You are on page 1of 6

IEEE T r a n s a c t i o n s on Power ApparatusandSystems, vol. PAS-94, no.

2 , MarchiApril 1975
NORMAL MODES AND MODESHAPES APPLIED TO DYNAMIC STABILITY ANALYSIS
R. T. Byerly D. E. Sherman D. K. McLain
Westinghouse Electric Corporation
Pittsburgh, Pennsylvania

ABSTRACT It is the intent of this paper and a companion paper5 to show


thatthe analysis ofdynamic oscillations involving low oscillating
Apowersystemof n generatorshas n-1 normalmodesof frequencies is considerably improved if the normal modes and mode
oscillation. If anyofthesemodes is poorlydamped,persistent or shapes are calculated by approximate methods in advance of detailed
disruptive oscillations mayoccur. Knowledge ofthenormalmodes simulation. Knowledge of normal modes and mode shapes enhances
and mode shapes of a specific system provides important insight into comprehensionofsystem behavior and provides data which can be
its dynamic behavior. Calculation of normal modes and mode shapes used as input to thesimulation program to make its use in this kind of
yields data that permits a systematic approach to dynamic stability study more efficient.
studies.
LINEARIZATION OF CLASSICAL EQUATIONS
The theoretical basis for normal mode analysis of power systems
is presented. Methods for calculating normal modes and mode shapes For undamped natural frequencies to exist in asystem, it is
are described,withillustrations to small systems.Techniques for necessary that the differential equations which describe it be of the
simulating normal modes for the analysis of mode damping are shown. form :

A companion paper describes the application of modal analysis x+ax=o (1)


to a large area in the MAPP system.
In order to achieve this, the generator acceleration equations are
INTRODUCTION written without adamping term and the turbine powei is assumed
constant. Generators are represented by the classical model, a constant
There have been numerous instances of dynamic oscillations in voltage magnitudebehindtransientreactance. As the power flow
North American power systems over the past ten years, primarily in equations are nonlinear functions of rotor angles, the system will be
the western United States andin parts of Canada. 1-4 Analyses of such linearized at an initial operating point. This is done by assuming that
oscillations ntay be quite complex and are largely dependent on digital the rotor angles are given by a steadystate value plus an incremental
simulation in which generators,excitationsystems,powersystem deviation.
stabilizers, speedgoverningsystems and loads are represented as
precisely as possible within the limits of feasibility. This attention to e 1. = e .10 + ai (2)
detail in simulation is required to account for various factors which
are known to influence the damping of oscillations. It is generally For each generator an acceleration equation can be written
assumed, however, and apparentlywithjustification, thatthe fre-
quency of an oscillation is primarily determined by generator inertias,
generationandnetworkreactances and average angular differences M~ ei = pTi -pEi
between generators. The authors of the four papers cited above, for
example, imply these assumptions in explaining actual system oscilla- Since Oio is a constant, itfollows from (2) that
tions they have observed over the range of 0.1 to 0.83 Hz. .. ..
8. = 6.
1 1 (4)
Beginning with a classical representation, it is possible to show,
with some assumptions, that an electric power system of n generators and (3) can be written as
has (n-1) normal modes of oscillation (natural frequencies). In many
power systems, all ofthesefrequencies are positively damped and Mi 6i -=PPET~~ (5)
spontaneous oscillations never occur. In other systems, most of them
are positively damped, but oscillations at some frequencies may occur
spontaneously or persist after a disturbance. Poorly damped oscilla- In order to eliminate pEi from (5) it is necessary to perform a network
tions may involve significailt speed variations for only one machine or reduction. If the loads are converted to impedances and +e generator
plant which moves with respect to the rest of the system. Such internal to terminal impedances are all considered to be XD, the nodal
oscillation frequenciestypically lie in the range of one to two Hz. admittance equations may be written and then simplified until only
Often, however, there may be significant speed deviations for a large the generator internal buses remain. The electrical generation for each
number of machines. Theseoscillationstend to be at lower fre- machine can then be written in terms of generator internal voltages
quencies, typically in the range of 0.1 to 0.5 Hz. The latter present the and transfer admittances.
most difficult analysis problems for a variety of reasons. Because of
the many machines involved, it may be difficult to obtain a clear n n
comprehension of what is happening, either from operating data or
from simulation, and the principal factorscontributing tothe
P E ~ = - E Gii
t -x
J= 1
EiEjGij cos(6i-0)) -x
j=1
EiEjBi, sin(ei-Oj) (6)
oscillation and best remedies are both obscured. System sizej prequired
i j#i
for simulation tends to be larger when more machines are involved,
and the time span of simulation must be longer because of the lower Gij and Bij are the real and imaginary components of the offdiagonal
oscillation frequency, both factors increasing the cost of simulation. terms of the reduced admittance matrix and Gii is the real component
of the diagonal term.The minus signs appear because the original
nodal equations were written in the form of current into the bus. This
means thatthe original offdiagonal terms are the actual branch
admittances and, in general, the imaginary componentsofthe
offdiagonal t e r m in the reduced matrix will have the same algebraic
signs as normal susceptance terms; that is, Bij < 0.
Equation ( 6 ) must be linearized. From (2)
Paper T 74 431-3, recommended and approved by the IEEE Power System e.-e. = (e.1 0-e.JO) + ( ~ ~ - 6 ~ )
Engineering Committee of the IEEE Power Engineering Society for presentation at 1 J
the IEEEPESSummerMeeting & Energy Resources Conf., Anaheim, Cal., July
14-19, 1974. Manuscript submitted January 31, 1974; made available for printing
May 10,1974. From (7) form the approximations
224
and

Using these approximations in ( 6 ) and then replacing PE in (5) yields b.. = -E.E.B..
U 1 J U
cos(eio-ejo)
the linearized acceleration equation:

n
M~ iii = pTi + q 2 G~ +x j= 1
E.E.G..
1 J 1J [cos(eio-ejo)-(6.-6.)
1 J sin(eio-ejo)l bii
n
=IEiEjBij cos(Oio-Bj0) -1U
j=1
=
j= 1
n
b..

j#i j#i j#i


(10)
n Unless there are phase shifters in the system,Gij= GjiandBij=Bji.Ako,
since sin(x) = -sin(-x) and cos(x) = cos (-x), it follows that the matrix
+ E ~ E [sin(eio-ejo)
~ B ~ ~ + ( ~ ~ - cos(eio-ejo)l
6 ~ ) [gijl is skew-symmetric andthe matrix [bij] is symmetric. As will
j= 1 become apparent, it is a computational necessity that the coefficient
j# i matrix be symmetric. With this need in mind, it is assumed that the
dominantpart of (14) is thematrix [bj] andthe problem can be
In the steady state therewill be no acceleration, and follows that: reduced to the symmetricsystem of equations:

n
pTi + ~~2 G~~+ 1E ~ E cOS(eio-ejo)
~ G ~ ~
j=1 Dividing each individual equation by thecorresponding Mi gives
j#i
A6 = 6
(1 1)
+xn

j= 1
~ B ~ ~ =o
E ~ E sin(eio-ejo)
The calculation of normalmodes and modesshapes can now be
j#i expressed as an eigetlvalue problem.* If the incremental rotor angle
Consequently, (10) simplifies to: oscillations are described by:

6 = p sin ( u t ) (22)
where 6 and p are vectors of dimension n,

then d = /Iw cos ( a t )


and i = -pw2 sin(wt)
Substituting (22) and (24) into (20)
3
Ap sin(wt) = - p a - sin(wt)

If (25) is to be valid for all values of time,


REDUCTION TO THE FORM OF AN EIGENVALUE PROBLEM
(A+w21)p=0 (26)
It is convenient to separate the 6i terms and rewrite ( 1 2) as:
must also be satisfied. If the eigenvalues Aj and eigenvectors $ are
found such that:
r
~ 1. 1g .
j= 1
n
=l-C E ~ E sin(eio-ejo)]
~ G ~ ~
-
6i +x
~

j= 1
n
E ~ E sin(eio-ejo)
~ G ~ ~ sj (A -$I) 4=0 (27)
then the natural frequencies are given by
j#i
WJ =q (28)

The corresponding eigenvectors indicate the relative amplitudes of


the generator rotor oscillations at a given frequency as can be seen
from (22) and also indicate the relative velocity deviations as shown
by (23).

By using (1 3) for each generator, the following system of differential


equations can be written: *The assumption that the matrix [gjl is neghgible is not required to
arrive at the form of an eigenvalue problem. It was made at this point
in thedevelopmentsimply to avoid increasing the notational
complexity. It would have been possible to assume at the outset that
conductancesare neghgible; however, the more general form is of
interest and would be used if the computatiorial difficulties could be
overcome.
225
SOLUTION OF THE EIGENVALUE PROBLEM matrix T as shown in (39) The eigenvalues and eigenvectors of
interest are computed by applying the Power Method to the matrix T.
Equation (27) must be modified several times before it is actually It should be noted that one eigenvector is known. The eigenvector
solved. The solution method chosen was the Power Method6 due to its corresponding to Aj = 0 for S and to Qj AB for T is given by
simplicity. Given an initial guess at an eigenvector, the method tends
to converge tothe eigenvalue having largest absolute value and
corresponding eigenvector. In order to find the next eigenvalue, it is
necessary to have an initial guess vector which does not have any
component in the direction of the first eigenvector. As shown in the The initial guess vector is arbitrary except that it should be chosen
Appendix, this can be done easily if the eigenvectors are orthogonal. such that it does not have a component in the direction of po. The
Orthogonality, inturn, requires thatthe matrix be symmetric. Power Method should then converge to g l , the f m t eigenvalue less
Although the matrix in (1 9) is symmetric, the matrix A introduced in than AB, and the corresponding eigenvector p 1. From these it follows
(20) is not. To return to asymmetric form, perform the following that :
change of variable :
X = D6 (29)
and the lowest natural frequency is
where D = diag (&, &2, ..., f i n ) (30)

[DADm1] [D61= Ds (3 1)
The corresponding eigenvector is
or sx=x (32)

where s = DAD- 1 (33)


The next lowest natural frequency is determined by starting with an
The matrix S is symmetric, with initial guess vector which does not have components in either the po
or p l directions, The process is continuedfor as many natural
% bii/Mi (34) frequencies as are of interest. Experience has shown that usually fewer
thanten eigenvalues need be calculated to ob* the natural
frequencies less than one Hz.
sij and = sji = b i j / a (35)
ILLUSTRATION OF THE USE OF NORMAL MODE DATA
The eigenvalues of S are still Xj, but the eigenvectors pj are rotated
such that: The hypothetical system of six generators shown in Figure 1 is
used to illustrate the use of normal mode data in dynamic stability
&=D-l& (36) analyses. This system has five natural frequencies, four of which are
calculated and presentedhere. The normalmode program is not
The second modification is required by the fact that the Power designed to calculate the relative velocity deviations for individual
Method converges to the eigenvalue having largest absolute value, but generators at the highest frequency and t h i s frequencydoes not
the natural frequencies of interest are the low ones. This problem is appear in t h i s discussion.
solved by forming a new matrix T such that the largest eigenvalues of
T are the smallest eigenvalues of S. An examination of (1 7) and (18) II 4 2 I
shows that there will be a zero eigenvalue and in general all others will
be negative.

Let

Then the eigenvalues of S are:


Xj< 0

Aj= 'Xj I

- AB ,...,-Aj,. ..,O
By applyingthe Power Method to thematrix
(37)

(38)

S it is possible to
7
determine AB. Then form

T=S+AgI (39)
Fig. 1. Hypothetical system for illustrating normal mode and mode
The eigenvectors of T are the same as those of S but the eigenvalues qj shape applications to dynamic stability analysis.
lie in the range
Detailed data for loads, lines, transformers and generating plants
0,..., ~ j ,...,AB (40) are omitted to conserve space. It is sufficient to note that the data for
machines at buses 3, 6 , 1 1 and 12 are typical for modem steam and
When an eigenvalue is found, the corresponding hj is .&en by hydro generating plants. The machine at bus 4 is asynchronous
condenser. Bus 1 represents a system equivalent which can be either a
hj = qj -AB (40) net load or generator. The inertias and internalimpedance of this
equivalent generator are selected to createwhat is essentially an
If there is a positive Xj, the method i s the same but some care must be infinite bus, and in all simulations described later, theequivalent
exercised as a positive Xj corresponds to a non-oscillatory solution. machine is represented classically.

COMPUTATIONAL PROCEDURE Two operating situations are presented to show the dependence
of normal modes and mode shapes on changing conditions. The first
The overall procedure is to first determine the linearized matrix situation, designated case 1, represents an approximate balance
Ashown in ( 2 0 ) . Performthe change of variable shown in (29) between load and generation so that the net flow to the system
through (35) to producethesymmetricmatrix S. Apply the Power equivalent bus is quite small. Thesecond situation, case 2, assumes
Method to S until an estimate for AB is determined. Then form the lighter loads with the original generation, so that a relatively high
226
power flow to the system equivalenf atbus 1 exists. Generator simulation of a normal mode. Figure 2 shows the results of simulating
internal voltages and angles behind XD are presented in Table 1 for the lowest frequency for case 1, while Figure 3 shows results of
both cases. simulating the highest frequency; that is, 2.016 Hz. Only three rotor
angle deviation curves are plotted to avoid complicating these figures.
TABLE I These particular simulations are made with classical generator models.
In lieu of the customary transient disturbance, each machine is given
Internal Voltages and Angles an initial velocity deviation proportional to its deviation in Table 11,
with all deviations scaled to obtain some desired peak angle swing. The
Generator result is a simulation of the desired natural frequency relatively free of
Bus No. case 1 Case 2 transients and other frequencies.

1 .................. 1.05; 00 1.05; 0'


3 .................. 1.05;6.2O 1.05; 38.6'
4 .................. 1.00;0.3° 1.OO; 5 1.2O
6 .................. 1.05;34.3' 1.05; 87.2O
11 . . . . . . . . . . . . . . . . . . 1.O5;6.Io 1.05; 62.6O
12 . . . . . . . . . . . . . . . . . . 1.O5;6.Io 1.05; 62.6'
The four lowest natural frequencies for case 1 are listed in Table
11, along with the relative peak generator speed deviations. Speed
deviations are normalized so that the largest deviation is unity. The
system equivalent generator at bus 1 is virtually stationary because of
its assumed large inertia.

TABLE I1
NORMAL MODES AND MODE SHAPES FOR CASE 1

Generator Relative Peak Speed Deviations

Natural I
Bus Bus
Frequency
Bus Bus Bus I 2 3 4 5
in Hertz 1 3 4 6 11 12 SECONDS
0.485 . . . . . . . .0 0.337 0.521 1.0 0.564 0.633 Fig. 2. Simulation o f lowest normal mode forcase 1. Generators
0.945 . . . . . . . .0 0.292 0.424 -0.463 0.568 1.0 are classical and loads are assumed to vary linearly with bus
1.627 ........ 0 1.0 0.387 -0.050 0.187 -0.270 voltage. Oscillation period from normal mode calculation
2.016 . . . . . . . .0 -0.582 1.0 -0.041 0.556 -0.224 is 2.06 seconds.

The mode shape forthe lowest naturalfrequency, 0.485 Hz.,


shows all machines oscillating coherently with respect to the system
equivalent at bus 1. The machine at bus 6 has the largest speed 20
IO-0.5 SECONDS-
deviation, and all other machines move in phase with bus 6, but with
lesser deviations. The mode shape forthe secondlowestfrequency u)
W
shows the generator at bus 12 having the largest deviation, with the W
a
generator at bus 6at theopposite extreme,and oscillating out of w IO
W
phase with bus 12. Each of the other frequencies shows a different a
machine having the maximumdeviation, and, of course, s h o w a f
different mode shape as well. Experience with very large systems has 2
shown that the same machine may have the maximum deviation at 0 0
I-
more than one natural frequency, although mode shapes are unique. s>
W
Table 111 presents natural frequencies and mode shapes for case a
2, in which power is being supplied from generators 3 , 6 , 11 and 1 2 to y -10
the system equivalent. The data in Table 111 is very similar to that in W
Table 11. The most significant change is thesubstantialreduction in 3
the lowest frequency. Judging from this, real systems may be expected
to exhibit some variation in lowest naturalfrequencies as operating 1
conditions change. Since lower frequenciestend to be less well a2 0.4 0.6 OB I.o 12
damped, such changes in operation may well cause a transition from SECONDS
dynamic stability to dynamic instability. Fig. 3. Simulation o f highest normal mode forcase 1. Generators
are classical and loadsare assumed to vary linearly with bus
TABLE 111 voltage. Oscillation period from normal mode calculation is
NORMAL MODES AND MODE SHAPES FOR CASE 2 0.496 seconds.
Natural a n e r a t o r Relative Peak Speed Deviations It should be notedthatthe nonlinearities of thesystem are
Bus Bus
Frequency
Bus Bus Bus Bus incompatiblewith the existence of pure sinusoids in simulation.
in Hertz 1 63 4 11 12 Maximum angle differences tend to be prolonged due to the nonlinear
power transfer characteristics of ac systems. In addition, the transition
0.355 . . . . . . . . 0 0.444 0.621 1.0 0.691 0.740 from stationary to oscillating conditionsproducesa change in the
0.903 . . . . . . . . 0 0.383 0.423 -0.565 0.597 1.0 average value of voltage sensitive loads which excites transients and
1 . 5 6 2 . . . . . . . . 0 1.0 0.393 -0.050 0.180 -0.325 other frequencies.
2 . 0 5 2 . . . . . . . . 0 -0.636 1.0 -0.047 0.681 -0.241
Themotivationforsimulating any naturalfrequency is to
Knowledge of naturalfrequencies and mode changes is of determine whether or not it is damped. The curves of Figure 4 show
considerable value in understanding large systembehavior, and in t h e simulated behavior of the generator at bus 1 1 forthe lowest
evaluating the effects of changes in operating conditions or in design. frequency of case 1 with three different conditions which alter system
The frequency deviations also provide data which permits the damping. Curve 1 is from a simulation in which all machines except
221
equivalent
the at1bus have detailed
generator models with frxed field APPENDIX
voltages. It is slightly damped. Curve 2 is from a simulation in which
all machines except bus 1 have automatic voltage regulators, and it is Choice of Initial Guess Vector
dynamically unstable. Curve 3 is from asimulation which includes
powersystem stabilizers at generators 3, 6, 11 and 12. Swing curves To avoid convergence tothe alreadyknown eigenvector po,it is
for generators at other buses are similar to thoseshown for bus 1 1. necessary to have an initial guess vector u which does not have a
component in thedirectionof po. An arbitraryvectorv may be
This method of forcing specific normalmodes provides an expressed as a linear combination of po andtheunknownvector u,
efficient and economicprocedure for analyzing systemdamping. which is orthogonal to po.

Since u is to be orthogonal to po, their inner productmust be zero.


T
(PO) u = 0 (I-3)
Consequently, (1-2) can be solved for the scalar k.

SECONDS

Fig. 4 . Sirnulotion of lowest normal mode forcase 1. Then (1-1) can be solved for the initial guess vector u.
These curves are for generator at bus 11. All loads are assumed to vary
linearly with bus voltage.
Curve 1 - Fixed field voltage on all generators
Curve 2 - Voltage regulators on all generators As more of the eigenvectors, become known, it is necessary that u be
Curve 3 - Power system stabilizers on all generators orthogonal to all known pJ. This can be achieved by repeating the
computations foreach pJ and replacing v with u at each step.
CONCLUSION

The analysis of dynamicstability is greatly improved by a


knowledge of the natural oscillation frequencies which may be excited
by a system disturbance or occurspontaneously. A mathematical basis
has been developed, and a procedure for calculating these frequencies
and the mode shapes which correspond to them has been presented.
Examples have been given to show the use of the mults of REFERENCES
such
computations in the investigation of system damping. The techniques
presented have been successfully applied to analyses of real power 1. F. R. Schleif and J. H. White, “Damping for theNorthwest-
tystems consisting of as many as 187 generators.
Southwest Tieline Oscillations-An Analog Study.”
IEEE
Trans.
Power Apparatus and Systems, Vol. PAS-85, pp 1239-1247.
SYMBOLS
2. F. R. Schleif, G. E. Martin and R. R. Angell, “Damping of System
Voltage Magnitude Oscillations with aHydrogenerating Unit.” IEEETrans. Power
Apparatus and Systems, Vol.PAS-86, pp 438442.
Identity Matrix
3. 0. Hanson, C. G. Goodwin and P. L. Dandeno,“Influence of
Generator Inertia Excitation and Speed Control Parameters in Stabilizing Intersystem
Oscillations.” IEEE Trans. Power Apparatus and Systems, Vol.
Generator Electrical Power PAS-87, pp 1306-1313.

Turbine Power 4. F. R. Schleif, H. D. Hunkins, C. E. Martin and E. E. Hutton,


“Excitation Control to Improve Powerline Stability.” IEEE Trans.
Generator Rotor Angle Power Apparatus and Systems, Vol. PAS-87, pp 14261434.

Steady State Rotor Angle 5. D. L. Bauer, Wm. D. Buhr, S. S. Cogswell, D. B. Cory, G. B.


Ostroski and D. A. Swanson,“Simulation of Low Frequency
Incremental Rotor Angle Change Undamped Oscillations in Large Power Systems.” Submitted for
the 1974 Summer Meeting of the IEEE Power Engineering Society.
Rotor angles and internal voltage angles are the same since the
classical generator model is used. 6. V. N. Faddeeva, Computational Methods of Linear Algebra. New
York: Dover Publications, Inc., 1959, pp 202-21 1.
Absolute Value of Largest Eigenvalue

228
Discussion REFERENCES
J. M. Unddl and F. P. deMello (Power Technologies, Inc., Schenectady, [ 1 ] William A. Mittelstadt, “Four Methods of Power System Damping”,
N. Y.): We compliment the authors onthis useful paper, and agree that IEEE Trans. on Power Apparatus and Systems, Vol. PAS-81, No. 5,
the direct analysis of the system’s inherent dynamical characteristicsis a May 1968, pp. 1323-1329. (In particular, Figures 7 and 14)
valuable aid to the interpretation of simulations, giving valuable insight [2] P.C. Magnusson, Discussion to Reference 1.
as to the origin of the phenomena observed.
In addition toexamining natural frequencies, we find the consider-
__
131P.C. Mamusson.“Transients in CouDled InductanceCaDacitance
Circuits b a l y z e d in Terms of a Roiling Ball Analogue”, AZEE
ation of system damping to be a valuable aspect of these direct analyses Trans., Vol. 69,pt. 11, 1950, pp.1525-1530.
The technique for determining the eigenvalues and eigenvectors (mode
shapes) of large systems including detailed generatorand control model-
ing is well established in papers such as References A and B.We agree
that the classical model used by the authors is adequate for the deter-
mination of rotor swinging natural frequencies but, because it neglects R T. Byerly, D. E. Sherman, and D. K.McLaii (Westinghouse Electric
all damping, we feel that the eigenvectors it produces mightbe distorted Corporation, Pittsburgh, Pa.): The authors wish to express their appre-
fromtheactual mode shapes existing in the presence of high gain ciation to Mr. Magnusson and to Mssrs. Undrill and de Mello for their
controls and supplementary stabilizing loops. The authors might like to interest in this paper.
commenton comparisonsof themode shapes computed fortheir The modeshapes calculated by methods described in this paper do,
classical modelandthose observed in simulation runs using detailed of course, differsomewhat from thoseobtainedfromsimulation.
system modeling. Consider, for example, the following four mode shapes for the hypo-
In our experience, the problem has not been finding the natural thetical system of Figure 1 for the natural frequency of 0.485 Hertz:
frequencies, buthas been to determine which of the modes are
negatively damped and hence are the source of negative damping on the Mode
system as a whole. Our studies of several large interconnected systemsBus ShaDe Bus Bus Bus Bus
have revealed that the lightly and negatively damped 6modes are not 4 No. 3 11 12
necessarily those with the lowest natural frequencies.
0.337
We have not found it necessary or desirable in our system damping 1 0.521 1.0 0.564 0.633
studies to set up a special fictitiousdisturbance to excitea specific 2
- 0.298 ~ 0.463 1.0 0.502 0.566
normal mode, although we agree that it is a great convenience when 3 0.456 0.283 1.0 0.482 0.563
concentrating on one troublesome mode. We have found that modes 0.298 4 0.441 1.0 0.609 0.507
with unacceptably low or negative damping grow quite rapidly to easily
observable magnitudes following practical system disturbances. The f i t of these is copied from Table I1 and is the mode shape obtained
using the models and methods of the paper. The second is derived from
REFERENCES asimulation using classical generator models. This simulation is the
basis for Figure 2 which shows an essentially undamped oscillation. The
[ A ] M. A. Laughton, “Matrix Analysis of Dynamic Stability in Syn- third mode shape is from a simulation in which generators and excita-
chronous Multi-Machine Systems”, Proceedings IEEE, Volume 1 13, tion systems are represented in detail. The system is dynamically un-
p. 325, 1966. stable, with a growing oscillation as shown by curve 2 of Figure 4. The
. . J. M. Undrill. A. E. Turner. “Construction of Power System
IB1 fourthmode shape is based onasimulation which includespower
Equivalents by Modal Analysis”, IEEE Transactions, Volume PAS- system stabilizers on each machine. The oscillation is damped as in-
90, pp. 2049-2059, 1972. dicated by curve 3 of Figure 4. This data is subject to various interpre-
tations, but it appears that the mode shapes based on simulation are
Manuscript received July 23, 1974. quite similar to each other in spite of differences in damping.
This isnotto say thatthe assumptions made in modeling the
system for eigenvalue calculations do not influence mode shapes. In-
tuitively, however, naturalfrequenciesand mode shapes are more
dependent on synchronizing torques than on damping torques. Thus,
Philip C. hiagnusson (Bonneville Power Administration, Portland, for example, the difference in the f i s t and second mode shapes is due
Oregon): The authors have presented an excellent analysis, in the light entirely to differences in synchronizing torques, since neither the
of modemcomputer capability, of aproblem which intrigued t h i s eigenvalue calculation northe classical simulation involve damping.
writer some years ago. Specifically, for a three-machine system,a There are two apparent reasons for the difference in these mode shapes.
three-space geometrical interpretation of the kinetic and potential ener- The f i s t is that the angle variations in the simulation are large enough
gies of the system, the “rolling-ball analogue”, is feasible and instruc- to constitute a significant departure from the assumption of linearity
tive.[ 1,2,3] Let the following generalized coordinates be adopted, and inherent in the eigenvalue calculation. The second is that real compo-
angle fi d e f i e d as indicated: nents of load influence the eigenvalue analysis only to the extent that
equivalent load resistances affect the susceptances of the reduced
network.
It should be noted that the assumption of constant voltage behind
transient reactance has the potential forintroducing significant errors in
the eigenvalue and eigenvector calculations at very low frequencies.
Without automatic excitation, generator internal voltages will vary sub
stantially during low frequency oscillations. On the other hand, terminal
voltages may be virtually constant under the influence of a strong auto-
matic excitation system.
Here 6 13 and 623 are the relative rotor angles between machines 1 and Whilewe agree with Undrill and de Mello that the techniques of
3, and 2 and 3, and the J’s are the moments of inertia of the respective eigenvalue analysis are well established, we are not aware of their prior
machines If these coordinates are oriented with respect to each other at use for systems as large as the MAPP system.[5] The classical system
the angle 6, the loci of motion in the normal modes will be perpendi- model was chosenas the basis for these calculations to provide the
cular to each other. Thepotential energy required to displace the economical computation and simple interpretation needed to support
system from the equilibrium arrangement may be plotted normal to the such large system planning and operating studies.
Plane of 413 and 923 to form a curved surface. For small displacements We agree thatthedetermination of damping is difficultand
from the equilibrium portion, contours of constant potential energy will expensive. In the case of the MAPP system, damping analysis depended
be concentric ellipses of the same eccentricity, and the normal-mode on manylong term dynamic simulations. Forthis very reason, the
loci will coincidewith the axesof those ellipses. (Themathematical eigenvalue analysis is crucial, providing comprehension of system be-
analysis has been developed more fully for the analogous problem of havior and an orderly and efficient method of simulating only the
oscillations in two inductively coupled L C circuits[3] ,) natural frequencies of interest.
Manuscript received August 1, 1974. Manuscript received October30,1974.

229

You might also like