torque component is
The only source of damping is due to field flux variation. Hence, the net damping
torque coefficient is
KD
KD(AJrfd)= 1.53
From Equation 12.81, the undamped natural frequency is
and from Equation 12.82, the damping ratio is
The above values of w, and
4 agree with those computed from the eigenvalues.
(b) The stability limit is determined by increasing P with Q=P/3 and E,=1.0 pu (EB
is allowed to take appropriate values so as to satisfy the network equations). The
results with and without saturation effects are as follows.
(i) With saturation effects:
The limiting P (within f0.025 pu) and the corresponding system conditions in per unit
are
The corresponding K constants are
see- 12.3
Hence, the steady-state synchronizing torque coefficient is

Ks
K1-K2K3K4 = -0.00 14 pu torquelrad
The eigenvalues of the system state matrix are

A,, h, = -0.226kj4.95
h3 = +0.00142
(a, = 0.79 Hz,
0.046)
The above represents conditions just past the stability limit. The system instability is
due to lack of synchronizing torque. This is reflected in the real eigenvalue becoming
slightly positive, representing a mode of instability through a non-oscillatory mode.
(ii) Without saturation effects:
The limiting value of P and the corresponding system conditions in this case are
The K constants are
The steady-state synchronizing torque coefficient is

Ks
K1-K2K3K4 = 0.0001 - pu torquelrad
The eigenvalues are
The system is on the verge of instability. The limiting rotor angle 6 is very close to
90'. With constant Efd and negligible saliency, the limiting rotor angle will be equal
to 90' if the values of Ld and Laq used to compute the initial operating condition are
the same as the values used to relate incremental flux linkages and currents.
In case (i), when we represented saturation, we made a distinction between total
saturation and incremental saturation. Hence, the limiting rotor angle was about 102",
significantly higher than 90'.
rn
Chap.
12.4 EFFECTS OF EXCITATION SYSTEM

In this section, we will extend the state-space model and the block diagram
developed in the previous section to include the excitation system. We will then
examine the effect of the excitation system on the sma@-signal stability performance
of the single-machine infinite bus system under consideration.
The input control signal to the excitation system is normally the generator
terminal voltage E,. In the generator model we implemented in the previous section
Et is not a state variable. Therefore, E, has to be expressed in terms of the stat;
variables A m , A&, and A ~ J ~
In Section 3.6.2, we showed that Et may be expressed in complex form:
Hence,
Applying a small perturbation, we may write
By neglecting second-order terms involving perturbed values, the above equation

reduces to
Therefore,
AE,
-edo~ e ~ +e * ~ 4e
Et 0
Et 0
In terms of the perturbed values, Equations 12.94 and 12.95 may be written as
Ae,
-RaAid+LlAi,-A$,,
Ae,
= -Ra A iq
-4A id +Aqad
Use of Equations 12.106, 12.107, 12.109, and 12.110 to eliminate Aid Ai,, Ayad and
Ayr,, from the above equations in terms of the state variables and substitution of the
set. 12.4
Effects of Excitation System
resulting expressions for Aed and Ae, in Equation 12.131 yield
For the purpose of illustration and examination of the influence on small-signal

stability, we will consider the excitation system model shown in Figure 12.11. It is
representative of thyristor excitation systems classified as type ST1A in Chapter 8.
The model shown in Figure 12.11, however, has been simplified to include only those
elements that are considered necessary for representing a specific system. A high
exciter gain, without transient gain reduction or derivative feedback, is used.
Parameter TR represents the terminal voltage transducer time constant.
I'erminal voltage
transdilcer
Exciter
Figure 12.11 Thyristor excitation system with AVR

The only nonlinearity associated with the model is that due to the ceiling on
the exciter output voltage represented by EFMM and EFm For small-disturbance
studies, these limits are ignored as we are interested in a linearized model about an
operating point such that Efd is within the limits. Limiters and protective circuits
(UEL, OXL, VIHz) are not modelled as they do not affect small-signal stability.
From block 1 of Figure 12.11, using perturbed values, we have
Hence
760
Chap.
Substituting for AE, from Equation 12.132, we get
From block 2 of Figure 12.11,

Efd =
KA( r'ef
In terms of perturbed values, we have

AEfd
KA(-Avl)
The field circuit dynamic equation developed in the previous section, with the effect
of excitation system included, becomes
where
remain unchanged and are given by Equation

The expressions for a,,, a32and
12.116.
Since we have a first-order model for the exciter, the order of the overall
system is increased by 1; the new state variable added is Av,.
where
and K, and Kg are given by Equations 12.133 and 12.134.
Set. 1 2.4
Since pAw, and p A 6 are not directly affected by the exciter,
a14 -- a%
The complete state-space model for the power system, including the excitation system
of Figure 12.11, has the following form:
With constant mechanical torque input,
Block diagram including the excitation system

Figure 12.12 shows the block diagram obtained by extending the diagram of
Figure 12.9 to include the voltage transducer and AVR/exciter blocks. The
K4
vref
Exciter
AvL
1+sT3
K3
K2
Field circuit
1 KD
2Hs+
A%
-OO
s
Kl
I
Voltage transducer
Figure 12.12 Block diagram representation with exciter and AVR
762
Chap. 12
representation is applicable to any type of exciter, with Gex(s)representing the transfQ

function of the AVR and exciter. For a thyristor exciter,
G,(s)
K,
The terminal voltage error signal, which forms the input to the voltage transducer
block, is given by Equation 12.132:
The coefficient K6 is always positive, whereas Kg can be either positive or negative

depending on the operating condition and the external network impedance
The value of K5 has a significant bearing on the influence of the AVR on the damping
of system oscillations as illustrated below.
Effect of AVR on synchronizing and damping torque components
With automatic voltage regulator action, the field flux variations are caused by
the field voltage variations, in addition to the armature reaction. From the block
diagram of Figure 12.12, we see that
By grouping terms involving
Ay- and rearranging,
The change in air-gap torque due to change in field flux linkage is
As noted before, the constants K2, K3, K4, and K, are usually positive; however, K5
may take either positive or negative values. The effect of the AVR on damping and
synchronizing torque components is therefore primarily influenced by K, and Gex(s).
We will illustrate this by considering a specific case with parameters as follows:
~ e c .12.4
763
hi^ represents a system with a thyristor exciter and system conditions such that K5
is negative.
a. steady-state synchronizing torque coefficient:
(
From Equations 12.143 and 12.144, with s =jw =0, AT, due to Avfd is
Hence, the synchronizing torque coefficient due to Avfd is
We see that the effect of the AVR is to increase the synchronizing torque component
at steady state. With KA=O (i.e., constant EP), KqAItii) = -0.9. When KA = 15, the AVR
compensates exactly for the demagnetizing effect- of the armature reaction. With
K~ =2009 Ks(~qfd)
= 0.529 and the total synchronizing torque coefficient is
Here, we considered a case with K5 negative. With a positive K5 the AVR would have
an effect opposite to the above; that is, the effect of the AVR would be to reduce the
steady-state synchronizing torque component.
Although we have considered a thyristor exciter in our example, the above
observations apply to any type of exciter with a steady-state exciter1AVR gain equal
to KA.
(b) Damping and synchronizing torque components at the rotor oscillation frequency:
Substitution of the numerical values applicable to the specific case under

consideration in Equation 12.143 yields
Chap.
We will assume that the rotor oscillation frequency is 10 rad/s (1.6 Hz). With
s=ja =jlO,
With K,= -0.12 and KA=200,
Thus the effect of the AVR is to increase the synchronizing torque component and
decrease the damping torque component, when K, is negative.
The net synchronizing torque coefficient is
Ks
K~+Ks(~ud= 1.591+0.2804
1.8714
pu torquelrad
The damping torque component due to Aty- is
UA6 )
K ~ ( ~ I V= l -0.3255
d)
Since A o , =sM/o,=jaAGlo,
Set. 12.4
with
= 10 radls,
the damping torque coefficient is
KD(Aqfd)
-12.27
the absence of any other source of damping, the total KD = KDcA6,.

It is readily apparent that, with K5 positive, the synchronizing and damping
torque components due to Ayfd would be opposite to the above.
For the system under consideration, Table 12.1 summarizes the effect of the
AVR on Ks and KD at w =10 radls for different values of KA.
Table 12.1
?
KA
Kf(~qfd)
Ks = KI K S C A ~ ,
+
K~(~qfd)
0.0
10.0
15.0
25.0
50.0
100.0
200.0
400.0
1000.0
Infinity
-0.0025
-0.0079
-0.0093
-0.0098
0.0029
0.0782
0.2804
0.4874
0.5847
0.6000
1.5885
1.5831
1.5817
1.5812
1.5939
1.6692
1.8714
2.0784
2.1757
2.1910
1.772
0.614
0.024
-1.166
-4.090
-8.866
-12.272
-9.722
-4.448
0.000
With KA=O,Ay- is entirely due to armature reaction. The effect of the AVR
is to decrease KD for all positive values of KA. The net damping is minimum (most
negative) for KA=200,and is zero for KA=m. For low values of KA,the effect of the
AVR is to decrease Ks very slightly, the net Ks being minimum at KAof about 46. As
KA is increased beyond this value, Ksincreases steadily. For infinite value of KA,the
torque due to Ayfd is in phase with A6, and hence has no damping component.
We are normally interested in the performances of excitation systems with
moderate or high responses. For such excitation systems, we can make the following
general observations regarding the effects of the AVR:
With K5 positive, the effect of the AVR is to introduce a negative
synchronizing torque and a positive damping torque component.
The constant K5 is positive for low values of external system reactance and
low generator outputs.
The reduction in Ks due to AVR action in such cases is usually of no
particular concern, because K, is so high that the net Ks is significantly greater
766
Chap. ,2
than zero.
With Kg negative, the AVR action introduces a positive synchronizing to,
9ue
component and a negative damping torque component. This effect is
pronounced as the exciter response increases.
For high values of external system reactance and high generator outputs K~ is
negative. In practice, the situation where KQis negative are commonl
Y
encountered. For such cases, a high response exciter is beneficial in increasing
synchronizing torque. However, in so doing it introduces negative damping
We thus have conflicting requirements with regard to exciter response. one
possible recourse is to strike a compromise and set the exciter response so that
it results in sufficient synchronizing and damping torque components for the
expected range of system-operating conditions. This may not always be
possible. It may be necessary to use a high-response exciter to provide the
required synchronizing torque and transient stability performance. With a very
high external system reactance, even with low exciter response the net
damping torque coefficient may be negative.
An effective way to meet the conflicting exciter performance requirements with
regard to system stability is to provide a power system stabilizer as described in the
following section.
12.5 POWER SYSTEM STABILIZER

The basic function of apower system stabilizer (PSS) is to add damping to the
generator rotor oscillations by controlling its excitation using auxiliary stabilizing
signal(s). To provide damping, the stabilizer must produce a component of electrical
torque in phase with the rotor speed deviations.
The theoretical basis for a PSS may be illustrated with the aid of the block
diagram shown in Figure 12.13. This is an extension of the block diagram of Figure
12.12 and includes the effect of a PSS.
Since the purpose of a PSS is to introduce a damping torque component, a
logical signal to use for controlling generator excitation is the speed deviation Am,.
If the exciter transfer function G,(s) and the generator transfer function
between AEfd and AT, were pure gains, a direct feedback of Am, would result in a
damping torque component. However, in practice both the generator and the exciter
(depending on its type) exhibit frequency dependent gain and phase characteristics.
Therefore, the PSS transfer function, G A S ) , should have appropriate phase
compensation circuits to compensate for the phase lag between the exciter input and
the electrical torque. In the ideal case, with the phase characteristic of Gpsds) being
an exact inverse of the exciter and generator phase characteristics to be compensated,
the PSS would result in a pure damping torque at all oscillating frequencies.
set. 12.5
Power System Stabilizer
Voltage transducer
Figure 12.13 Block diagram representation with AVR and PSS
It should be recognized that the generator model assumed in the representation

shown in Figure 12.13 neglects amortisseurs to simplify the system model and allow
its representation in the form of a block diagram. However, amortisseurs could have
a significant effect on the generator phase characteristics and should be considered in
establishing the parameters of the PSS. Representation of amortisseur effects will be
considered in Section 12.6.
We will illustrate the principle of PSS application by considering the system
used in the previous section for examining the effect of the excitation system. The
system parameters, as before, are
Since TR is very small in comparison to T,, we will neglect its effect in

examining the PSS performance. This simplifies the analysis without loss of accuracy.
From the block diagram of Figure 12.13, with TRneglected, Ayfd due to PSS
is given by
768
Chap. 12
Therefore,
Let us examine the PSS phase compensation required to produce damping

torque at a rotor oscillation frequency of 10 rads. With s =jo=jl0,
AT,,
ATe due to PSS
K2(Aqfd due to PSS)
Therefore, at a frequency of 10 radls,
If AT,,, has to be in phase with Ao, (i.e., a purely damping torque), the Am, signal
should be processed through aphase-lead network so that the signal is advanced by
8 =42.3" at a frequency of oscillation of 10 rad/s. The amount of damping introduced
depends on the gain of PSS transfer function at that frequency. Therefore,
AT,,
(gain of PSS at o = 10) (3.522) (Ao,)
With the phase-lead network compensating exactly for the phase lag between AT, and
Av,, the above compensation is purely damping.
The damping torque coefficient due to PSS at o =10 rads is equal to
K,(PSS)
(gain of PSS) (3.522)
Set. 12.5
769
In the previous section, KD due to AVR action was found to equal -12.27. Therefore,
the et KD including the effects of AVR and PSS is
KD
-12.27 + (gain of PSS at o = 10) (3.522)
with a gain of 12.2713.522 =3.48, the PSS produces just enough damping to
for the negative damping due to AVR action. As the PSS gain is
further, the amount of damping increases.
If the phase-lead network provides more compensation than the phase lag
between AT, and Avs, the PSS introduces, in addition to a damping component of
torque, a negative synchronizing torque component. Conversely, with undera positive synchronizing torque component is introduced. Usually, the
~ S is
S required to contribute to the damping of rotor oscillations over a range of
frequencies, rather than a single frequency,
We will illustrate the basic structure, modelling, and performance of power
system stabilizers by considering a thyristor excitation system. Figure 12.14 shows the
block diagram of the excitation system, including the AVR and PSS. Since we are
~oncernedwith small-signal performance, stabilizer output limits and exciter output
limits are not shown in the figure. The following is a brief description of the basis for
the PSS configuration and considerations in the selection of the parameters.
The PSS representation in Figure 12.14 consists of three blocks: a phase
compensation block, a signal washout block, and a gain block.
The phase compensation block provides the appropriate phase-lead
characteristic to compensate for the phase lag between the exciter input and the
generator electrical (air-gap) torque. The figure shows a single first-order block. In
practice, two or more first-order blocks may be used to achieve the desired phase
compensation. In some cases, second-order blocks with complex roots have been used.
Terminal voltage
transducer
Gain
Washout
vref
Exciter
Phase
compensation
Power system stabilizer

Figure 12.14 Thyristor excitation system with AVR and PSS
Chap.
Normally, the frequency range of interest is 0.1 to 2.0 HZ, and the phase-lead
network should provide compensation over this entire frequency range. The phase
characteristic to be compensated changes with system conditions; therefore,
compromise is made and a characteristic acceptable for different system conditions
is selected. Generally some undercompensation is desirable so that the PSS, i,
addition to significantly increasing the damping torque, results in a slight increase
the synchronizing torque.
The signal washout block serves as a high-pass filter, with the time constant
Twhigh enough to allow signals associated with oscillations in a, to pass unchanged
Without it, steady changes in speed would modify the terminal voltage. It allows
PSS to respond only to changes in speed. From the viewpoint of the washout function
the value of Tw is not critical and may be in the range of 1 to 20 seconds. The
consideration is that it be long enough to pass stabilizing signals at the frequencies
of interest unchanged, but not so long that it leads to undesirable generator voltage
excursions during system-islanding conditions.
The stabilizer gain KmB determines the amount of damping introduced by the
PSS. Ideally, the gain should be set at a value corresponding to maximum damping;
however, it is often limited by other considerations.
In applying the PSS, care should be taken to ensure that the overall system
stability is enhanced, not just the small-signal stability [9]. The performance objectives
of the PSS, alternative input signals, and a detailed description of the procedure for
selection of its parameters so as to enhance overall system stability are presented in
Chapter 17.
System state matrix including PSS
From block 4 of Figure 12.14, using perturbed values, we have
Hence
Substituting for pAo, given by Equation 12.115, we obtain the following expression
for pAvz in terms of the state variables:
set. 12.5
since pAv2 is not a function of Av, and Av3, a54'a56'0.

From block 5,
Hence
substitution for pAv2, given by Equation 12.145, yields
where
AEfd
KA(Avs-Avl)
The field circuit equation, with PSS included, becomes
Smal.1-Signal Stability
Chap, 1
where
The complete state-space model, including the PSS, has the following form (with
AT, =O):
Alternative methods of treating blocks 4 and 5:
Block 4 may be considered to be made up of two blocks:
In this case, Av; becomes the state variable, with
and the output Av2 of the block is given by
set 12.5
773
The advantage of this approach is that ,the expression for the derivative of the
input variable to the block is not required. This is important in situations where the
input is not a state variable, in which case the expression for its derivative is not
available. Similarly, block 5 may be treated as follows:
In this case, v,' is the state variable, with
and the output Av, is given by
The disadvantage of this approach is that the output of the block is not a state
variable and hence cannot be monitored directly by computing the state variable.
For excitation system models with higlier-order transfer function blocks the
following general procedure may be used.
Higher-order transfer function blocks:
Consider an nth order transfer function whose block diagram is shown in
Figure 12.15.
Figure 12.15
774
Chap, 1
The above system has n poles and m zeros (m<n). The transfer function ,p
Figure 12.15 may be divided into two parts:
Hence the nth order transfer function may be represented in terms of two blocks as
shown in Figure 12.16.
\-Y
Figure 12.16

x l ( i + T ~ ~ s + T-..~
+ ~T s~n,~~ +=s V ,
Let
From Equation 12.153, in the time domain

x1+TDIx2+TD2x3+"'+TD(n-1)
Therefore,
x n +Thin = vi
Set 12.5
combining Equations 12.154 and 12.155, we have
From block 2 of Figure 12.16, the expression for the output is
Equations 12.156 and 12.157 have the state-space form
The variables xl ,x2, ...,x, defined above are referred to as phase variables. They are
a particular set of state variables that consists of one variable and its n-1 derivatives.
Example 12.4
This example is an extension of the previous two examples which analyzed the smallsignal stability of the system shown in Figure E12.5. Here we consider the effect of
two alternative types of excitation control:
(a)
A thyristor exciter with AVR as shown in Figure 12.11, with
(b)
The above excitation system with PSS added as shown in Figure 12.13, with
The generator parameters and saturation data are as given in Example 12.3. With the
transmission circuit 2 out of service, the plant output in per unit on 2220 MVA and
24 kV base is as follows:
P = 0.9
E, = 1.O136"
Chap.
Q = 0.3 (overexcited)
EB = 0.99510'
Analyze the small-signal stability of the system with each of the two types of
excitation control, by determining
(i)
Elements of the state matrix
(ii)
The constants of the block diagram of Figure 12.12
(iii)
Eigenvalues of the state matrix, participation &%trix, and freguency and

damping ratio of each oscillatory mode
(iv)
Synchronizing and damping torque coefficients at the rotor ~ ~ ~ i l l a t i ~ ~

frequency
Compare with the results of Examples 12.2 and 12.3.

Solution
(i) From equations developed in Section 12.3.3, the state matrix of the system
including the AVR and PSS is
States
A "r
A6
A+fd
A"1
A",
A",
With AVR only (i.e., without PSS), the last two rows and columns disappear.
(ii) The constants of the block diagram are
(iii) (a) With AVR only, the eigenvalues are
A,,
h, =
h3 =
h4 =
+0.504kj7.23
-20.202
-31.230
and the participation matrix is
(w,=1.15 Hz, C,=-0.07)
Set. 12.5
Power System Stabilizer,
Only the magnitudes of the participation factors are shown above; the angles do not
provide any useful information.
With AVR only, the system becomes unstable through an oscillatory mode of 1.15
Hz. From the participation factors, we see that this mode is associated primarily with
the rotor angle and speed. The two non-oscillatory modes, both of which decay
rapidly, are associated with the AVR and the field circuit.
(b) With AVR and PSS, the eigenvalues are
A, = -39.097
h, h3 =
-1.0055j6.607
3L4 = -0.739
A,,& = -19.797kj12.822
(md=1.05 Hz, &=0.15)

(md=2.04Hz,<=0.84)
and the participation matrix (magnitudes only) is
With the addition of the PSS, the system has become very stable. There are two
oscillatory modes: one is the rotor angle mode with a frequency of 1.05 Hz; the other
has a frequency of 2.04 Hz and is associated with the excitation system and field
circuit. The two non-oscillatory modes are associated with the excitation system.
(iv) (a) Synchronizing and Damping Torque Coeflcients with A VR Only
From the block diagram of Figure 12.12, field flux linkage change due to AVR and
armature reaction (AR) is
Chap. 12
Substituting the values of the constants summarized in (i) above and simplifying,
get
From the eigenvalue calculations, the complex frequency of rotor oscillation is

0.504+j7.23. With s equal to this complex frequency, the above expression for Ay,fd
simplifies to
Hence, AT, due to Ayfd is
Now,
Hence,
Substituting for jA8 in the expression for AT,, we get
Therefore,
Ks
KD(AVR+ARl
0.2 115 pu torquelrad

pu torque/pu speed change
= -7.06
set. 12.5
779
The field flux linkage variation is due to armature reaction and AVR. In Example
12.3, we saw that the effect of the armature reaction alone is to decrease Ks by
0.00172 and increase KD by 1.53 at the rotor oscillation frequency of 6.41 rad/s. The
effect of AVR alone is therefore to increase Ks by nearly 0.2 and to decrease KD by
nearly 8.5.
The total synchronizing and damping torque coefficients are
and
K~
=+-7.06
,
From Equation 12.82, the corresponding damping ratio is
This agrees with
computed from the eigenvalues.
(b) Synchronizing and Damping Torque Coeflcients with PSS

The complex frequency of rotor oscillation in this case is -1.005 cj6.607. From the
J to~ PSS is
block diagram of Figure 12.13, A ~ due
with
Therefore,
Chap. 1
Substituting the numerical values of the parameters, setting s = - 1.005 +j6.607, and
simplifying, we get
IPSS
(24.0+j8.089)Awr
24.0 Aor+8.089(-0.0179 A6 -0.1521 A",)
Now,
Hence,
IPSS
The synchronizing and damping torque coefficients due to PSS are
Ks(pss, = -0.145
KD( p s s ) = 22.77
pu torquelrad
and K' and KD due to AVR and armature reaction (AR)must now also be evaluated
at s = - 1.005 +j6.607. Following the same procedure used before, we have
Ks(AvR+AR)= 0.21
KD(AVR+AR)
= -8.69
pu torquelrad
The total synchronizing and damping torque coefficients are
KD = K D , + ~ + K D ~ S S )
= -8.69 + 22.77
= 14.08
Set. 12.5
From Equation 12.81, the undamped natural frequency is
The above results agree with
C; and cod computed from the eigenvalues.
We see that the PSS has increased KD by 22.77 and reduced Ks by 0.145. The
reduction in Ks indicates that phase lead compensation used results in slight
overcompensation at the rotor oscillation frequency of 6.61 radls. By adjusting TI
and/or T2, the compensation may be haltered to give zero or even positive
synchronizing torque component.
The following is a summary of frequency of rotor oscillation, damping ratio, Ks and
KD with different types of excitation control and with classical model, based on the
results of this and the previous two examples.
Model
Classical Model
(KD =o>
Constant Efd
AVR only
AVR and PSS
ad
1.02 Hz
1.07 Hz
1.15 Hz
1.05 Hz
0.0 17
Ks
0.757
0.763
KD
1.53
-0.07
0.976
-7.06
0.15
0.829
14.08
782
Chap. 12
12.6 SYSTEM STATE MATRIX WITH AMORTISSEURS

When generator arnortisseurs are included, the system equations developed in
Section 12.3.2 change as follows. We will assume that the model includes one d-axis
amortisseur and two q-axis amortisseurs, as shown in Figure 12.17.
Figure 12.17 Synchronous machine equivalent circuits
Rotor circuit equations
The rotor currents are given by

1
ifd = -(qfdfd-qad)
4
ild =
1
Lld
Set. 12.6
System State Matrix with Amortisseurs
n e d- and q-axis mutual flux linkages are given by

@ad
-Lads id + La h ifd +Lads iId
Lfd
Lld
The expressions for id and i, (corresponding to Equations 12.103 and 12.104) become
where
Lfd
Lld
AS discussed in Chapter 5, i is assumed to be 1.0 pu to counterbalance the effects

of neglecting stator transients.
784
Chap. 1
Expressing Equations 12.164 and 12.165 in terrns of perturbed values, we get
with
XTq
m2=--
L*
Lrd
XTq
Lld
RT
n3 =--
The expressions for Avndand
L:kr
m3=--
L:*
Lld
Avaqare given by
set. 12.6
he expression for AT, (corresponding to Equation 12.112) is given by
where
L"aqs .
K22= n4(@adO+ L~qsidO)-m4(@aqOfL&iqO)--
ldo
L"aqs
K23 = ~ 5 ( @ d + L ~ q s i d o ) - ~ S ( @ a q o + L ~ i ld0
qo)--
From Equations 12.83 and 12.170
where
785
786
As before,
where
From Equations 12.158, 12.159, and 12.168,
where
a3,= O
a35= -
=-
o0 Rf d m 'L"
r
1 ads
o0R
1
fdm
4 ads
Similarly,
where
a,,
=-
o0 Ri d m L"
1 ads
Lld
Chap. 1
Set 12.6
a44= - W ~ R l d ( l : + m
3 L a&)
H
Lld
a45
o.OR I d , L"
4
U4ik
a46= -
Lld
o0R I d , L"
5
adrs
Lld
where
a53
o 0 R19
2 aqs
L19
where
L"
- - o 0 R1qn L"
a54 -
3 aqs
L19
Chap. 12
The complete state equation is given by
With constant input torque, AT, =O. Similarly, with constant field voltage, AEfd=O.
If AVR action is to be represented, we need the expression for AE,. Following
the procedure used in Section 12.3.3,
with
By using Equations 12.166, 12.168, and 12.169, the perturbed value of the terminal
voltage may be expressed in the following form:
It is left to the reader as an exercise to develop the expressions for the K constants of
the above equation in terms of system parameters.
set. 12.6
789
In this example, we extend the analysis presented in Examples 12.3 and 12.4 by
including the effects of generator amortisseurs. In addition, we examine the effects of
varying the exciter gain and stabilizer gain to provide additional insight into how the
AVR and PSS influence the small-signal dynamic characteristic.
The per unit generator parameters related to the amortisseurs are as follows:
All other data (including generator and excitation system parameters, and system
conditions) are as considered in Example 12.4.
(a)
With the excitation system on manual control, determine the state matrix and
its eigenvalues. Identify the state variable with the highest participation for
each mode.
(b)
With AVR in service and PSS out of service, examine the effect of varying
the exciter gain KA on the eigenvalues. Assume TR=0.02 s.
(c)
With AVR and PSS in service, examine the effect of varying the stabilizer
gain KSAB on the eigenvalues. Assume all other parameters of the excitation
system are fixed at values given in Example 12.4.
Solution
(a) The per unit values of the leakage inductances and resistances of the amortisseur
circuits are computed following the procedure used in Example 4.1 as follows:
Based on equations developed in Section 12.6, the state matrix of the system
including amortisseurs, but with constant Efl,is
The eigenvalues of A and the corresponding state variable with highest participation
are
Eigenvalues
2'
'17
'3
'4
5'
'6
Dominant State(s)
-0.171kj6.47
-0.200
-2.045
-25.01
-37.85
Chap,
c;
A ~ f d
A ~ ~ q
A~2q
A ~ l d
Comparing with the results of Example 12.3, we see that the effect of the amortisseuls
is to increase the damping ratio from 0.017 to 0.0265 and to increase the frequency
very slightly from 1.02 to 1.03 Hz.
(b) Table E12.1 shows the effect of varying KA on the eigenvalues of the state matrix
with PSS out of service. The table also indicates for each mode the state variable with
highest participation as well as any other state with significant participation.
Table E12.1
Eigenvalues
KA
A,?
0.1
5.0
10.0
15.0
30.0
50.0
100.0
110.0
125.0
130.0
150.0
200.0
400.0
Dominant
States
Significant
States
h,
-0.171kj6.47
-0.170kj6.47
-0.112kj6.46
-0.050kj6.46
0.01 1kj6.46
0.182kj6.50
0.357kj6.62
0.552&j6.95
0.564kj7.01
0.573kj7.09
0.574kj7.11
0.570kj7.21
0.528kj7.38
0.315kj7.66
Am,, A6
3L4
3L3
'5
-25.01
-37.85
-37.84
-25.01
-25.02
-37.34
-25.02
-36.83
-36.34
-25.02
-34.86
-25.03
-32.87
-25.06
-25.54
-26.81
-25.47kj0.96
-24.15kj0.88
- 13.03
- 14.14
-22.92
-24.18
-24.72
- 18.09kj5.09
-0.20
-0.20
-0.59
-0.98
-1.34
-2.94
-4.60
-9.40
- 10.64
-24.86
-24.92
- 17.63rtj11.33
- 16.40kj22.63
A ~ f i $A
Av1
d A~2q
A,
-2.04
-2.05
-2.06
-2.10
-2.16
-1.86
-1.94
-1.96
A,
-
-50.00
-50.22
-50.42
-50.79
-51.12
-51.70
-52.84
- 1.97
-53.30
-1.97
- 1.97
-1.97
-1.97
-1.98
-53.71
-54.41
-56.38
AV~q
Av1
AVI~
12.6
791
We see that the effect of the AVR is to introduce negative damping. For values of KA
higher than about 14, the system is unstable. The system is most unstable for a KA of
about 125; as KA is increased further, the system becomes less and less unstable.
With KA of about 110, two real eigenvalues (h4 and h5) merge to become a complex
pair of eigenvalues. At KAof about 130, these eigenvalues become real again. For KA
of about 150 and higher, the real eigenvalues h3 and h4 merge to form a complex pair.
The oscillatory mode represented by these eigenvalues is restricted to the field circuit,
d-axis amortisseur, and the AVR.
(c) Table E12.2 shows the effect of varying the stabilizer gain KSTAB7
with KA=200.
Table E12.2
Eigenvalues
KSTAB
h7
12
h37 h4
A5
0.53kj7.4
- 17.6kj11.3
-24.9
- 1.97
-54.4
0.1
0.5 1kj7.4
-17.6kj11.4
-24.9
-1.97
-54.4
-30.4
-0.71
1 .O
0.34kj7.4
-1 7.1kj11.8
-24.9
- 1.97
-54.4
-30.9
-0.72
3 .O
-0.01kj7.4
-16.2kj12.7
-24.9
-1.97
-54.5
-3 1.9 -0.72
5.0
-0.37kj7.2
-15.4kj13.6
-24.9
-1.97
-54.5
-32.8
-0.73
9.5
-1.08kj6.8
-13.9kj15.5
-25.0
-1.96
-54.5
-34.3
-0.74
15.0
-1.75kj6.2
-12.5kj17.7
-25.0
-1.95
-54.6
-35.6
-0.76
21.0
-2.20kj5.5
-1 1.4kj20.0
-25.0
-1.94
-54.7
-36.7
-0.78
Am,, A6
Av2
Avs
Dominant
States
,I
'6
~Avlq
q
8'
'9
From the results summarized in Table E12.2, we see that as KsTABis increased the
damping of rotor angle mode increases and the frequency decreases slightly. The
reduction in frequency is indicative of reduction in synchronizing torque; this is due
to the PSS phase lead circuit providing overcompensation at the rotor oscillating
fiequency. The frequency of the second oscillatory mode increases and its damping
decreases as KSTABis increased. This mode is associated with the field circuit. All
W
other modes are not affected by the variation of KSTAB.
792
Chap. 1
12.7 SMALL-SIGNAL STABILITY OF MULTIMACHINE
SYSTEM^
Analysis of practical power systems involves the simultaneous solution of

equations representing the following:
Synchronous machines, and the associated excitation systems and prime
movers
Interconnecting transmission network
Static and dynamic (motor) loads
Other devices such.as HVDC converters, static var compensators
Figure 12.18 shows the general structure of the complete system model.
As discussed in Chapter 5, for system stability studies it is appropriate to
neglect the transmission network and machine stator transients. The dynamics of
machine rotor circuits, excitation systems, prime mover and other devices are
represented by differential equations. The result is that the complete system model
consists of a large number of ordinary differential and algebraic equations.
-.-
1 Stator equations and I UR,MI
1 1 axes tra;sforrnation I MR, MI

I ------------------------------------------------,
!
I
I
-.-
Rotor
circuit
equations
Acfeleration. or
swing equation
Transmission
network
equations
including
static loads
-t
, :
Other generators
Motors
I /
lHPrimel1 1 1 1
Other dynamic
devices, e.g.,
HVDC, SVC
governor
.I .I
!--------------------------------------,----
Individual machine
reference frame: d-q
*
**
Common reference
frame: R-I
Algebraic equations
Differential equations
Figure 12.18 Structure of the complete power system model
Set. 12.7
Small-Signal Stability of Multimachine Systems
793
Each machine model is expressed in its own d-q reference frame which rotates
with its rotor. For the solution of interconnecting network equations, all voltages and
must be expressed in a common reference frame. Usually a reference frame
rotating at synchronous speed is used as the common reference. Axis transformation
are used to transform between the individual machine (d-q)reference frames
and the common (R-I) reference frame as shown in Figure 12.19. For convenience in
the organization of the complete set of algebraic equations, the machine stator
equations are also expressed in the common reference frame.
The R-axis of the common reference frame is usually used as the reference for
measuring the machine rotor angle. For a machine represented in detail including
dynamics of one or more rotor circuits, the rotor angle F is defined as the angle by
which the machine q-axis leads the R-axis, as shown in Figure 12.19. For a machine
represented by a classical model, the rotor angle is the angle by which the&ltage E'
leads the R-axis. Under dynamic conditions, the angle 6 changes with rotor speed.
Figure 12.19 Reference frame transformation
The formulation of the state equations for small-signal analysis involves the
development of linearized equations about an operating point and elimination of all
variables other than the state variables. The general procedure is similar to that used
for a single-machine infinite bus system in the previous sections. However, the need
to allow for the representation of extensive transmission networks, loads, a variety of
excitation systems and prime mover models, HVDC links, and static var compensators
makes the process very complex. Therefore, the formulation of the state equations
requires a systematic procedure for treating the wide range of devices. The following
is a description of one such procedure [lo, 11,121.
794
Chap,
Formulation of the state equations

The linearized model of each dynamic device is expressed in the following
form:
where
xi are the perturbed values of the individual device state variables

ii is the current injection into the network from the device
v is the vector of the network bus voltages
In Equations 12.186 and 12.187, Bi and Yi have non-zero elements
corresponding only to the terminal voltage of the device and any remote bus voltages
used to control the device. The current vector ii has two elements corresponding to
the real and imaginary components. Similarly, the voltage vector v has two elements
per bus associated with the device. Such state equations for all the dynamic devices
in the system may be combined into the form
where x is the state vector of the complete system, and AD and CD are block diagonal
matrices composed of Ai and Ci associated with the individual devices.
As described in Chapter 6, the interconnecting transmission network is
represented by the node equation:
The elements of YN include the effects of nonlinear static loads as shown later in this
section.
Equating Equation 12.189 associated kith the devices and Equation 12.190
associated with the network, we obtain
Hence,
Substituting the above expression for Av in Equation 12.188 yields the overall system
state equation:
Set. 12.7
where the state matrix A of the complete system is given by
The method of building Ai, Bi, Ci, and Yi matrices for the synchronous machine and
the associated controls can follow the general approach described in the previous
sections. Motor loads can be treated in a similar way. Reference 12 gives details of
treating HVDC links and SVCs.
The MASS (Multi-Area Small-signal Stability) computer program described
in references 10 and 11 uses the above general approach for formulating the system
state matrix. The program computes all eigenvalues of the matrix by using the QR
transformation method discussed in Section 12.2.9. Since this approach cannot exploit
sparsity, it cannot be used for analysis of very large systems. Depending on the
capability of the computer used, the maximum system size is limited to a few hundred
states. This is satisfactory for the analysis of stability of small systems or local
stability problems in large systems. Since all modes of the system are computed, it
is also ideally suited for the design and coordination of controls. In the next section,
we will describe techniques for the analysis of global problems in very large systems.
Representation of static loads

(a) Constant impedance (linear) load:
The shunt admittance to ground representing the load is computed as

Load bus
Figure 12.20
where
PLO = initial value of the active component of load

QLo = initial value of the reactive component of load
V, = 'initial value of the bus voltage magnitude
796
Chap.
(6) Nonlinear load:

Let us consider the load whose voltage dependent characteristics ae
represented as
where V is the magnitude of the bus voltage given by
The R and I components of the load current are
Figure 12.21
Linearizing, we find
AiR
v~~
-APL+
v:
Air
v~~
-AQL+
pLO
-AvR+
QLO
-Avz+
v:
v:
v~~
v~~
pLO
-APL--AQL+-AvZ--AvR+
QLO
v:
v:
v:
v:
set. 12.7
*here
substitution of Equations 12.196 and 12.197 in Equation 12.195 yields
where
2
QLO
PLO
GRR- -
v:
BIR
- QLO
-
As an example, consider a load such that
PLO = 5.0 pu
Q,
= 2.0
pu
Go = 1.0L20
798
Chap. 1
The real and imaginary components of the initial bus voltage are
vRo
1.0cos2O0
0.94
vzo
1.0sin2O0
0.342 pu
pu
From Equation 12.199, with constant impedance characteristic (m =n=2.0),
Similarly, with constant current characteristic (m =n= 1.O),
and with constant MVA characteristic (m =n=O.O),
The equivalent admittance matrix of Equation 12.198 representing a static load may
be directly implemented in the network admittance matrix. However, as evident from
the above example, the equivalent admittance matrix representing nonlinear loads is
not symmetrical and does not represent a simple shunt admittance to ground as in the
case of a constant impedance load.
Redundant state variables
The formulation of the system state equations described above uses absolute
changes in machine rotor speed and angle as state variables. With such a formulation,
the state matrix of a system which does not contain an infinite bus will have one or
two zero eigenvalues [13].
One of these zero eigenvalues is associated with the lack of uniqueness of
absolute rotor angle. In other words, if rotor angles of all machines are increased by
a constant value, the system stability is not affected. The redundancy in rotor angle
states can be eliminated by choosing one of the machines as a reference and
expressing angle changes of all other machines with respect to this reference as
follows.
For the reference machine R,
for any other machine i (i=l , ... , n; i#R),
Set. 12.8
Special Techniques for Analysis of Very Large Systems
pA8i
799
( A o r of machine i ) - ( A o r of machine R )
6
The second zero eigenvalue exists if all the generator torques are assumed to be
independent of speed deviations, i.e., if a damping term represented by KD is not
included in the swing equation and a speed governor is not represented. This zero
,igenvalue can also be avoided by measuring speed deviations with respect to that of
a reference machine. Mathematically, the process of referring rotor angles or speed
deviations to a reference machine is equivalent to a similarity transformation.
The zero eigenvalues, however, may not be computed exactly because of
mismatches in the power flow solution and the limited accuracy of eigenvalue
calculation routines. They may therefore appear as small eigenvalues.
12.8 SPECIAL TECHNIQUES FOR ANALYSIS

OF VERY LARGE SYSTEMS
Analysis of interarea oscillations in a large interconnected power system
requires a detailed modelling of the entire system. System representations with as
many as 2,000 dynamic devices and 12,000 buses are not uncommon for such studies.
With an average of 15 states per device, the number of state variables required for
modal analysis may thus be on the order of 30,000. This is well outside the range of
the conventional eigenvalue analysis methods. Special techniques have therefore been
developed that focus on evaluating a selected subset of eigenvalues associated with
the complete system response.
One such technique is the AESOPS algorithm originally presented in reference
14. It uses a novel frequency response approach to calculate the eigenvalues
associated with the rotor angle modes. References 10 and 15 describe improved
implementation of the AESOPS algorithm.
Several powerful methods for the computation of eigenvalues associated with
a small number of selected modes of oscillation have been published in the literature
on power system stability. Reference 16 describes the application of two sparsitybased eigenvalue techniques: simultaneous iterations and modified Arnoldi methods.
The S-method described in reference 17 is particularly suited for finding the unstable
modes. The selective modal analysis (SMA) approach described in references 2 and
18 computes eigenvalues associated with selected modes of interest by using special
techniques to identify variables that are relevant to the selected modes, and then
constructing a reduced-order model that involves only the relevant variables.
The PEALS (Program for Eigenvalue Analysis of Large Systems) described in
references 10 and 11 uses two of these techniques: the AESOPS algorithm and the
modified Arnoldi method. These two methods have been found to be efficient and
reliable, and they complement each other in meeting the requirements of small-signal
stability analysis of large complex power systems. The following is a description of
these two techniques:
Chap.
(a) The AESOPS algorithm
The acronym AESOPS stands for the Analysis of Essentially Spontaneous

Oscillations in Power Systems [14]. It is based on an ingenious approach developed
using sound engineering judgement rather than a rigorous mathematical procedure
The algorithm computes eigenvalues associate@only with rotor angle
one complex conjugate pair of eigenvalues at a time. The eigenvalues are determined
by applying to the rotor of a selected generator an external torque having a complex
sinusoidal form, expressed in terms of an initial estimate of an eigenvalue. Under
steady-state conditions, all incremental variables of the linear system model will be
complex sinusoids similar in form to the external torque. The complex frequency
response of the system is calculated by solving an appropriate set of complex
algebraic equations. A revised estimate of the eigenvalue is determined from the
linear system response, and the process is repeated until successive estimates converge
within a desired tolerance. The resulting eigenvalue represents a mode of oscillation
in which the selected generator participates significantly. If the generator has several
dominant modes of rotor oscillations, the eigenvalue computed depends on the initial
estimate.
The procedure may be applied to different generators to compute the
eigenvalues associated with the modes in which the generators have significant
participation.
The AESOPS algorithm is derived from the linearized equation of motion of
a generator:
modes,
Taking the Laplace transform, we get
The above equation recognizes that synchronizing and damping torque coefficients Ks
and KD are functions of the frequency of oscillation.
Rearranging Equation 12.200, we have
The characteristic values (eigenvalues) of the system are given by the zeros of
set. 12.8
801
The characteristic values are thus those values of s which force AT, to be zero,
,,vided Aw, is not zero. This is used by the AESOPS algorithm to determine zeros
P
of Equation 12.202.
With a finite external torque, the magnitude of system variables tends to be
unbounded at complex frequencies close to an eigenvalue. From a computational
it is preferable to limit the extent of system response and allow the
magnitude of external torque to go to zero at an eigenvalue. This is achieved by
setting the complex speed deviation of the perturbed generator to 1.0+j0.0 per unit.
he external torque AT, is determined in the solution process. As iterations converge,
the magnitude of the external torque tends to zero and provides an indication of the
,,curacy of the eigenvalue.
In practice, the model representing the system has'poles as well as zeros. Zeros
,lose to a pole prevent the AESOPS algorithm from converging to the corresponding
eigenval~e~.
Poles associated with modes in which the machine has a significant
participation are not cancelled by zeros. Thus, for each generator in the system, a
subset of the overall system modes may be determined.
Iterative Procedure: The zeros of AT,(s) niay be determined by using
Newton's method. This requires the derivative of 4T, with respect to s. From
\
Equation 12.201,
This can be simplified by making the following approximations.

Close to a system eigenvalue, from Equation 12.202,
I f KD(s)is small,
Therefore,
802
Chap. 12
Further, - and - are small compared to 4H. Therefore,

8s
as
The nth iteration of Newton's method for computing the eigenvalues, i.e., the values
of s, is given by
The momentum of the disturbed machine is
Mn =JAW, = 2HAor
Hence,
For modes that involve many machines, this makes too large a change in the
eigenvalue at each iteration. In such cases, aKs/as and aKD/ds are not small.
Therefore, a modified momentum is used based on an equivalent inertia defined so
that kinetic energy associated with the change in speed at all machines in the system
is equal to the equivalent inertia multiplied by the square of the speed change of the
disturbed machine. The equivalent inertia is given by
with N, =number of machines. Equation 12.208 is then modified to
Set. 12.8
~f as
Special Techniques,,f'or Analysis of Very Large Systems
803
discussed earlier Ao, is set to 1.0 +j0.0 pu,
~~plementation
of the AESOPS algorithm in PEALS
The application of the AESOPS algorithm as developed above requires the
calculation of
(a)
The torque necessary to keep the speed change of a chosen disturbed machine
equal to 1.0+j0.0 per unit; and
(b)
Speed changes of all the other machines and the equivalent inertia He.
These depend on Equation 12.186 of each dynamic device and Equation 12.190 of the
interconnecting network.
We will first organize the device equations in a suitable form to perform the
above calculations. From Equation 12.186, for any complex frequency s, for each
dynamic device, we may write
Therefore,
Substitution of the above in Equation 12.187, and expressing in the s domain yields
where
804
Chap. 1
with
Ai
ai =
diagonal matrix of eigenvalues of Ai of the device

corresponding eigenvector matrix
Computation of Yi,(s) using Equation 12.213 is sigPlified due to sI-Ai being

-.
diagonal.
For the machine k selected for applying the external torque, the dynamic
equations, including the effect of the applied torque AT,, may be witten in the
partitioned form:
where xris a vector representing all state variables of the machine, except for Ao, and
AS. For the disturbed machine, Am, is assumed to be equal to 1.0 pu, and hence
A6 =a,/s.
By manipulation and rearrangement of Equations 12.214 and 12.215 to
eliminate A6 and x, in terms of Am,, and by setting Ao, to 1.0 pu, we obtain the
following expressions for AT, and Aik as functions of Av:
AT,
where
2H s - a l l -
a12u0
I)?+
- a 1 ( s 1 - A [arl
(12.216) -.
Set. 12.8
Special T e c h n i q u e ~ ~ f o
Analysis
r
of Very Large Systems
805
in Equation 12.213, computations associated with the above equations are

,implified by expressing (sI - A ~ ) in diagonal form in terms of the eigenvalues of 4.
The current injections into the network are given by Equation 12.217 for the
disturbed machine and by Equation 12.212 for all other machines and dynamic
devices. The equivalent device admittances Yie(s) and the equivalent current source
&(s) for the disturbed machine are functions of the parameters of the devices, initial
condition, and the complex frequency s.
The interconnecting network equations may be written as
-'
where subscript D refers to buses with dynamic devices and the subscript L refers to
buses with static loads, i.e., buses with no dynamic devices.
Combining the device equations 12.212 and 12.217 and the network equation
12.220 yields
where Y,,(s) is a block diagonal matrix of the device equivalent admittance matrix
Y,(s), and AIDe(s) is the device current source vector with non-zero element only for
the disturbed machine given by Equation 12.218.
Equation 12.221 is solved for bus voltages Av. The network matrix is very
large and sparse. PEALS exploits this by using an efficient sparsity-based network
solution technique.
The following is a summary of the computation steps associated with the
AESOPS algorithm:
1.
Form the device models: Ai, Bi, Ci, Yi; compute eigenvalues and eigenvectors
of Ai using the QR transformation method.
2.
Set the initial value of complex frequency s equal to the specified initial
estimate of the eigenvalue.
3.
Compute AIke(s) and Yke(s) associated with the disturbed machine, using
Equations 12.218 and 12.219, and Yie(s) for all other dynamic devices, using
Equation 12.213.
4.
Solve Equation 12.221 to compute bus voltage Av.
806
Chap. 12
5.
Compute AT, for the disturbed machine, using Equation 12.216; Aw, for all
machines, using Equation 12.211; He, using Equation 12.209.
6.
Compute next estimate of s, using Equation 12.210.
7.
If A s between successive iterations is within the specified tolerance, set

eigenvalue to s and stop; otherwise, go to step %d repeat the process.
The above approach to the calculation of eigenvalues corresponding to rotor

angle modes involves the solution of only the algebraic Equations 12.221. It does not
require the formation of the state matrix A of the overall system. Instead, the state
equations for each dynamic device are formed and treated as a small eigenvalue
problem to form the equivalent admittance matrix Yi,(s). The overall admittance
matrix of Equation 12.221 is highly sparse. Hence, with the use of efficient sparsitybased solution techniques, this approach has been applied to power systems with over
2,000 dynamic devices and 12,000 buses [lo].
One of the limitations of the AESOPS algorithm is that, unless the general
characteristics of critical modes are known a priori, a considerable amount
searching is required to find all the critical modes. In fact, there is never any
assurance that all critical modes have been identified. The method is best suited for
tracking changes in specific modes as system conditions change. The modified
Arnoldi method described next overcomes these limitations of the AESOPS method.
Calculation of eigenvectors:
In the AESOPS approach, the speed deviations Am, for all the machines
computed in the last iteration directly give the corresponding component of the right
eigenvector.
The left eigenvector is calculated by using the transposed system dynamic
model of Equations 12.188 and 12.189:
and the transposed network model of Equation 12.190:
It can be readily shown that the AESOPS algorithm when applied to the above model
will converge to a system eigenvalue and the corresponding left eigenvector. With the
initial eigenvalue estimate equal to the correct value, one iteration is usually sufficient '
for the algorithm to converge to the left eigenvector.
set. 12.8
807
Frequency response calculation:

The frequency response can be calculated directly from the system equations
in the form used by the AESOPS algorithm with s replaced by jo.The overall system
has the general form of Equation 12.221. Instead of applying an external torque as
input to one of the machines, the input signal may be applied to any device and the
outputs monitored at the desired locations. Since the network equations are retained
explicitly, the input and outputs may be associated with the network as well as the
dynamic devices.
/
(6) The modifid Arnoldi method (4M)

The Arnoldi method was first presented in reference 19. However, in the
original form it had poor numerical properties, the main problems being loss of
orthogonality and slow convergence if several dominant eigenvalues are needed.
These problems were solved in reference 20 by using complete reorthogonalization
and an iterative process. The following is a description of the method as implemented
in PEALS [lo, 111.
The modified Arnoldi method is based on a reduction technique in which the
matrix A, whose eigenvalues are to be computed, is reduced to an upper Hessenberg
matrix by the recurrence:
where
hjPi= ~
A (superscript
V ~
H denotes conjugate transpose)
and
is an arbitrary starting vector with 1 vl 1 = 1
v,
h i is a scaling factor to make 1 v , +1~, = 1
m
is a prespecified order of the reduced Hessenberg matrix
Equation 12.222 can be rearranged and assembled for all m equations to give
where
Chap. 1
Here Hm is an upper Hessenberg matrix. We can show that ideally the vector
sequence vi generated by Equation 12.222 is orthonormal. Therefore, if m=N (the
order of A), we have
and Equation 12.223 becomes
Thus A is reduced to the upper Hessenberg matrix HN whose eigenvalues are the
eigenvalues of A. An important feature of the method is that the values of h,+l,, the
subdiagonal elements of Hm, decrease very rapidly as m increases; therefore, a good
approximation can be made for m<<N by dropping out the second term in Equation
12.223:
The eigenvalues of Hm, which is of a very low order, approximate a subset of the
eigenvalues of A. The corresponding eigenvectors of A are given by
-.
W z Vm P
(12.226)
where P is the eigenvector matrix of H,. To improve the accuracy of the eigenvalues
of A, the above procedure can be iterated with a new starting vector vYWderived
from a linear combination of the columns of Vm:
where ai are coefficients calculated from the modal properties of Y, [16].
set. 12.8
Special Techniques .for Analysis of Very Large Systems
809
Equation 12.225 holds only if the vector sequence vi is kept orthogonal at each
step of calculation [20]. In practice, the orthogonality may be lost rapidly due to
numerical cancellation and roundoff errors. The remedy is to introduce a
reorthog~nalizationprocess [16] after the calculation of each vi from Equation 12.222.
An important property of the method is that the eigenvalues of Y, in Equation
12.225 usually converge to those eigenvalues of A which have largest (and smallest)
modulus. Thus, if eigenvalues of A around a specified point h, are desired, the
Ransformation
can be used to magnify the eigenvalues of A close to h,, since
where hi is an eigenvalue of A, and h, is the corresponding eigenvalue of A,. Then

A, is the suitable matrix to which the method can be applied in order to find the set
of eigenvalues of A close to A,, which is termed the shift point.
With the transformation 12.228, the only computation involving A in
constructing the vector sequence vi is the solution of the equation
Equation 12.222 may be written in terms of ui, as
and the upper triangular elements in H, become
hJ71. .
v.H u
J
Application of the MAM to power system model:

Substituting the expression for the state matrix A of a power system given by
Equation 12.192 in Equation 12.230, we get
810
(AD- A , I ) u ~ + B ~ ( Y ~ + Y
cDui
~ ) - ='
Chap.
Vi
This may be written as
where
Rearranging,
CDui-(YN+YD)qi= 0
(12.234)
Equations 12.233 and 12.234 may be combined to give the matrix equation
The following are the computational steps associated with the MAM:
1.
Form the device models Ai, Bi, Ci, Yi.
2.
Compute the device admittance matrix corresponding to the chosen shift point
A,:
3.
Solve for qi:
4.
Calculate ui:
set. 12.8
81 1
Compute vi+, using Equation 12.231 and the related elements in Hm using
Equation 12.232, with the order m equal to the specified value. Repeat steps
3 to 5 to form the complete V, and 33,.
6.
Compute eigenvalues of H,- using the QR transformation method. If the

desired set of eigenvalues has converged within the specified tolerance, stop;
otherwise, compute a new stwing vector v r w using Equation 12.227, and go
to step 3 for the next iteration.
The factorization of the network equivalent admittance matrix YN+Yk(ht)in Equation

12.237 needs to be performed only once since h, does not change during the iteration
Process; this adds to the efficiency of the method. As the admittance matrix is very
sparse, the method can be applied to very large systems by using sparsity-based
techniques for solution of algebraic equations.
The MAM, unlike the AESOPS algorithm, can compute eigenvalues associated
with any system mode, not just the rotor angle mode. In addition, it does not require
apriori knowledge about the mode characteristics of the system. Normally, the shifi
point A, is given in order to compute a particular set of eigenvalues close to it. By
slightly modifying the algorithm, the method can be adopted to provide the capability
for scanning eigenvalues over a frequency range on the complex plane with assurance
that all critical eigenvalues within the specified range have been computed.
Calculation of eigenvectors:
The right eigenvectors are given directly by Equation 12.226 fkom the last
iteration. The left eigenvectors are calculated by inverse iteration. This requires the
solution of Equation 12.235 transposed as follows:
with the computed eigenvalue h, and an arbitrary vector yo. Normally, one iteration
is sufficient to give the left eigenvector.
Comparison of results
Table 12.2 compares the eigenvalue calculations using the QR transformation,

AESOPS, and MAM. The test system consisted of 94 generators and about 1,400
buses. Eighteen of the generators were represented by detailed models and the other
generators were represented by the classical model. This resulted in a system model
with 366 states. The MASS and PEALS referred to above were used to compute the
eigenvalues.
81 2
Chap.
Table 12.2
MAM
Mode
QR Method
AESOPS
-0.122kj1.650
-0.121kj1.650
-0.121kj1.6T
-0.1 17kj3.073
-0.1 17fi3.0%
-0.1 17kj3.073
-0.27635.212
-0.276kj5.212
-0.276kj5.212
All eigenvalues of the state matrix and the corresponding participation factors
were first computed using the MASS program, which uses the QR transformation
method. Then three pairs of complex eigenvalues associated with rotor angle modes
were selected for comparison with AESOPS and MAM. The initial estimates of the
eigenvalues for both methods were set so that they deviated from the true values by
about 10%. From Table 12.2, we see that all three methods give practically identical
results. This is particularly reassuring in view of the widely differing theories on
which the three methods are based.
As a further validation of the above results, frequency response was computed
using PEALS with the change in mechanical torque of a large generator (with high
participation in the three modes) as the input signal and the speed change of the same
generator as the output signal. Figure 12.22 shows a plot of the computed frequency
response, which exhibits resonant frequencies at about 0.26 Hz, 0.49 Hz, and 0.83 Hz.
These are in agreement with the frequencies of the selected modes.
0.26
0.49
0.83
Frequency in Hz
Figure 12.22 Frequency response
Set 12.8
813
In this example, we will analyze the small-signal stability of a simple two-area system
shown in Figure E12.8. This system is similar in structure to the one used in
references 23 and 24 to study the fundamental nature of interarea oscillations.
+
+
Area 1
Area 2
Figure E12.8 A simple two-area system
The system consists of two similar areas connected by a weak tie. Each area consists
of two coupled units, each having a rating of 900 MVA and 20 kV. The generator
parameters in per unit on the rated MVA and kV base are as follows:
1.8
X, = 1.7
Xz = 0.25
X
: = 0.25
T&b = 0.03 s T$ = 0.05 s
H = 6.5 (for G1 and G2)
Xd
XI= 0.2
R,
X& = 0.3
Xi = 0.55
0.0025 Tio = 8.0 s

A , = 0.015 Bsat = 9.6
H = 6.175 (for G3 and G4)
=
Tio = 0.4 s
yTI= 0.9
KD = 0
Each step-up transformer has an impedance of 0 +j0.15 per unit on 900 MVA and
201230 kV base, and has an off-nominal ratio of 1.0.
The transmission system nominal voltage is 230 kV. The line lengths are identified
in Figure E12.8. The parameters of the lines in per unit on 100 MVA, 230 kV base
are
r
0.0001 pulkm
x,
0.00 1 pulkm
b,
0.00175 pulkm
The system is operating with area 1 exporting 400 MW to area 2, and the generating
units are loaded as follows:
GI:
G2:
G3:
G4:
P = 700 MW,
P = 700 MW,
P = 719 MW,
P = 700 MW,
Q
Q
Q
Q
185 MVAr,
235 MVAr,
= 176 MVAr,
= 202 MVAr,
=
=
E,
E,
Et
E,
=
=
=
1.03L20.2"
1.01L10.5"
1.031-6.8"
1.011-17.0"
The loads and reactive power supplied (Qc) by the shunt capacitors at buses 7 and 9
are as follows:
BUS7:
BUS 9:
PL = 967 MW,
PL = 1,767 MW,
QL = 100 MVAr,
QL = 100 MVAr,
Qc
Qc
Chap, 12
200 MVAr
= 350 MVAr
=
(a)
If all four generators are on manual excitation control (constant E'), compute
the eigenvalues of the system state matrix representing the small-signal
performance of the system about the initial operati,ng condition. For each
eigenvalue, identify the system state variables a h high participation.
Determine the frequencies, damping ratios, and mode shapes of the rotor
oscillation modes. Assume that the active components of loads have constant
current characteristics, and reactive components of loads have constant
impedance characteristics.
(b)
Determine the eigenvalues, frequencies, and damping ratios of rotor oscillation

modes when all four generators are equipped -with the following types of
excitation control:
0)
Self-excited dc exciter (see Figures 8.38 and 8.40):

TA = 0.055
TE = 0.36
KA = 20.0
A, = 0.0056 B, = 1.075
TF = 1.8
TB, Tc, Re, and Xc are not used
KF = 0.125
TR = 0.05
(ii)
Thyristor exciter with a high transient gain:
(iii)
Thyristor exciter with a transient gain reduction (TGR):
(iv)
Thyristor exciter with high transient gain and PSS:
The block diagram of thyristor excitation system with PSS is shown

in Figure E12.9.
Solution
(a) System modes with generators on manual excitation control:
Table E12.3 summarizes the eigenvalues of the system state matrix and the state
variables that have high participation in each mode. The first two eigenvalues
represent the zero eigenvalues due to the redundant state variables. As described in
Section 12.7, one of these zero eigenvalues is due to lack of uniqueness of absolute
rotor angle (there is no infinite bus, and the rotor angles are referred to a common
reference frame). The other zero eigenvalue is due to the assumption that the
generator torques are independent of speed deviation (speed governors are not
modelled and KD=O).
Exciter
TGR
Power system stabilizer

Figure E12.9 Thyristor excitation system with PSS
Table E12.3 System modes with manual excitation control
No.
Eigenvalues
Real
Imaginary
Frequency Damping
Ratio
(HZ)
Dominant States
A o and A6 of
G1, G2, G3, G4
1,2
-0.76E-3
+0.22E -2
0.0003
0'331
-0.96E - 1
-0.1 11
-0.1 17
11
k3.43
0.545
0.032
11
11
475
6
amortisseur
81 6
Chap. 12
From the table, we see that the system is stable. There are three rotor angle modes of
oscillation. Their mode shapes (normalized eigenvector components corresponding to
rotor speeds of the four machines) are shown in Figure E12.10. From the mode shapes
we see that the 0.55 Hz mode is the interarea mode, with generators G1 and G2 of
area 1 swinging against generators G3 and G4 of area 2. The 1.09 Hz mode is the
intermachine oscillation local to area 1, with G1 swinging against G2. The third rotor
angle mode, with a frequency of 1.12 Hz, is the internchine
mode local to area 2.
tl
(a) Interarea mode

f=0.545 Hz, <=0.032
(b) Area 1 local mode

f=1.087 Hz, c=0.072
(c) Area 2 local mode

f=1.117 Hz, c=0.072
Figure E12.10 Mode shapes of rotor angle modes with manual excitation control
(b) Rotor angle modes with diffent types of excitation control:
Table E12.4 summarizes the eigenvalues, frequencies, and damping ratios associated
with the rotor oscillation modes for the four alternative forms of excitation control.
We see that the local intermachine modes of oscillation have the same degree of
damping with the dc exciter and thyristor exciter (with and without TGR). The
interarea mode has a small positive damping with the dc exciter. It is unstable with
high gain thyristor exciters. The TGR makes the interarea mode more unstable. The
interarea mode as well as the local modes are very well damped when power system
stabilizers are added to the thyristor exciters.
Table E12.4 Effects of excitation control on rotor oscillation modes
<
with PSS
(f=0.60, <=0.13)
(f=l.28, <=0.22)
(f=1.33, <=0.22)
Set. 12.9
Characteristics of .Small-Signal Stability Problems
12.9 CHARACTERISTICS OF SMALL-SIGNAL

STABILITY PROBLEMS
In large power systems, small-signal stability problems may be either local or
global in nature.
Local problem
Local problems involve a small part of the system. They may be associated
with rotor angle oscillations of a single generator or a single plant against the rest of
the power system. Such oscillations are called local plant mode oscillations. The
stability problems related to such oscillations are similar to those of a single-machine
infinite bus system as studied in Sections 12.3 to 12.6. Most commonly encountered
small-signal stability problems are of this category.
Local problems may also be associated with oscillations between the rotors of
a few generators close to each other. Such oscillations are called intermachine or
interplant mode oscillations. Usually, the local plant mode and interplant mode
oscillations have frequencies in the range of 0.7 to 2.0 Hz.
Other possible local problems include instability of modes associated with
controls of equipment such as generator excitation systems, HVDC converters, and
static var compensators. The problems associated with control modes are due to
inadequate tuning of the control systems [2 11. In addition, these controls may interact
with the dynamics of the turbine-generator shaft system, causing instability of
torsional mode oscillations 1221. Torsional oscillation problems are described in
Chapter 15.
Analysis of local small-signal stability problems requires a detailed
representation of a small portion of the complete interconnected power system. The
rest of the system representation may be appropriately simplified by use of simple
models and system equivalents. Usually, the complete system may be adequately
represented by a model having several hundred states at most.
Global problem
Global small-signal stability problems are caused by interactions among large
groups of generators and have widespread effects. They involve oscillations of a
group of generators in one area swinging against a group of generators in another
area. Such oscillations are called interarea mode oscillations.
Large interconnected systems usually have two distinct forms of interarea
oscillations:
(a)
A very low frequency mode involving all the generators in the system. The
system is essentially split into two parts, with generators in one part swinging
against machines in the other part. The frequency of this mode of oscillation
81 8
Chap. 1
is on the order of 0.1 to 0.3 Hz.

(b)
Higher frequency modes involving subgroups of generators swinging against

each other. The frequency of these oscillations is typically in the range of 0
to 0.7 Hz.
Illustration of global problems

We will consider two large systems with distinct characteristics to illustrate the
two forms of interarea oscillations.
Test system A
This test system consists of 2,310 buses, 375 generators, and 3 HVDC links.
The 3 dc links and 291 generators are modelled in detail. The remaining 84 generators
represent equivalent machines and are modelled by the classical model.
To identify the critical modes of interarea oscillation, eigenvalues of system
modes with frequencies in the range of 0.2 to 1.0 Hz are scanned, using the modified
Arnoldi method. Table 12.3 lists the eigenvalues, frequencies, and damping ratios of
the modes identified by this mode search. Figure 12.23 shows a plot of these
eigenvalues on a complex plane. Box A in the figure represents the region on the
complex plane within which all eigenvalues have been computed; outside this area,
there may exist eigenvalues other than those shown on the complex plane.
Real axis
Figure 12.23 Eigenvalues computed by frequency scan using MAM
Set, 12.9
Characteristics of Small-Signal Stability Problems
81 9
Table 12.3 Modes of system A computed by using frequency scanning technique
- ode1 no.
fl
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
/
Eigenvalue
-0.7567kj1.4163
-0.7193kj1.4605
-0.7999kj1.5623
-0.7806kj1.5776
-0.1265kj1.9894
-0.8248kj2.3509
- 1.0698kj2.3642
-0.1434kj2.7826
-0.2871kj3.2495
-0.1609kj3.7797
-0.4382kj4.6754
-0.0470kj4.7439
-0.3336kj4.8262
-0.5420kj5.2074
-0.055 1kj5.4200
-0.3 116kj5.6159
-0.1451kj5.8825
-0.2385kj6.1228
-0.2243kj6.1745
-0.2552kj6.1814
-0.1961kj6.3504
-0.1457kj6.4356
Frequency (Hz)
0.2254
0.2324
0.2486
0.25 1 1
0.3 166
0.3742
0.3763
0.4429
0.5172
0.6016
0.744 1
0.7550
0.7681
0.8288
0.8626
0.893 8
0.9362
0.9745
0.9827
0.9838
1.01 07
1.0243
Damping Ratio
0.4712
0.44 19
0.4558
0.4435
0.0634
0.33 1 1
0.4123
0.05 15
0.0880
0.0425
0.0933
0.0099
-0.0689
0.1035
0.0102
0.0554
0.0247
0.0389
0.0363
0.0412
0.0309
0.0226
The 22 modes represented by the eigenvalues shown in Figure 12.23 and Table
12.3 can be divided, based on mode shapes, into three categories:
(a)
Interarea modes. Examples are modes 5 and 8 to 12.
(b)
Local plant modes. Examples are modes 13, 21, and 22.
(c)
Control modes. Examples are modes 1 to 4, and 6.
820
Chap. 1
Modes 10, 11, and 12 represent the critical interarea modes. These modes have
frequencies in the 0.6 to 0.75 Hz range. Each represents oscillation of a subgroup of
generators swinging against another subgroup of generators across a weak or heavily
loaded transmission interface. Figure 12.24 shows how the interfaces associated with
the three critical interarea modes geographically divide the overall system. ~ h ,
stability characteristics of each interarea mode of oscillation depend on the power
transfer across its interface and on the controls associated with the subgroups of
generating units on the two sides of the interfa@.
Interface
of mode 11
\.I_)
of mode 12
Figure 12.24 Interfaces of critical interarea modes

of oscillation of system A
Test system B
This system consists of nearly 3,000 buses and 300 generators. It has a
dominant 0.2 Hz interarea mode in which practically all the generators in the system
participate. The mode shape of the interarea mode i s such that its interface splits the
system into two parts as depicted in Figure 12.25. When the system is perturbed, the
generating units on the two sides of the interface swing against each other with a
frequency of about 0.2 Hz.
Table 12.4 summarizes the results of analysis carried out to investigate the
effects of excitation control of the generators at a large thermal plant having a high
participation in the interarea mode. This plant is identified as GS "Dmin Figure 12.25
and has a thyristor excitation system.
We see from the results that the control of excitation of generators at the
thermal plant has a very significant effect on the stability of the 0.2 Hz interarea mode
of oscillation. With a high gain thyristor exciter, for the system conditions considered,
Set, 12.9
Characteristics of Small-Signal Stability Problems
Figure 12.25 Interface of 0.2 Hz interarea

mode of oscillation of system B
Table 12.4 Effect of excitation control of generating units

at GS "D" on the stability of interarea mode
Type of Excitation Control
(a) Thyristor exciter with
high transient gain
(b) Thyristor exciter with
transient gain reduction
(c) Thyristor exciter with
power system stabilizer
Interarea Mode
Frequency (Hz)
Damping Ratio
0.192
0.009
0.187
-0.057
0.179
0.122
the interarea mode has a small positive damping (<=0.009). With a transient gain
reduction, the interarea mode actually becomes unstable (<= -0.057). When a power
system stabilizer (designed as described in reference 9) is added, the damping
increases significantly (t;=O.122).
Factors influencing interarea modes of osciZZation

The characteristics of interarea modes of oscillation are very complex and in
some respects significantly differ from the characteristics of local plant modes. Load
characteristics, in particular, have a major effect bn the stability of interarea modes.
822
Chap, 1
The manner in which excitation systems affect interarea oscillations depends on the
types and locations of the exciters, and on the characteristics of loads [23].
Speed-governing systems normally do not have a very significant effect on
interarea oscillations. However, if they are not properly tuned, they may decrease
damping of the oscillations slightly. In extreme situations, this may be sufficient aggravate the situation significantly. In the absence of any other convenient means of
increasing the damping, adjustment or blocking of the governors may provide some
relief [26].
A mode of oscillation in one part of the system may interact with a mode of
oscillation in a remote part due to mode coupling. This occurs when the frequencies
of the two modes are nearly equal [24,25]. Care should be exercised in interpreting
results of analysis in such cases.
The controllability of interarea modes with R B is a complex function of many
factors:
-
Location of unit with PSS

Characteristics and location of loads
Types of exciters on other units
On some units, the PSS does not have the desired effect on the damping of interarea
oscillations. Reference 24 presents results of a detailed study of factors influencing
PSS performance in damping interarea and interplant modes of oscillation.
Other effective means of stabilizing interarea modes of oscillation include
modulation of HVDC converter controls and static var compensator controls.
Analysis of interarea oscillations requires detailed representation of the entire
interconnected power system. Models for excitation systems and loads, in particular,
should be accurate, and the same level of modelling detail should be used throughout
the system.
Reference 27 provides a detailed account of a comprehensive study of the
interarea oscillation problems, including the fundamental nature of the problem,
methods of analysis, and control design procedures to mitigate the problem.
Enhancement of small-signal stability is discussed in Chapter 17.
REFERENCES
[I]
A.M. Lyapunov, Stability of Motion, English translation, Academic Press, Inc.,

1967.
[2]
G.C. Verghese, I.J. Perez-Arriaga, and F.C. Schweppe, "Selective Modal

Analysis with Application to Electric Power Systems, Part I: Heuristic
Introduction, Part 11: The Dynamic Stability Problem," IEEE Trans., VO~.,
PAS- 101, No. 9, pp. 3 117-3134, September 1982.
~eferences
[3]
823
J.G.F. Francis, "The QR Transformation - A Unitary Analogue to the LR

Transformation," Parts 1 and 2, The Computer Journal, Vol. 4, pp. 265-271,
1961; pp. 332-345, 1962.
[4]
J.H. Wilkinson, The Algebraic Eigenvalue Problem, Clarendon Press, Oxford,

1965.
[5]
W.G. Heffron and R.A. Phillips, "Effects of Modern Arnplidyne Voltage

Regulator in Underexcited Operation of Large Turbine Generators," AIEE
Trans., Vol. PAS-71, pp. 692-697, August 1952.
[6]
F.P. deMello and C. Corcordia, "Concepts of Synchronous Machine Stability

as Affected by Excitation Control," IEEE Trans., Vol. PAS-88, pp. 316-329,
April 1969.
S.B. Crary, Power System Stability, Vol. 11, John Wiley & Sons, Inc., 1955.
F.P. DeMello and T.F. Laskoswski, "Concepts of Power System Dynamic
Stability," IEEE Trans., Vol. PAS-94, pp. 827-833, MayIJune 1975.
P. Kundur, M. Klein, G.J. Rogers, and M.S. Zywno, "Application of Power
System Stabilizer for Enhancement of Overall System Stability," IEEE Trans.,
Vol. PWRS-4, pp. 614-626, May 1989.
P. Kundur, G.J. Rogers, D.Y. Wong, L. Wang, and M.G. Lauby, "A
Comprehensive Computer Program for Small Signal Stability Analysis of
Power Systems," IEEE Trans., Vol. PWRS-5, pp. 1076-1083, November 1990.
EPRI Report EL-5798, "The Small Signal Stability Program Package," Vol.
1, Final Report of Project 2447-1, Prepared by Ontario Hydro, May 1988.
S. Arabi, G.J. Rogers, D.Y. Wong, P. Kundur, and M.G. Lauby, "Small Signal
Stability Program Analysis of SVC and HVDC in AC Power Systems," IEEE
Trans., Vol. PWRS-6, pp. 1147-1153, August 1991.
N. Martins and L.T.G. Lima, "Eigenvalue and Frequency Domain Analysis of

Small Signal Electromechanical Stability Problems," Eigenanalysis and
Frequency Domain Methods for System Dynamic Performance, IEEE
publication 90TH0292-3-PWR.
R.T. Byerly, R.J. Bennon, and D.E. Sherman, "Eigenvalue Analysis of

Synchronizing Power Flow Oscillations in Large Electric Power Systems,"
IEEE Trans., Vol. PAS-101, pp. 235-243, January 1982.
Chap. 12
N. Martins, "Efficient Eigenvalue and Frequency Response Methods Applied

to Power System Small-Signal Stability Studies," IEEE Trans., Vol. PWRS.~~
pp. 2 17-225, February 1986.
L. Wang and A. Semlyen, "Application of Sparse Eigenvalue Techniques to

the Small Signal Stability Analysis of Large Power Systems," IEEE Trans.,
Vol. PWRS-6, pp. 635-642, November 1990.
N. Uchida and T. Nagao, "A New Eigen-Analysis Method of Steady-state
Stability Studies for Large Power Systems: S Matrix Method," IEEE Trans,
Vol. PWRS-3, pp. 706-714, May 1988.
F.L. Pagola, L. Rouco, and I.J. Perez-Arriaga,@Analysis and Control of SmallSignal Stability in Electric Power Systems by Selective Modal Analysis;
Eigenanalysis and Frequency Domain Methods for System Dynamic
Performance, IEEE publication 90TH0292-3-PWR.
W.E. Amoldi, "The Principle of Minimized Iterations in the Solution of the
Matrix Eigenvalue Problem," Quart. Appl. Math., Vol. 9, pp. 17-29, 1951.
Y. Saad, "Variations on Arnoldi's Method for Computing Eigenelements of
Large Unsymrnetric Matrices," Linear Algebra and Its Applications, Vol. 34,
pp. 184-198, June 1981.
P. Kundur, D.C. Lee, and H.M. Zein El-Din, "Power System Stabilizers for
Thermal Units: Analytical Techniques and On-site Validation," IEEE Trans.,
Vol. PAS-100, pp. 81-95, January 1981.
P. Kundur and P.L. Dandeno, "Practical Application of Eigenvalue Techniques
in the Analysis of Power System Dynamic Stability Problems," Proceedings
of Fi$h Power System Computation Conference, Cambridge, England,
September 1975.
-.
M. Klein, G.J. Rogers, and P. Kundur, "A Fundamental Study of Inter-Area
Oscillations," IEEE Trans., Vol. PWRS-6, No. 3, pp. 914-921, August 1991.
M. Klein, G.J. Rogers, S. Moorty, and P. Kundur, "Analytical Investigation of
Factors Influencing Power System Stabilizers Performance,"
Paper
92WM016-6EC, Presented at the IEEE PES Winter Meeting, New York,
January 1992.
D.K. Mugwanya and J.E. Van Ness, "Mode Coupling in Power systems," *
IEEE Trans., Vol. PWRS-2, pp. 264-270, May 1987.
References
825
[26]
O.W. Hanson, C.J. Goodwin, and P.L. Dandeno, "Identification of Excitation

and Speed Control Parameters in Stabilizing Inter-System Oscillation," IEEE
Trans., Vol. PAS-87, pp. 1306-1313, May 1968.
[27]
Canadian Electrical Association Report, "Investigation of Low Frequency

Inter-area Oscillation Problems in Large Interconnected Power Systems,"
Report of Research Project 294T622, prepared by Ontario Hydro, 1993.

Chapter 12 Small-Signal Stability

Uploaded by

Chapter 12 Small-Signal Stability

Uploaded by

small-Signal Stability

Chapter 2 provided a general introduction to the power system stability

12.1 FUNDAMENTAL CONCEPTS OF STABILITY

The behaviour of a dynamic system, such as a power system, may be described

We are often interested in output variables which can be observed on the

Fundamental Concepts of Stability of Dynamic Systems

where xo is the state vector x at the equilibrium point.

The stability of a linear system is entirely independent of the input, and

Fundamental Concepts of Stability of Dynamic Systems

Let us perturb the system from the above state, by letting

where the prefix A denotes a small deviation.

Since iio= f; (xo,u,) , we obtain

with i=1,2,..., n. In a like manner, from Equation 12.4, we have

12.1.4 Analysis of Stability

Lyapunov 's first method [I]

Lyapunov 's second method, or the direct method

Eigenproperties of .the State Matrix

of system response are related to the eigenvalues. Analysis of the

12.2 EIGENPROPERTIES OF THE STATE MATRIX

To find the eigenvalues, Equation 12.16 may be written in the form

For a non-trivial solution

Expansion of the determinant gives the characteristic equation. The n solutions of

bi has the form

Since Equation 12.17 is homogeneous, k9i (where k is a scalar) is also a solution.

is called the lefr eigenvector associated with the eigenvalue hi.

However, in the case of eigenvectors corresponding to the same eigenvalue,

where Ciis a non-zero constant.

I 2.2.3 Modal Matrices

diagonal matrix, with the eigenvalues Al, A2, ..., An

Eigenproperties of th,e State Matrix

from Equation 12.27

12.2.4 Free Motion of a Dynamic System

A set of equations of the above form, derived porn physical considerations,

The new state equation can be written as

In view of Equation 12.29, the above equation becomes

Eigenproperties of the State Matrix

The time dependent characteristic of a mode corresponding to an eigenvalue

A real eigenvalue corresponds to a non-oscillatory mode. A negative real

Complex eigenvalues occur in conjugate pairs, and each pair corresponds to an

has the form

(a) Mode shape and eigenvectors

Eigenproperties of the State Matrix

prernultiplying by @ i , and noting that qi+i

I& (A-IiI) = 0, we see that the

(c) Participation factor

The element pki=$kiyik

As we will see in a number of examples in this chapter, the participation

Expressing them in terms of the transformed variables z defined by Equation 12.31

Eigenproperties of the State Matrix

12.2.7 The Concept of Complex Frequency

Consider the damped sinusoid

With s =o+jo, we have

where Z(s) is the generalized impedance.

In the factored form,

Eigenproperties of the State Matrix

12.2.8 Relationship between Eigenproperties and Transfer Functions

This has the general form

If D(s) and N(s) can be factored, we may write

and Riis known as the residue of G(s) at pole pi.

Since A is a diagonal matrix, we may write

Eigenproperties of the State Matrix

This may be written in the standard form

undamped natural frequency

The output variable is given by

These have the standard state-space form:

The eigenvalues of A are given by

Eigenproperties of the State Matrix

Solving for the eigenvalues, we have

The right eigenvectors are given by

This may be rewritten as

and the eigenvector corresponding to 3L2 is

The location of the eigenvalues in the complex plane with respect to

< and anis

If the vector representing the initial condition is not an eigenvector, it can be

A, and A, real and negative;

12.2.9 Computation of Eigenvalues