Power

Graham Rogers *

ower systems are among the most ues that maintain the required power
complex dynamic systems created flows through the transmission net-
by man, and the insidious nature of work and supply the system’s loads.
oscillatory instability has mystified If the disturbance is large, say a pro-
many practicing engineers. With mod- longed three-phase fault on the trans-
ern computer programs, a thorough mission system, the nonlinear nature
study of power system oscillatory sta- A common problem of the synchronizing torque may not
bility is possible. The necessary com- in small and large be able to return the generator angles
plex mathematics are made
transparent to the program user, and
power systems is the ttors o a steady state. Some or all genera-
then lose synchronism and the
the emphasis of studies may be placed insidious nature o f system exhibits transient instability.
on the physical nature of the oscillato- oscillatory instu bility On-theother-hand, if the disturbance is
ry phenomena and the accuracy of the small, the synchronizing torques keep
model data. Careful control design and the generators nominally in synchro-
consistent control commissioning and tuning a r e nism, but the generator relative angles oscillate. In a cor-
required to damp power system oscillations sufficiently rectly designed and operated system, these oscillations
so that system operating limits are caused by other phe- decay: t h e system is small-signal stable. In an over-
nomena. stressed system, small disturbances may result in oscil-
lations that increase in amplitude exponentially; the
Nature of Power System Oscillations system is then said to be small-signal unstable.
Smaller power systems have in the order of hundreds of Unstable power system oscillations have occurred all
miles of transmission lines, while the largest (the eastern over the world in the last 30 years. They appear first
U.S./Canadian interconnected system) has thousands of when a power system is pressed to supply increasing
miles of transmission lines. The electric power is largely load. As transmission lines are loaded more and more,
alternating and at a frequency that is almost uniform the generators need to rely more heavily on their excita-
over the whole network. This is achieved by using syn- tion systems t o maintain synchronism, and a t some
chronous ac generators. System frequency is held within point, without supplementary control, the synchronizing
tight limits by speed governing the generator prime- oscillations become unstable. Also during the last 30
movers, and system voltages are held by generator exci- years, many power systems have become interconnect-
tation system colntrol. In small systems, there may be ed so as to be able to exchange power to keep operating
only tens of generators; in large systems, there are thou- costs to a minimum. However, the interconnecting ties
sands. between neighboring power systems, although not over-
Interconnected ac generators produce torques that loaded, are often relatively weak when compared to the
depend on the relative angular displacement of their connections within each system. The synchronizing
rotors. These torques act to keep the generators in syn- torques are lower across these weak ties, and this, cou-
chronism (synchronizing torques), thus, if the angular pled with the high aggregate inertia of each of the sys-
displacement bel ween generators increases, an electri- tems being interconnected, leads t o low frequency
cal torque is produced that tries to reduce that angular interarea oscillations. Many of the early instances of
displacement. It is as though the generators were con- oscillatory instability occurred at low frequencies when
nected by torsional springs, and, just as in mass-spring interconnections were made.
systems, the moment of inertia of the rotors and the syn- A common problem is the insidious nature of oscilla-
chronizing torqules cause the angular displacements of tory instability. Power flow over a tie may be increased
the generators t o oscillate following a system distur- to supply remote load with no noticeable problems until
bance. The angular displacements should settle to val- the stability limit is reached. A slight increase in power
flow beyond this limit results in oscillations in which
* Cherry Tree Scientific Software amplitude increases quickly with no need for any system

30 IEEE ComputerAp,ulications in Power ISSN 08950156/96/$5.0001996 IEEE

fault. At best, system nonlinearities used to determine possible steady-
limit oscillation amplitudes. At state conditions that satisfy all the
worst, the oscillation amplitudes systems steady-state design con-
reach levels at which protective straints. The nature of power sys-
relays trip lines and generation, and tem dynamics is nonlinear, but for
this in turn causes partial or total small changes to an operating point,
system collapse. the nonlinearities may be removed.
The differential equation model of
Need for Ainalysis the system may be linearized, using
Two types of synchronizing oscilla- t h e first term of a Taylor series
tion are common in all interconnect- expansion. It has been shown over
ed ac power systems. The first is many years that, in almost every
associated with a single generator case, the analysis of a linear power
(or a plant of identical generators) acting against the sys- system model gives results that agree with observations.
tem. The second is more complex and involves many gen- Small interconnected power systems may have, say,
erators in one area of the power system oscillating against 20 generators. About 10 first-order differential equations
other generators in other areas of the power system. are required to describe the dynamics of each generator
Local or plant modes of oscillation have natural frequen- and its controls. Thus, the small system model would
cies of about 1 to 2 Hz. Interarea modes of oscillation have require 200 equatioas to describe the system dynamics.
lower natural frequencies in the order 01 0.1 to 0.7 Hz. In The eastern U.S./Canadian interconnected power system
small systems, interarea oscillations generally have higher is often modeled with 2,000 generators, requiring up to
natural frequencies than those of large systems. 20,000 first order equations to describe its dynamics.
The total number of modes of synchronizing oscilla- The process of modeling and analysis is clearly a job for
tion is equal i o one less than the number of intercon- computers, even when the equations are linearized.
nected gener,ators. In a system having thousands of Linear equations have properties that enable their
generators, there are thousands of modes of oscillation. characteristic dynamic modes of oscillation to be calcu-
All of these mlodes of oscillation must decay following a lated in a systematic manner. The method of calculation
system disturbance. If any one mode increases in ampli- is based on the determination of functions of the follow-
tude, the system's operators would have to take action ing general form, which satisfy all the unforced system
to prevent either local or system-wide collapse. equations simultaneously:
Oscillations must be designed to be stable under as
wide a range d system loads and operating conditions exp(alt) (At sin(w,t) + B, cos(w,t))
as possible. Generally, if the operation of the system is
constrained, those constraints should be caused by the If the value of a is negative for all functions, then the
thermal operaiting limits of the transmission system or equilibrium point is stable. If the value of a is positive for
loss of synchironism (transient instability) and not by any function, then t'he system is unstable; that is, the tran-
oscillatory instability. sient following a small disturbance will grow in magnitude,
To determine the nature of system oscillations, analy- rather than decay. 'The value of w may be zero, in which
sis is requiredl that shows the following system charac- case the corresponding transient is aperiodic. If w is not
teristics: zero, the correspoinding transient will oscillate at a f r e
IFrequency and damping of the sysi ems synchroniz- quency of w/27c. Each function that satisfies the system
ing oscillations equations is called a characteristic mode of the system.
I Pattern of generators that take part in each oscilla- Normally a, and w, ;He given as a set of complex numbers
tion. (1,= a, + wJ, which are called characteristic values or
Generators that are able to have a control effect on eigenvalues. The number of characteristic values is equal
the oscillationis must also be identified, and tools must to the number of first order differential equations required
be provided tlo allow for efficient and robust oscillation in the system model. The values of A, and B, are arbitrary
damper control design. constants. For a particular case, they depend on the char-
acteristic value, the location, and type of disturbance, and
Analysis To~ols the point in the system at which the mode is observed.
Stable oscillations remain small in amplitude or decay. If a disturbance is applied and then quickly removed
Following a disturbance, the system settles to a steady (a common occurrence in power systems), the relative
equilibrium state. Power systems, however, are continu- amplitude of each characteristic mode throughout the
ously disturbed by changes in loads. These changes are system's dynamic variables is a property of the system
normally smalll, and the system's operating state is very called a characteriistic vector or eigenvector. In power
close to a constant equilibrium. Power flow analysis is systems, this is often called t h e mode shape. The

July 1996 31
 Determine the cause of unstable
oscillations
Design controls t o extend t h e
range of oscillatory stability.
A simple example (a drastic equiva-
lent of the New England / New York /
Ontario system) is used; the one-line
diagram for this system is shown in Fig-
ure 1. While the system is small and so
drastically simplified as to be somewhat
artificial, it does show clearly many of
the characteristics of larger power sys-
tem oscillations.

Stability of the Operating Point

Unstable modes are identified by char-
acteristic values having positive or zero
real parts. In power systems, it is nor-
Figure 1. One-line diagram of the New England standard test case mal to operate the system, or design the
systems controls, so that each mode
response to more complex disturbances can be calculat- decays by at least exp(-O.lz) during every cycle of the
ed in terms of the basic system modes. oscillation. If all modes have real parts more negative
Given the point at which an observation is made and than the limit, we consider that the system is sufficiently
the point at which a disturbance is applied, it is possible stable at that operating point. Normally, in any study, a
(for a stable system) to calculate the maximum ampli- number of different operating points thought to be criti-
tude of any mode caused by an impulse at the distur- cal a r e studied. These generally have high loading
bance point. The mathematical term f o r this is t h e and/or have critical transmission lines out-of-service.
residue; it is the numerator of the term in the partial frac- To illustrate this, the eigenvalues of the test system
tion expansion of the transfer function between the dis- are shown on an argand diagram (x-axis real, y-axis imagi-
turbance and the observation. For example, given a nary) in Figure 2. In this system model, the 9 eigenvalues
disturbance in the controls of one generator, by using associated with the electromechanical synchronizing
residues, we can determine the size of oscillations that oscillations are insufficiently damped, indeed a number
may be observed in any of the systems generators. have positive real parts showing that the system has
Eigenvalue (olr modal) analysis is performed using unstable oscillations. This is because, in this model of the
special linear power systems analysis programs. These
determine the eigenvalues and eigenvectors associated
with the system model, linearized about the operating
point. The examples in this article have been analyzed I
i
using CEPEL’s PACDYN program, which h a s been 4-
designed for the analysis of large power system models.
It gives all eigenvalues and eigenvectors for systems hav-
+
ing up to 800 equations, and special subsets of eigenval-
ues and eigenvectors for larger systems. Participation
factors, and the residues associated with specific system
inputs and outputs may also be calculated. Step respons-
es for better visualization and frequency responses for
use in control design are also available. A feature of the
program is that it automatically calculates those modes 4-
that dominate the dynamics associated with a particular
input/output combination. +
Understandinlg Oscillations
Eigenvalue analysis gives us the necessary information
to: Figure 2. Eigenvalues of test case (x represents unsatis-
Determine whether a particular operating point is factory damping, + represents satisfactory damping,
sufficiently stable and -represents boundary line satisfactory dampin$

32 IEEE Computer Applications in Power

system, generators 30 to 38 have fast stat- While mode shape relates to what can
ic excitation s,ystems, but are not fitted be seen in a transient simulation, it is
with power system stabilizers. sometimes not very helpful in determin-
ing where to place power system damp-
Cause of Unstable or ing controls. The problem is this: a small
Lightly Damped Modes generator placed at one end of a large sys-
Determining the cause of unstable oscilla- tem may oscillate with higher amplitude
tions starts by examination of the associ- than a large generator placed in the same
ated mode shape. The mode shape tells area but closer to the tie with other areas.
us whether the oscillation is a local mode However, the oscillation energy associat-
or an interareis mode. Local modes have ed with the large machine may be much
eigenvectors that are dominated by only higher than that associated with the small
a few variables associated with a single machine and, hence, controls on the large
generator or a closely connected group of generators. machine are likely to be able to damp the oscillation
Interarea modes have eigenvector components distrib- more effectively than controls on the small machine.
uted more equally throughout areas of tlhe system. Since Participation vectors are uscd to give an indication of
the eigenvector gives the relative amplitude distribution the relative importance of any generator to the control of
of an oscillation throughout the system, it can be used to an electromechanical oscillation. For example, the rotor
recognize the mode in transient stability simulation speed components of participation vectors indicate the
results. Table 1 shows the generator speed eigenvector
components magnitude and phase for the two eigenval- grm 30 against gen 35 & gen 36
ues 0.1022 f j’7.215 and 0.037 i j4.301, respectively. The
relative amplitudes of oscillation defined by the eigen-
vectors are shown in Figures 3 and 4. The first mode is,
essentially, an oscillation between generator 30 and gen-
I 2i
erators 35 and 36. The second mode is a lower frequency
(0.68 Hz) interarea mode. In this mode generators 30 to
38 oscillate algainst generator 39. Generator 39 repre-
sents the equivalent dynamics of a large part of the origi-
nal power system.

Table 1. Generator
speed eigenvector components
Machine Number
Mode 1 -
-
30
-
~
- Magnitude

10
-
- -
-
Phase
-

00
-
(degrees)
-
- L machine number time s

Figure 3. Shape mode I generator speeds, generator 30

31 0 1408 -44 5 speed = exp (1.022t,)
sin (72149t)
32 0 0797 241 9
33 0 1851 152 3 gens 30-38against gen 39
34 0 4777 -32 1
35
36
37
0 7935
0 7797
0 3468
1702
1705
10 1
I *I
38 0 1664 1114
39 - -
00170 - -
191 3- -
ModeP
- - - - - - -
30 5574 -9 9
31 0 4757 -3 4
32 0 5208 -55
33 0 7601 53
34 10 00
35 0 7961 -5 7
36 0 7977 -68
37 0 5084 -124 machine number time s
38 0 6694 -3 3
39 0 4052 -1 79 6 Figure 4. Shape made 9 generator speeds, generator 34
speed = exp (0.036Qt)sin (4.3008t)

July 1996 33
sensitivity of the eigenvalue to changes in mechanical oscillations. If modes a r e unstable, o r t o o lightly
damping applied1 to the generator shaft. If a speed com- damped, these measures used together clearly indicate
ponent of participation is zero, then power system stabi- where controls may be effectively placed t o increase
lizers placed on that generator will not affect the modal damping.
damping. A speed participation factor that is high and The magnitude of participation factors for speed for
positive shows that the generator is a good candidate the two modes are shown in the Figures 5 and 6. For the
site for a power system stabilizer to add damping to the higher frequency mode, the maximum participation
mode. occurs for generator 30, but the sum of the participation
Residues give more specific information. They indi- at generators 35 and 36 is almost equal to that of genera-
cate whether the mode can be controlled with a signal tor 30. For the lower frequency mode, the maximum par-
measured at a slpecific point, and fed back to a specific ticipation occurs for generator 39. The damping effect on
control input. If, for a specified input/output pair, the mode 1 of control at generator 30 is about the same as
residue of a mode is zero, then feedback will not have that of controls placed at both generator 35 and genera-
any effect on that mode. Using eigenvalues, eigenvectors, tor 36. For the second mode, controls at generator 39
participation velctors and residues gives the analyst a will have most effect. However, this is an artificial equiva-
good appreciation of the nature of the power systems lent generator. The sum of the participation of the other
generators is almost equal to that of generator 39. This
means that controls placed on all generators except 39
speed participation factors Mode 1
will provide about the same damping as a single control
placed on generator 39. All the electromechanical modes
of this example system are unsatisfactorily damped, and
detailed study shows that controls on generators 30 to
38 are necessary to completely stabilize modes 1 to 8.
These same controls stabilize the low frequency inter-
area mode without requiring an additional stabilizer on
generator 39.

Damping Controls
Power system stabilizers are the most cost-effective
power system oscillation damping controls. Essentially,
they use the power amplification capability of the gener-
ators t o generate a damping torque in phase with the
speed change. This is achieved by injecting a stabilizing
I machine number
signal into the excitation system voltage reference sum-
ming junction. The stabilizing signal is most often the
Figure 5. Speed participation mode 1 change in generator rotor speed, phase advanced t o
counteract the phase lag between the exciter voltage ref-
erence and the generator electrical torque. There have
I ' - - - - - ' ' ' ~ ~ '
speed participation factors - Mode 9
been problems in the past with power system stabilizers
causing steam turbine shaft torsional modes to become
unstable, but this risk has been eliminated in many mod-
ern power system stabilizer designs. However, there are
still many problems with installed power system stabiliz-
ers that have been introduced by ineffective commis-
sioning and tuning of the devices. Generally, in systems
with both local and interarea modes, power system sta-
bilizer parameters are best determined through off-line
analysis, and then set to these values on commissioning.
?2 0 3 The validity of the model used in the off-line studies
02 should be checked on commissioning. Setting power sys-
tem stabilizers to typical values is particularly dangerous
01 for systems in which interarea modes are of concern: it
0 is very easy for the stabilizer t o have a destabilizing
30 31 32 33 34 35 36 37 38 39
effect at low frequencies that cannot be observed during
i machine number
on-line commissioning tests.
Figure 6. Speed participation mode 9 The system's eigenvalues with power system stabiliz-

34 IEEE ComputerApplications in Power

ers at generators 30 to 38 are shown in Figure 7. With In general, generator models supplied by manufactur-
these stabilizers, all the system’s modes, including the ers in the last 20 years are satisfactory. Exciter models
interarea mode, decay sufficiently quickly. are less easily obtained. However, dynamic models can
Other power system devices, such as HVDC links, be obtained from t!he study of the manufacturers sup-
static var compensators, and FACTS, can be used t o plied circuit diagrams and a knowledge of the control’s
enhance the stability of power system oscillations. potentiometer sett:ings. Models of exciters supplied
However, the cost of these devices requires that their more than a few years ago were often drastically simpli-
use is justified by something other than added power fied to fit the capabilities of the stability programs which
system damping. This means that they may not be sited were then used for analysis, and should be treated with
at a point in the system best suited to oscillation d a m p suspicion. Today’s analysis programs a r e versatile
ing. The probllem is to find their effectiveness in a par- enough to be able t o model any significant feature of
ticular case. Again, eigenvalue analysis (can give a good power system dynamics.
indication of the feasibility of their use for damping It is easy for controller settings to be inadvertently
control. changed. It is not until problems occur that the settings
Damping control design using these special devices is are checked. Planner’s models based on original manu-
often best achieved using more advanced methods than facturers informatioln may not be accurate 5 years later.
are necessary for power system stabilizers. Power sys- Data for the controls of critical system devices should be
tem stabilizers, if designed correctly, are generally quite tested periodically. Indeed, with this in mind, some mod-
robust to changes in system parameters and operating ern exciters have in-built monitoring facilities.
conditions. This is not always true of damping con- In large systems, such as the U.S./Canada intercon-
trollers that use these special devices. Thus a robust nected systems, a full model of the system is beyond the
controller design method, such as H,, is often required. capability of any system simulation program; the largest
This method requires knowledge of the systems transfer system model that can be currently studied has 50,000
function and its variation over the expected range of buses. Thus, it is necessary to reduce the system models
operating conditions. This information is easily pro- so that they fit the program’s capacity. Reduced order
duced by modal analysis programs. system models are not straightforward to produce. It is
certainly unacceptable, for oscillation studies, t o pro-
System Models duce a reduced maldel by reduction of the steady-state
Power system stability programs are as accurate as the network. This technique distorts the interarea modes of
system models that they use. The correct prediction of oscillation and generally gives very pessimistic stability
the stability of power system oscillations is very sensi- models. The reduction should be done in such a way
tive to the models of the generators and their controls. that the lowest frequency modes of oscillation are p r e
Excitation system model accuracy is crucial to obtain a served in the reduction. Dynamic power system reduc-
good small signal stability indication. tion based on thi:: principal h a s been shown t o b e
accurate and forms t h e basis of t h e EPRI program
DYNRED.
I least damped eiaenvatues - with pss I
Acknowledgment
Visualization of the cornputer results was obtained using MATLAB
from The Mathworks Inc. PACDYN may be obtained from Cherry
Tree Scientific Software:. DYNRED may be obtained through EPRI.

Further Reading
System Oscillations Working Group, Inter-Area Oscillations in Power
Systems, IEEE Power Engineering Society, 95TP101, October 1994.
Prabha Kundur, Power System Stability and Control, McGraw-Hill,
1993.

Biography
Graham Rogers has hald a varied career in electric power systems
engineering spanning over 35 years. He has worked as an educator
and researcher a t Southampton University, as an engineering
mathematician at AEI(Rugby)Ltd., and as a power system analyst
at Ontario Hydro. He formed Cherry Tree Scientific Software in
1993 and through them develops and markets computer programs
real palt for power system analysis, gives courses and seminars, and p r e
vides engineering support services. He is Associate Professor
Fpigure 7. Eigenvalues of least damped modes with (part-time) at McMaster University, Adjunct Associate Professor at
power system stabilizers at generators 30 to 38 f- r e p University of Toronto and Associate Editor of the IEEE Control Sys-
resents boundary line satisfactory dampin& tems Society Transactions on Control Systems Technology.

July 1996 35