You are on page 1of 13

IEEE Transactions on Power Apparatus and Systems, Vol. PAS-100, No.

6 June 1981
APPLYING POWER SYSTEM STABILIZERS
PART III: PRACTICAL CONSIDERATIONS

3034

D.A. Swann (Member)


E.V. Larsen (Member)
General Electric Company, Schenectady, New York

STABILIZER IMPLEMENTATION PROCEDURES

ABSTRACT
The practical considerations associated with applying
system stabilizers are addressed in this final
part of the paper. Procedures are described whereby
the tuning concepts developed in Part II may be
implemented in the field. An approach is described
for determining the "plant" characteristics for which
Guidelines are prea stabilizer must compensate.
sented for adjustment of stabilizer parameters,
inclduing frequency- response, gain, and output limits.
Techniques are described for verification of
proper stabilizer set-up.

As used in this paper, implementation of a power


system stabilizer implies adjustment of its frequency
characteristic and gain to produce the desired damping
of system oscillations in the frequency range of 0.2 to
2.5 Hz. The transfer function of a generic power system
stabilizer may be expressed as per equation (1):

In addition to tuning, this part also deals with the


potential for torsional interaction with the various
types of stabilizers, viz. speed, frequency, and
power/speed input signals, and the filtering required
to alleviate diffi-culties. The effects of electrical
noise from the power system on stabilizer performance
is described, as is the susceptability of specific
types of stabilizers and transducers to noise problems. For power input stabilizers, an approach for
representing mechanical power variations and the
consquences of an imprecise representation are discussed.

where

power

INTRODUCTION
This portion of a three-part paper deals with
practical aspects of applying power system stabilizers.
Guidelines are presented for tuning the three types of
stabilizers described in Part I, viz., utilizing speed,
frequency, or power input, which enable the user to
achieve satisfactory stabilizer performance with limited
effort. These guidelines are based on field experience
[1,2,3,], including that of others reported in the
literature [4,5,6,7], and extensive analytical studies
leading to- the concepts described in Parts I and II of
this paper.

Realization of the system performance benefits


available by applying power system stabilizers requires
attention to such practical considerations as influence
of the stabilizer on turbine-generator shaft torsional
modes of vibration, the effects of power system noise
and,
for power input stabilizers, mechanical power
variations. These concerns relate primarily to equipment design; the approaches taken to alleviate these

potential difficulties

80

SM

560-3

paper

are discussed.

recommended

and

approved by

IEEE Power Generatf'on Committee of the IEEE

the

Power

Engineering Society for presentation at the IEEE PES


Summer Meeting, Minneapolis, Minnesota, July 13-18,

1980. Manuscript submitted March 14, 1980; made


available for printing May 7, 1980.

PSS

KS
KS

(1 +sT )(I
sT2)(1
(1
+

+
+

sT3)
sT4) FILT(s)

(1)

stabilizer gain

FILT(s)= combined transfer function of


sional filter
transducer

and

input

tor-

signal

The stabilizer frequency characteristic is adjusted by


It
T
T , and T
varying the time constants T , T
will be noted that the sta%il2zer transfer function
includes the effect of both the input signal transducer
and any filtering required to attenuate the stabilizer
gain at turbine-generator shaft torsional frequencies.
These effects, dictated by other considerations, must be
considered in addition to the "plant" (as defined in
Part I) for which the stabilizer must compensate.
A power system stabilizer can be most effectively
applied if it is tuned with an understanding of the
associated power system characteristics and the function
to be performed by the stabilizer, as described in Parts
I and II of this paper. A knowledge of the modes of
power system oscillation to which the stabilizer is to
provide damping establishes the range of frequencies
Simple analytover which the stabilizer must operate.
ical models, such as that of a single machine connected
to an infinite bus, can be useful in determining the
frequencies of local mode oscillations during the planning stage of a new plant. It is also desirable to
establish the weakest power system conditions and associated loading for which stable operation is expected as
the adequacy of the power system stabilizer application
will be determined under these performance conditions.
Since the limiting gain of some stabilizers, viz., those
having input signals from speed or power, occurs with a
strong transmission system, it is necessary to establish
the strongest credible system as the "tuning condition"
for these stabilizers. Experience suggests that designing a stabilizer for satisfactory operation with an
external system reactance ranging from 20% to 80% on the
unit rating will ensure robust performance.

Many power system stabilizers have been tuned with


the aid of extensive analytical studies involving detailed models of the power system including the generators, exciters, stabilizers, and even the turbinegenerator torsional dynamics. While such studies can be
very valuable in attaining the best possible stabilizer
performance, satisfactory performance can usually be
attained with less effort by following appropriate
guidelines. Such guidelines are suggested by the analytical results presented in Parts I and II of this
paper, and by experience in implementing many of these
concepts in the field. These guidelines are described
in the paragraphs which follow.

Performance Criteria

The stabilizer tuning yielding the best overall


performance has been found to provide a local modt
damping constant in the range of -1 to -2 seconds
(i.e., oscillation decay time constants between 1 second
and 0.5 second) for moderate to weak ac systems. Less
damping results in quite oscillatory performance, but a
stabilizer providing damping in excess of this tends to
give poor performance for large system transients as it
detracts too much from the voltage regulation function
of the excitation system. To ensure damping over the
broadest possible frequency range, and thereby minimize
the effect of system changes on stabilizer performance,
the stabilizer time constants T , TV T,, T3 and T
should be set to compensate for tWe phase lag associated
with the plant such that the net phase lag is:
0

1.

Less than 90
possible.

2.

Between 00 and 450 from the lowest intertie mode


frequency to the highest local mode frequency.

for as broad a frequency range as

System Condition
The stabilizer should be adjusted to give the above
characteristic for the condition which represents the
highest stabilizer loop gain and greatest phase lag.
For all three types of stabilizers (i.e., speed input,
frequency input, and power input), full load on the
generator yields the highest loop gain. For speed and
power input stabilizers, the strongest ac system presents the highest loop gain and greatest phase lag. For
frequency input stabilizers, the highest loop gain
occurs with the weakest ac transmission system, but as
noted in Figure 3 of Part I, increases only slightly for
external reactance greater than about 40 percent.
Plant Identification

The desired time constants of the stabilizer depend


upon the phase characteristics of the open-loop transfer
function of the plant, i.e., from stabilizer output to
torque with constant rotor speed:

GEP(s)
where

= AT

ep/Epss

GEP(s)
T

ep

pss

the rotor speed


making measurements in
transfer function from
electrical torque will
noted in Part I, GEP(s)

(2)
the plant through which
stabilizer must operate.

the

component of electrical torque


solely to stabilizer path
(i.e., with constant rotor
speed).

due

stabilizer output signal.

cannot be held constant while


the field, measurement of the
the stabilizer output signal to
not yield GEP(s). However, as
is proportional to the closedloop voltage regulator characteristic with constant
rotor speed, and as shown in Appendix A, this transfer
function is affected relatively little by shaft motion.
The transfer function from voltage reference to terminal
voltage therefore provides the best measure of the
The data is generally
required phase information.
accurate for frequencies beyond the local mode resonance, which is the most critical frequency range when
tuning the stabilizer to keep the net phase lag less
than 900 to as high a frequency as possible. Interpolation from data points on either side of the resonance
will yield a fairly good estimate of the phase of GEP(s)
at the local mode frequency.

Since

3035
This measurement is often obtained by introducing
sinusoidal signals into the voltage summing junction of
the regulator and utilizing a transfer function analyzer
to determine the gain and phase of the resulting terminal voltage variations with respect to the driving
signal. Recently, techniques have been developed to
utilize noise signals which are processed by digital
computer based instrumentation to identify this transfer
function [8]. Such techniques hold promise for simplifying field measurements of this type in the future.

Setting Time Constants


The stabilizer time constants are set to provide
the desired phase compensation. It is desirable to have
a slightly undercompensated system, as a phase lag of
between 200 to 400 at the system oscillation frequencies
will result in a component of electrical torque due to
the stabilizer which is largely damping with some positive synchronizing. Note that perfect phase compensation implies zero phase lag for a speed or frequency
input stabilizer, and 900 of lag for a power input
stabilizer (as described in Part I). The stabilizer
lead/lag stages must also compensate for the phase lag
introduced by the transducer and torsional filtering.
The frequency range over which the phase compensation is required depends on the stabilizer input signal
employed. All stabilizers should be adjusted to produce
some phase lag at the lowest frequency interarea mode of
system oscillation, typically in the neighborhood of 0.2
to 0.5 Hz, since phase lead at these frequencies will
cause the stabilizer to produce a desynchronizing component of torque. This may reduce interarea power transfer capability by having an adverse impact upon transient stability. For power and speed input stabilizers,
which are equally sensitive to local modes and modes
between units in a plant, the stabilizer compensation
should extend to as high a frequency as possible, since
destabilization of the higher frequency intra-plant
modes will limit the stabilizer performance. The results presented in Part II indicate that stabilizer
adAustment that produces a net phase lag of less than
90 up to 3 to 3.5 Hz will give good results. Limiting
the phase lag at these higher frequencies is more important than obtaining perfect phase compensation at lower
frequencies with speed and power input stabilizers.
With a frequency input stabilizer, which is insensitive
to the local modes of oscillation within the plant, it
is not as important to compensate at high frequencies;
the phase lag can normally be permitted to go through
900 in the frequency range of 1.5 to 2 Hz. This allows
the stabilizer to be tuned with smaller ratios of
lead/lag time constants and consequently less high frequency gain.
The adjustment of the stabilizer time constants to
produce proper compensation at a local mode frequency
can be verified by a simple field test. The test involves the determination of points on a "root locus" by
field measurement from which the proper action of the
stabilizer may be deduced. The initial migration of the
local mode eigenvalue, in terms of change in frequency
and damping, can be determined by stimulating the local
mode oscillation with zero stabilizer gain and again
with a few low values of gain which cause a noticeable
change in damping. The oscillations can be stimulated
by step changes to the voltage regulator reference, line
switching, or low level sinusoidal stimulation of the
local mode by modulation of the voltage reference. A
recording of the resulting oscillation will yield frequency and damping information; frequency by timing the
length of a number of cycles, damping by sketching the
decay envelope and plotting its magnitude versus time or
cycles on a semi-log graph [1]. Modern techniques
utilizing random noise signals with microcomputer based
instrumentation are currently emerging, which will

3036
likely

prove to be a
tests in the future.

effective alternative for such


These tests can be designed to be
safe, and are easily performed and interpreted. The
initial direction of eigenvalue motion should be to
produce increased damping and a slightly higher freAs indicated in Part I, the
quency of oscillation.
direction of this eigenvalue migration in the s-plane is
directly related to the compensated phase angle of the
stabilizer loop. Thus, results of this test can be used
to verify proper phase compensation at the local mode
frequency.
very

Setting Gain

The desired stabilizer gain may be set based on the


stabilizer gain which produces instability. The latter
is obtained by a field test commonly referred to as the
This test consists of
gain margin test [1,2,4,6].
slowly increasing the stabilizer gain until an instability is observed. This instability is characterized
by growing oscillations at a frequency greater than the
local mode of oscillation at the plant, typically in the
Stabilizer output and exciter field
range of 2 to 4 Hz.
voltage are the most sensitive signals from which to
monitor the onset of instability. Once an instability
is detected, the stabilizer is switched out of service.
Reduction of the stabilizer output limits will minimize
the potential disturbance level during this test, and
contribute to a safe procedure. Figure 1 shows an
example of exciter and speed signals as instability is
reached in a gain margin test. It is seen that each
time the gain is increased, particularly as the instability is approached, some oscillations are stimulated
which initially grow, but then die away. These can be
mistaken as the instability and result in a measure of
instability gain lower than actual. This would, however, be conservative.
During the gain setting tests, torsional oscillations in the stabilizer output signal should be monitored to ensure that the torsional filtering is performing properly.
As described in Part II, there is a relationship
between the instability gain and the desired stabilizer
gain. This relationship is dependent upon the high
frequency filter characteristics and stabilizer input
signal. With the high frequency filtering assumed in
Part II, the desired gain is about one-third of the
instability gain for speed input stabilizers, and about
two-thirds of the instability gain for frequency input.
This characteristic will generally be valid for most
practical equipment utilizing speed or frequency input.
However, the lower torsional filtering requirements
associated with power input stabilizers allows a much
higher gain before onset of an instability. The results
of Part II showed an 8:1 ratio between instability and
desired gain, but this will vary with the specific
application.
The instability gain must be determined at the
operating conditions causing the highest stabilizer loop
As indicated previously, with the frequency input
gain.
stabilizer, highest loop gain occurs with a weak system
condition, under which it may be undesirable to perform
gain margin tests. However, the relative loop gain can
be calculated as a function of system strength, similar
to Figure 3 of Part I, with a simple model [9]. The
system can then be tested under a fairly strong condition with the observed instability gain used in conjunction with this calculation to extrapolate to the instability gain which would occur with the weak system.
This same approach can be followed should testing at
full load prove difficult; tests can be performed at
lighter loads with calculations used to extrapolate to
full load.

An additional point should be made here regarding


the gain margin of the frequency input stabilizer. The
results presented in Part II suggest that optimum performance occurs at two-thirds of the instability gain
under the tuning condition. Hence, only a 4 db (1.5:1)
gain margin exists, as opposed to the 10 db (3:1) margin
for optimum gain with a speed input, and which is typical of current industry practice. Conservative control
system design practice of the system operators may
therefore impose a restriction limiting performance with
a
frequency input stabilizer to less than optimum.
However, optimum damping contribution with frequency
input is nearly twice that with speed input for weak
systems, so using a 10 db gain margin will result in
performance at least as good as speed input stabilizers
for the performance condition of weak systems.

SPE ED

.05%

rIp l%V M

FIGURE 1. GAIN MARGIN TEST

Setting Output Limits

The output of the stabilizer must be limited to


prevent damping signals from saturating the excitation
system and thereby defeating the voltage regulation
function. This is of particular concern with excitation
systems which have more dynamic range in the down direction than the up direction. In response to a fault on
the power system, the exciter will initially go to
ceiling, but on the rotor backswing the stabilizer will
call for and receive from the excitation system more
forcing in the down direction than was available in the
up direction. The result is a lowering of the average
terminal voltage following the first cycle of the system
oscillation. This lower voltage will decrease power
transfer limits between areas.
This effect has been experienced in studies of
large systems. In one particular case the first swing
between two areas following a fault was stable, but
local oscillations in the receiving area were very
lightly damped. Addition of power system stabilizers to
damp the local oscillations caused a reduction in the
average voltage of the receiving system and the first
swing between the areas then became unstable.
As a general guide, when the stabilizer output is
driven to its limits at frequencies between 0.5 and 2
Hz, the excitation system response should be within its
limits. This criteria should be met with full load on
the unit and with a moderate to weak transmission system
(X > 0.5 p.u.).
e
Summary of Field Implementation Procedure
Power system stabilizer tuning begins with an
understanding of the application. Field tests are
performed to obtain the phase lag characterization of
the "plant" (i.e., excitation system, generator, and
power system) through which the stabilizer must operate.
The best stabilizer damping characteristics over a broad
range of system conditions are obtained by adjusting the

stabilizer time constants such that the net phase lag of


the plant and stabilizer (including the stabilizer torsional filter and transducer) is between 00 and 450 over
the frequency range from approximately 0.2 Hz to 1.5 Hz.
The phase lag should be kept less than 90 up to approximately 2 Hz for frequency input stabilizers and 3.5 Hz
for speed and power input stabilizers. The stabilizer
thereby produces a positive synchronizing component of
torque which reduces the tendency towards an adverse
influence upon large transient performance.
The stabilizer gain for speed and frequency input
stabilizers is established as a percentage of the instability gain. The latter is determined by use of the
gain margin test in which the stabilizer gain is increased until instability is observed. This test is
ideally performed at full load and with strong transmission conditions when using a speed or power input stabilizer; at full load and with moderate to weak transmission conditions when using a frequency input stabilizer.
Calculations with simple models can be used to extrapolate from measurements at other test points to the
lowest possible instability gain should testing under
these ideal tuning conditions prove inconvenient. With
a properly adjusted stabilizer, the local mode decay
rate fhould typically be in the range of -1 to -2 seconds
under moderate to weak system conditions, which
should be checked by a final test to verify performance.

TORSIONAL INTERACTION
The potential for interaction between'power system
stabilizers and turbine-generator shaft torsional modes
of vibration was observed by Ontario Hydro at their
Lambton station [10], and subsequently by the authors'
Company at Salt River Project's Navajo station [3].
Analysis has revealed that such interaction can occur on
nearly all modern excitation systems as they have relatively high gain at high frequencies. A stabilizertorsional instability with a high response ratio excitation system may result in shaft damage, particularly
at light generator loads where the inherent mechanical
damping is small. Even if shaft damage does not occur,
such an instability can cause saturation of the stabilizer output causing it to be ineffective, and possibly
also cause saturation of the voltage regulator resulting
in loss of synchronism and tripping the unit. It is
imperative that stabilizers do not induce torsional
instabilities.

3037
shift characteristics at low frequencies were
close to the equivalent used in Part II for the torsional filtering, i.e.,

phase

FILT(s) _ 570/(570 + 35s +

(3)

for frequencies below 3 Hz

Hence, attainment of the performance shown by the analytical results in Part II is ensured. The characteristic of equation 3 is also a good approximation of the
filter for two-pole units described in [3].
Alternative Input Signals
The research leading to the torsional filter design
for four-pole units involved consideration of speed,
frequency, and a combination of power and speed as
stabilizer input signals. The relative torsional interaction characteristics associated with each of these
input signals is worth noting here.
Speed input stabilizers generally require the
greatest amount of torsional attenuation. Although the
filtering requirements could be minimized by positioning
a speed sensor at an appropriate location on the turbine-generator shaft, for ease of installation it is
desirable to utilize a speed pickup mounted at either
end of the shaft, usually the front standard. The
torsional mode-shapes of typical turbine-generators
indicate that there is less torsional motion at the
generator than at the ends of the shaft. As a consequence, the frequency input signal, which is comparable
to a speed pickup on the generator rotor, typically has
a lower torsional frequency content than a speed signal
taken from the front standard or exciter. On some untts
this can amount to a 20 db difference in torsional
attenuation requirements. In addition, as noted in Part
II, the tuning of a frequency input stabilizer need not
accommodate the higher frequency local modes, including
intraplant modes of oscillation, which results in approximately 8 db less torsional interaction than a speed
input stabilizer having a pickup mounted on the generator shaft.

The non-minimum phase characteristic which has been


shown to characterize stabilizers using accelerating
power as an input results in much lower torsional interaction than either the speed or frequency input stabilizers, which have minimum phase characteristics. This
To ensure that a stabilizer does not induce tor- characteristic results in a stabilizer interaction with
sional oscillations, torsional filtering schemes have torsional oscillations having a declining gain with
been developed. Reference 3 describes a torsional band frequency at the rate of 20 db/decade for the ideal
reject filter designed for two-pole turbine-generator power input stabilizer without filtering. By contrast,
units utilizing front standard speed input, and similar the torsional interaction gain increases with frequency
filters have been designed for four-pole units utilizing at the rate of 20 db/decade for the ideal minimum phase
speed and frequency input stabilizers without filtering.
frequency input.
Assuming equivalent performance at a local mode freTwo basic criteria have been established for de- quency of 1 Hz, the torsional interaction for the power
input stabilizer would be lower than for the speed or
signing such a filter:
1.
Ensure that the maximum possible change in damping frequency input stabilizer by 40 db at 10 Hz and by 52
of any torsional mode is less than some fraction of db at 20 Hz. Even with this advantage, however, utilizing a power input stabilizer with a high performance
the inherent torsional damping.
GENERREX* excitation system does not eliminate the need
2.
Minimize the phase lag of the filter in the fre- for torsional filtering on four-pole turbine-generator
quency range from 1 to 3 Hz to ensure the attain- units, due to the relatively low first torsional mode
frequencies in the neighborhood of 6 to 8 Hz.
ment of adequate system performance.
In meeting the first criteria, the stabilizer
settings arrived at in Part II of this paper were utilized to establish the torsional interaction for a wide
range of system operating conditions. These were compared to estimated inherent damping for those operating
conditions and the filter attenuation requirements were
established to provide a margin of safety in excess of
20 db.
This attenuation was achieved for four-pole
units utilizing' frequency input such that the filter

An additional aspect of torsional interaction with


power input stabilizers must be considered when using
the method proposed by deMello [11] to compensate for
variations in mechanical power. This method involves
obtaining a measure of shaft speed and differentiating
it to determine the true accelerating power, then subtracting electrical power to determine the mechanical
contribution. Direct use of this technique is equiva-

Trademark of General Electric Company.

3038
2.

Fast valve closure, as by relaying following a


fault or for overspeed protection, on a unit will
likely cause saturation of the stabilizer. The
rapid reduction in power following valve closing
will raise the voltage and could cause excessive
voltages on the back swing. When the valves open
again and load begins building up, a voltage reduction will be imposed on the system and, as was the
case with the fast load pickup, could aggravate a
stability problem.

3.

A potential exists for interaction between the


power system stabilizer and governor controls with
fast acting governor systems.

Pa

NOTE:

ACCELERATING POWER Po - M s w Pm Pe
APPROXIMATE MECH. POWER Pm (M s w + Pe) Fm(s)
APPROXIMATE ACCEL. POWER P Pm- Pe
:

Aeref

(A)

(B)
(C)

FIGURE 2 POWER PLUS SPEED FUNCTIONAL IMPLEMENTATION

lent to a minimum phase stabilizer with speed input and


its associated torsional interaction characteristics.
However, since the mechanical power component varies
relatively slowly, the signal generated to simulate
mechanical power can be filtered quite heavily. This
implementation is illustrated in Figure 2, with the
filtering of the mechanical power signal represented as
F (s). With F (s) = 1, this stabilizer reduces to a
speed input stabilizer as mentioned previously. With
F (s) = 0, a power input stabilizer results having the
torsional interaction characteristics described previously. Since, as described in Part I, the torsional
interaction characteristics of the power and speed input
are diverging at a rate of 40 db/decade with equivalent
gain at the local mode frequency, the filtering required
in the simulated mechanical power signal must have a
second order lag characteristic for the power, speed
combination stabilizer to have the same torsional interaction advantages associated with the accelerating power
A double time constant low pass
input stabilizer.
filter with break frequencies between 5 and 10 rad/sec,
i.e., near the local mode frequency, will attentuate the
torsional interaction magnitude to approximately the
Additional
same level as with electrical power alone.
filtering of the simulated mechanical signal alone is
therefore of little value and increases the potential
for difficulty with mechanical power variations..
EFFECT OF MECHANICAL POWER VARIATIONS

Imperfect compensation of mechanical power variations with a power input stabilizer, including transient
variation due to filtering of the simulated mechanical
power signal, will result in some transient voltage
offsets during fast mechanical power changes. Three
situations are of concern:
1.

Rapid

response of the governor on a unit to a


sudden decline in system frequency will cause a
rapid increase in the power output of the unit.
This might occur as the result of the loss of
another unit on the system or the loss of ties to
an adjacent area.
Stabilizer action may then cause
a reduction in voltage which, in turn, may aggravate the system stability problem.

In addition to these transient effects, imperfect


steady-state compensation for mechanical power will
cause steady-state offsets in the stabilizer input
signal, which must be corrected by utilizing an additional washout stage. As indicated in Part I, the
additional phase lead associated with this washout stage
may have a detrimental impact on interarea modes of
oscillation.
These concerns must be resolved when
applying power input stabilizers. Some of them are
strongly influenced by the nonlinear characteristics of
the particular installation, and hence may require
custom design.
NOISE CONSIDERATIONS

Noise content of the power system stabilizer input


signal can cause significant problems in applying equipment in certain locations. This is particularly true
with plants located near large industrial loads such as
arc furnaces. For minimum phase stabilizers which have
a gain which increases with frequency, the noise is
amplified considerably and can cause saturation of the
stabilizer. In addition to "true" noise signals caused
primarily by intermittent loads on the power system,
there may also be "false" noise signals due to aliasing
produced by sampling types of transducers.
Front standard shaft speed from a multi-tooth wheel
is relatively noise free, which is the reason it has
been used in most applications. Use of electrical
power, with its non-minimum phase characteristic and
declining loop gain as a function of frequency, is
likely to be relatively free of noise problems. As
would be expected, ac bus frequency is the most sensitive signal with respect to power system noise. Large
industrial loads can significantly distort the voltage
waveform from which frequency is being sensed, causing
zero-crossing types of transducers to give false indications and potentially prevent the stabilizer from
improving system performance. In addition, these transducers have an aliasing problem. The best that a zero
crossing transducer can do is to sample every half-cycle
per phase, and hence experiences an aliasing effect with
any signal above 60 Hz. The ac voltage signal will in
general always have signals superimposed on it which are
above 60 Hz due to normal system stimulations. This
noise problem has prevented the use of frequency as an
input in several locations in the experience of both the
author's company and others [6,7].
Much of the noise content, particularly in the
speed signal, is due to torsional oscillations. Hence,
applying torsional filtering to minimize the torsional
interaction with the stabilizer also helps considerably
with respect to reducing the noise. This has been
observed with all of our applications of the torsional
band reject filter [3], and also by others [6,7].

3039

NOTE:

f (Hz)

10

20 30 50

(C) TERMINAL BUS FREQUENCY,


SINGLE PHASE ZERO-CROSSING
DETECTOR

db REPRESENTS 1.0 P.U. RMS SHAFT OSCILLATION MAGNITUDE


FIGURE 3 SPECTRUM MEASUREMENTS AT WORST NOISE

In anticipation of using ac frequency as a stabilizer input, a new frequency transducer has been developed and tested at several locations, including that
where the author's Company has experienced the most
severe noise. Figure 3 shows frequency spectrums* as
measured on signals taken from (a) shaft speed, (b) the
new transducer, and (c) a zero crossing detector frequency transducer with and without 60 Hz bandpass filtering on the ac voltage (intended to minimize aliasing). As expected, the frequency signal is less sensitive to local and torsional modes of oscillation than
the speed signal, but has more noise in the frequency
range from 10 to 50 Hz. The new frequency transducer
has considerably less noise than the zero crossing type,
even when the latter is augmented by a filter.

To examine the impact of this noise on power system


stabilizer performance with the new frequency transducer, the time response of the stabilizer with the
adjustments determined in Part II and torsional filtering discussed in the previous section were calculated
from the frequency spectrum. Figure 4 shows that without the torsional filter, the stabilizer output consists
largely of noise. With the torsional filter, the noise
content is reduced dramatically leaving only the local
mode of oscillation. Since these tests were taken at
the location which was found to have the most severe
noise characteristics, confidence is established that
the frequency input stabilizer using the new transducer
and torsional filtering will not suffer from limitations
due to noise.

1.0

WITH AC FILTERS

2 3

f (Hz)
(B) TERMINAL BUS FREQUENCY,
NEW TRANSDUCER

f (Hz)
(A) SHAFT SPEED

NO AC FILTER

NO
FILTER

WITH
KFILTERj

I0.0
T IME -

SEC

FIGURE 4 CALCULATED EFFECT OF TORSIONAL FILTER ON NOISE


MEASURED AT WORST LOCATION FOR FREQUENCY INPUT PSS

LOCATION

SUMMARY AND CONCLUSIONS

The major practical considerations associated with


applying power system stabilizers have been discussed.
Basic guidelines for tuning stabilizer equipment in the
field have been described, based upon field experience
and the understanding gained from analyzing stabilizer
performance over a wide range of system conditions. In
particular, the relationship established in Part II
between system performance and the phase compensation
The procedure
characteristics have been utilized.
consists of two basic steps:
1.

Setting the stabilizer time constants to yield a


phase compensation characteristic specified for
best performance.

a)

Maximize frequency at which the compensated


phase lag passes through 90 . (3 to 3.5 Hz
for speed or power input, 2 Hz for frequency

input.)

b)
2.

Compensated phase lag at local mode frequency


between 00 and 450, preferably near 200.

Setting the gain based upon a measurement of the


gain which causes an instability in the stabilizer
loop.

Several testing techniques have been discussed,


although a particular stabilizer application may require
only a few of them. Utilization of digital computerbased instrumentation currently being introduced holds
the potential for making stabilizer commissioning a
relatively straightforward process in the future, with
minimal need for system studies for each application.
The major practical considerations associated with
equipment design for successful stabilizer application
have been discussed, viz., minimization of torsional
interaction, the effects of power system noise, and, for
power input stabilizers, mechanical power variations.
Torsional filtering schemes have been developed by the
author's Company which ensure a sufficient margin of
safety for torsional interaction, while at the same time
having minimum phase shift at low frequencies so that
good stabilizer performance may be achieved. Although
the power input stabilizer requires less torsional
filtering than a stabilizer utilizing speed or frequency
as an input, such stabilizers require additional cir-

3040
cuitry and input signals to compensate for the effects
It is difficult to
of mechanical power variations.
compensate perfectly for mechanical power variations,
and imperfect compensation introduces the potential for
other system interactions. These must be resolved to
ensure successful application of a power input stabilizer, and may require custom designs tailored to specific installations.
Noise on the power system can have an adverse
effect on stabilizer performance. Some stabilizer types
are more susceptible to noise problems than others.
Historically the frequency input stabilizer using zero
crossing detection to measure frequency has performed
poorly in the presence of noise caused by large intermittent loads such as arc furnaces. However, a new
frequency t-ransducer has been developed which promises
to significantly reduce the impact of this noise on the
measured frequency signal.
Further, the filtering
required to minimize torsional interaction also significantly reduces the effect of power system noise.

ACKNOWLEDGEMENTS

The analysis and field test work which forms the


basis for the results presented in this paper involved
contributions from several individuals, notably G.F.
Wright of Large Steam Turbine-Generator Department, R.A.
Lawson of Drive Systems Department, and M.L. Crenshaw,
J.M. Cutler, D.H. Baker, and D.D. Durbak of Electric
Utility Systems Engineering Department. In addition,
excellent cooperation has been received from several
utility companies, notably Consumers Power Company,
Commonwealth Edison Company, The Montana Power Company,
and Salt River Project in permitting tests to be conducted on their systems.
REFERENCES

1)

P.H. Beagles, E.V. Larsen, "Field Test and Analytical Experience with Dynamic Stability of a Loosely
Interconnected Power System", Paper A 78 519-1
presented at IEEE PES Summer Meeting, Los Angeles,
July 1978.

2)

S.T. Naumann, G.L. Landgren, R. Jovanovich, E.V.


Larsen, D.A. Swann, "Underexcited Operation and
Stability Tests at the Powerton Station", Proceedings of the American Power Conference, 1979.

3)

R.A. Lawson, D.A. Swann, G.F. Wright, "Minimization


of Power System Stabilizer Torsional Interaction",
IEEE Trans., Vol. PAS-97, Jan./ Feb. 1978, pp.

8)

G.S. Hope, S.T. Nichols, "Measurement of Transfer


Functions of Power System Components Under Operating Conditions", IEEE Trans., Vol. PAS-96, Nov.!
Dec. 1977, pp. 1798-1808.

9)

C. Concordia, F.P. deMello, "Concepts of Synchronous Machine Stability as Affected by Excitation


Control," IEEE Trans. Vol. PAS-88, April 1969 pp.

316-329.

10)

W. Watson, M.E. Coultes, "Static Exciter Stabilizing Signals on Large Generators - Mechanical Problems", IEEE Trans., Vol. PAS-92, Jan./Feb. 1973,
pp. 205-212.

11)

F.P. deMello, L.N. Hannett, J.M. Undrill, "Practical Approaches to Supplementary Stabilizing from
Accelerating Power", IEEE Trans., Vol. PAS-97,
Sept./Oct. 1978, pp. 1515-1522.
APPENDIX A - MEASUREMENT OF GEP(s)

For tuning the power system stabilizer, it is


desired to measure the transfer function GEP(s) from the
stabilizer output signal (usually the same as the voltage setpoint) to the component of torque which can be
controlled via excitation modulation, i.e., the torque
change which would result from exciter modulation with
rotor speed constant. In practice, it is impossible to
hold the rotor speed constant and thus measuring the
transfer function from the stabilizer output signal to
electrical torque is not equivalent to measuring the
desired transfer function, although it is related to
that transfer function. As indicated in Part I, GEP(s)
is proportional to the transfer function from voltage
reference to terminal voltage for -constant rotor speed.
Rotor motion has less influence on the latter transfer
function than on the transfer function from reference
signal to torque, and hence the voltage measurement
provides the best measure of GEP(s).

ATm

183-190.

4)

Kundur, D.C. Lee, H.M. Zein El-Din, "Power


System Stabilizers for Thermal Units: Analytical
Techniques and On-Site Validation", Paper F80-227-9
presented at IEEE PES Winter Meeting, New York,
Feb. 1980.

5)

J.P. Bayne, D.C. Lee, W. Watson, "A Power System


Stabilizer Stabilizing Signal for Thermal Units
Based Upon Derivation of Accelerating Power", IEEE
Trans., Vol. PAS-96, Nov./Dec. 1977, pp. 1777-1783.

6)

Schleif, R.K. Feeley, W.H. Phillips, R.W.


Torluemke, "A Power System Stabilizer Application
with Local Mode Cancellation", IEEE Trans., Vol.
PAS-98, May/June, pp. 1054-1060.

7)

P.

F.R.

E.L. Busby, J.D. Hurley, F.W. Keay, C. Raczkowski,


"Dynamic Stability Improvement at Monticello Station - Analytical Study and Field Tests", IEEE
Trans., Vol. PAS-98, May/June 1979, pp. 889-901.

AET

FIGURE Al. SIMPLIFIED MODEL SINGLE MACHINE TO INFINITE BUS


The relationships can be understood with the aid of
Figure Al. This figure represents a simplified model of
a single machine to an infinite bus, similar to that
used in Part I, which identifies the components of
torque, terminal voltage, and the desired transfer
function GEP(s). From this figure, the transfer functions to torque and voltage are obtained as:

3041

ATe/AEPSS =(ATeo +Tep)/AEPSS


= GEP(s) Ms /[Ms

wb

(Ala)
K

le

2
=,GEP(s)[ /K _K
hEt/ EPSS =GEP(s){K6/K2 K5 Wb/[Ms

AE /AE

where

K1 (s) =
e

*(Defined

(Alb)

(s)]
+

22

1~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~I20
800T

(s)]j(A2)AES
WbK(s)
K

400

inertia.

PSS

effective
synchronizing
torque
coefficient,
T
including
demagnetizing effects, amortisseur
circuits, voltage regulator characteristics, and electrical loads.

e/86,

Wb

system radian frequency (377 rad/sec


@ 60 Hz).

Pt

DE /8E'

K
-5
in

8E t /86

I -400
I -80 _

-1I60o

-200_,
0.01

istics:

1.

With no load on the generator, K.5 is positive and


approaches zero as the transmission system becomes
weaker and finally becomes an open circuit.

2.

Under load, K is positive for strong systems but


passes throug? zero and becomes negative as the
transmission system becomes weaker. It is therefore possible to find an operating condition where
K 5 is zero and hence the measure from stabilizer
output to terminal voltage is proportional to the
desired transfer function. Typically, the effect
is small under the conditions where the transfer
function must be measured.

Figures A2a and b show a comparison of the phase


versus frequency characteristics for the transfer functions froth stabilizer output to torque and voltage, and
in addition show the desired transfer function GEP(s).
These are calculated characteristics for a large fossil
unit connected to a strong (A2a) and weak (A2b) system.
It can be seen that the measure of voltage is a good
approximation for the strong system over the entire
frequency range, while the measure of torque is considerably different than the desired transfer function
for frequencies below the local mode. For the weak
system, the measure of voltage becomes less accurate
below the local mode frequency, but is very close beyond
the resonance.

0.05

0.1

(a)

0.2
FREQUENCY

0.5

(Hz)

STRONG SYSTEM

1.0

2.0

5.0

10

2.0

5.0

Xe 0.2 p u1

1200 L

800

The denominator term of Equation Al represents the


characteristic equation of the local mode of oscillation
without the power system stabilizer, while the double
lead term in the numerator implies that electrical
torque cannot be changed in steady-state by modulation

From Equation A2, it can be seen that the transfer


function from stabilizer output to terminal voltage will
be proportional to GEP(s) for the case where K is zero.
This gain represents the effect of rotor ang.e changes
on terminal voltage, which has the following character-

0.02

[9])

mode, with identical phase characteristics beyond the


local mode frequency. However, the voltage regulator
has a significant impact on the term K (s) and invalie
dates this conclusion.

GEP(j)

1I200

K* = 8T/8E'

of the excitation system, which obviously is correct.


For a lightly damped local mode, Equation Alb would
suggest that the transfer function from stabilizer
output to torque is 1800 offset from the desired transfer function GEP(s) for frequencies up to the local

AEP/AEpss

ATe

AEPSS

40

Oa

CD

GEN(jw)

-80

I
< 120 _
a-

1600

-2000

-240
0.01

0.02

0.05

0.

0.2

0.5
P.
(b) WEAKSYSTEM (Xe 0.6p.u.)

FIGURE A2. PHASE CHARACTERISTICS OF MEASURABLE VS DESIRED TRANSFER FUNCTIONS


CALCULATED FOR IOOOMVA FOSSIL UNIT

3042
Combined Discussion 1, 2, 3
C. Concordia (Consulting Engineer, Venice, Florida): The function of a
Power System Stabilizer (PSS) is to provide more positive damping in
order to improve stability in those cases when negative damping determines the stability limit. (As it often does when high-response generator
voltage regulators are used.dl1) Since the PSS operates through the
voltage regulator and excitation system as well as the generator field,
which at the frequencies of interest have an appreciable phase lag; since
the object is to produce a damping torque, or a variation of torque
roughly in phase with generating-unit speed; since this is accomplished
by controlling the field flux; and since it is always easier to produce a
phase lag rather than a phase lead; it would seem that the ideal signal
would be acceleration, which always leads speed by 900, rather than
speed itself. Thus, electrical active power (measured in the direction of
rotation), which is an approximation to electrical torque, which in turn
is an approximation to accelerating torque and thus to acceleration
itself, would seem to be a better signal than speed, which might be, and
has often been, considered as the obvious one. This argument is even
more persuasive when one considers older voltage regulating systems,
where some phase advance may be required even with the
"accelerating" power signal. (Of course, in these cases the voltage
regulation does not produce so much negative damping anyway.) This
argument has led me to prefer power as the signal right from the beginning, in spite of its drawbacks.
If one insists on using a signal that looks more like speed, it has long
been appreciated (and the authors have strongly confirmed) that frequency has the great advantage over speed, that it behaves more or less
like speed for the oscillation modes of most interest, but is greatly attenuated for higher-frequency modes corresponding to oscillations between nearby generators. Since these modes are usually not subjected to
so much negative damping because of the smaller voltage regulator
response, and in addition may have appreciable positive amortisseur
damping, the PSS is not likely to be needed for them. Thus, the PSS
design is made easier by the smaller range for which it must function effectively.
In view of these considerations, I have never understood why speed
seems to have been chosen in so many cases for the PSS input. Either
power or frequency would be better. (I and my coauthor may be accused of having advocated speed in the authors' reference 2 of Part I
(reference 3 of Part II and reference 9 of Part III); but, at least for my
own part, this was not at all intended, and in fact, was specifically
disclaimed in the paper.)
This has been said without regard to the possible effect of the PSS on
generating-unit torsional oscillations. When this problem came to light,
it became evident that, with a speed-fed PSS, about as much effort had
to be expended to ensure that the PSS did not do the wrong thing in the
higher-frequency torsional range as to ensure effective operation in the
intended range of usefulness. To me, this seemed finally to have given
the speed signal the coup-de-grace. It became apparent that there are
many speeds, and it is not always easy to find that any particular one is
appropriate. It is true that filters have been applied, but this added further complication to a device that is often itself regarded as a complication by operators.
In comparing the two remaining signals, power and frequency, we
must point out that power also has the ability to be summed for nearby
generators (easily in the case of generators in the same power stations,
but not generally so easily as frequency). Also, we believe that frequency will require some signal filtering that will aggravate the already 90
phase penalty that it suffers relative to power. On the other hand, use of
electrical power may require some (very approximate) signal of
mechanical input power. (Incidentally, we feel that the power-input
PSS will require a washout, since a perfect steady-state compensation
for power change is not conceivable to us, but we do not share the
authors' fear that the desynchronizing effect may have to be serious or
even appreciable. In our opinion, the authors PSSp example, equation
6, Part II, should have included an appropriate washout to be more
realistic.)

E. V. Larsen and D. A. Swann, Applying Power System Stabilizers,


Pt. I: General Concepts, this issue, pp. 2999-3006 .
2 E. V. Larsen and D. A. Swann, Applying Power System Stabilizers,
Pt. II: Performance Objectives and Tuning Concepts, this issue, pp.
3007-3015 .
3 E. V. Larsen and D. A. Swann, Applying Power System Stabilizers,
Pt. III: Practical Considerations, this issue, PP.3016-3028-

In summary, we are in general agreement with the authors that speed


is not as appropriate as frequency or power, but do not lean as much
toward frequency as they seem to do. It depends on what are the critical
problems in each case. For rather low frequency oscillations between
relatively large areas, frequency may be best, but when torsional oscillations may be a problem power may be best (even though it may be
found advisable to use a "station" power rather than a "unit" power.)
The application of the PSS has so far been a long and fairly complicated procedure, and this has been a handicap to its general use.
Since electric power systems are very rarely well damped, a standardized
and simply-applied PSS, with no attempt made to optimize performance, but with its output comfortably within a range to give moderate
aid to both damping and synchronizing torque, would be extremely
welcome. We look forward to that day, and in the meantime, welcome
the present paper, which by its thorough analysis and explanations of
all aspects, makes the application more rational. I hope I can regard it
as a necessary first step towards such a "standard" PSS.
Following are comments on a few specific points:
1) Regarding the frequency range of concern, we should regard 2.5
Hz as unnecessarily large, as it is difficult to imagine the voltage
regulators producing a net negative damping at such a high frequency.
2) We should like to endorse fully, and emphasize, the remarks
deploring the use of unnecessarily large voltage-regulator gain.
3) We do not believe that any sensible engineer would conclude that
the performance of a power-input stabilizer is identical to that of
a speed-input stabilizer. First, it does not require phase lead in its
transfer function. (It would never be designed as might be implied
by equation 7 of Part I, i.e., starting with the speed signal, and
then integrating!) Second, it is precisely because it does not take
so much notice of the torsional oscillations that it is better. (Incidentally, we feel that the argument leading to the rather roundabout way of arriving at the conclusion that power can be treated
as the derivative of speed after all is completely unnecessary. We
have always found it best to follow the flow of physical action in
deciding how to control any process. In fact, changes in power do
cause changes in speed, which in turn cause changes in angle,
which in turn modify the power. But this last is a feedback
mechanism. From a dynamic point of view, it is power that determines angle, not the other way about.)
4) We cannot agree with the authors statement that the PSS is not intended to enhance transient stability. It was precisely to improve
transient stability that the PSS was first studied and later widely
used by the West Coast utilities. Although there are some cases
where transmission lines (usually tie lines) have been loaded so
severly that spontaneous oscillations have appeared, the more
common case is that these oscillations only appear after a fault has
tripped out a major line and so weakened the system, thus making
the prefault system transiently unstable.
Also, the PSS was not "extended" to the case of radial transmission. For example, we believe the Canadian Peace River application (which incidentally used power) antedated the U.S. West
Coast applications.
5) For the frequency-input PSS the desired gain is stated to be twothirds of the "instability" gain. This seems dangerously close,
especially from the viewpoint of determining the setting from a
single system test of the instability limit.

REFERENCE

11] C. Concordia, Steady-State Stability of Synchronous Machines as


Affected by Voltage Regulator Characteristics, AIEE Trans., Vol.
63, 1944, pp. 215-220, 490.
Manuscript received July 24, 1980.

F. P de Mello and J. M. Undrill (Power Technologies Inc., Schenectady, NY): The authors are to be congratulated on the preparation of
an excellent discussion of stabilization through excitation control.
Although there are a few points of emphasis with which we take issue as
noted further, this three part paper makes a significant contribution on
the subject, covering the many aspects to be considered in stabilizer application and, more importantly, relating cause and effect through a
searching analysis of the basic process and controls.

3043

In Part I, under "basic concepts," several observations are made on


the gain of GEP(s) as function of ac system strength, generator loading,
excitation system gain and bandwidth, and generator open circuit field
time constant. Whereas the statements by themselves are essentially correct, they can be taken out of context and cause concerns with factors
that appear tangential to the main problem.
As developed in the de Mello-Condordia paper (Ref.2), the basic
premise was that the voltage regulator loop (flux loop) should be adjusted to yield a well-damped closed loop response, and it was pointed
out that the criterion for regulator tuning is the open-circuit closed loop
performance. A good adjustment under open-circuit (K, = 1.0, K6 = 1.0)
would guarantee a more damped performance for the flux loop under
load. Voltage regulator transient gain should be proportional to T'do.
A recommendation of Ke = 25 p.u. was made for a machine with a T'do
of 6 sec. This yields a crossover under open circuit of about 4 rads/sec,
and perhaps 2 rads/sec under load as compared with 1 rad/sec given by
the authors for a nominal ac system. Variations in T'do should be offset
by corresponding variations in voltage regulator gain (a T'do of 10 sec
should call for a Ke = 40). It is therefore helpful to reduce the number of
independent parameters to those that are not susceptible to logical compensating adjustment.
The fact that GEP(s) increases with generator loading is fortunate
because the negative damping effect due to voltage regulator action
does generally increase with load, and it is therefore proper that GEP(s)
have the greatest effect for conditions where stabilization is most needed. Proper stabilizer design should make provisions for disconnecting
the signal at sustained power levels below 30 to 50%o load.
The point was also made that GEP(s) increases with ac system
strength. This is only partly true since the gain in question is affected by
the lower bandwidth of the flux loop for small values of K6, and the fact
that, with increased ac system strength, the angle-speed loop oscillation
frequency increases. Also, the need for additional damping through
supplementary stabilization disappears for the case of strong ac systems
(K, is small or even positive) hence the main concern is not that the
stabilization be properly phased for maximum effect under these conditions but merely that it not be harmful.
The authors correctly point out that the use of terminal frequency has
a compensating effect on the net gain of singal to torque as a function
of system strength. This endorses approaches where the frequency of a
voltage synthesized from terminal conditions (et + jix) is used as the
signal rather than shaft speed (Ref. 1). With the value of X =Xq, one
can obtain a voltage which closely follows the rotor angle and, with
smaller values or even negative values for X, the voltage could be made
to represent any point between the rotor and the infinite bus. The advantage of sensing frequency, whether of terminal voltage or internal
voltage, is also that it filters out torsional modes.
While much emphasis is given to the need for higher gain of the signal
as the system strength decreases, our experience has been that changes
in system strength cause changes in the oscillation mode of concern.
The higher the system reactance, the lower is the oscillation frequency
and, at these lower oscillation frequencies, the strength (gain) of the
signal should naturally be lower for the desired effect. This was illustrated in Figures 12 and 13 of Reference 2.
The discussion of power input stabilizers in Part I appears unnecessarily complicated and misleading. Equations 6a to 6c make use of
the relation, P = wY, which is universally true at all perturbation frequencies and mixes this with the relation STe/8d = Kle which is approximate and then only at low perturbation frequencies. Specifically,
this relationship is not valid in the bandwidth of shaft torsional modes,
the stability of which is a key factor raised by the authors in the selection of stabilizer inputs. The result of proper analysis of the torque/angle and torque/speed sensitivity in the broader bandwidth yields
terms with sum and difference frequencies, (wo + wO and (wo- wD, and
8 Te/8 d is not a scalar but rather is a complicated transfer function that
can yield positive or negative damping depending on the perturbation
frequency and network characteristics.(3,4) The Heffron and Philips
model used in Ref. (2) was never intended for analysis of interactions
with torsional modes. If torsional mode damping is to be discussed, the
modeling level which is the subject of this paper is entirely inadequate.
In the authors' discussion of the power plus speed functional implementation (Ref. 5), the problem of shaft torsional interaction is
again brought up. We would emphasize that in any such implementation, our intent is to use an inferential speed signal measured in such a
way as to be insensitive to torsional effect. This measure, along with the
reduced bandwidth needed in this scheme for the rate of change of
speed, should minimize the possibility of interaction problems with
shaft modes.

REFERENCES
1. F. P. de Mello, "The Effects of Control," tutorial paper on
"Modern Concepts of Power System Dynamics," IEEE Tutorial

70M62-PWR.
2. F. P. de Mello and C. Concordia, "Concepts of Synchronous
Machine Stability as Affected by Excitation Control," IEEE Transactions on Power Apparatus and Systems, Vol. PAS-88, April
1969, pp. 316-329.
3. J. M. Undrill and T. E. Kostyniak, "Subsynchronous Oscillations:
Part I, Comprehensive Stability Analysis," IEEE Transactions on
Power Apparatus and Systems, Vol. PAS-95, pp. 1446-1455.
4. J. M. Undrill and F. P. de Mello, "Subsynchronous Oscillations:
Part II, Shaft System Dynamic Interactions," IEEE Transactions
on Power Apparatus and Systems, Vol. PAS-95, pp. 1456-1464.
5. F. P. de Mello, L. N. Hannett and J. M. Undrill, "Practical Approaches to Supplementary Stabilizing from Accelerating Power,"
IEEE Transactions on Power Apparatus and Systems, Vol.
PAS-97, Sept./Oct. 1978, pp. 1515-1522.

Manuscript received July 24, 1980.

Kundur, and H. M. Zein El-Din (Ontario Hydro,


Toronto, Canada): The authors have done an excellent job of bringing
together information from many sources and providing an in depth
look at the application of several types of power system stabilizers.
As the authors have pointed out, in Ontario Hydro we have generally
used high excitation transient gain of the order 200 pu Efd/pu Etref.

D. C. Lee, P.

We use power system stabilizers almost exclusively to damp machinesystem (local) modes and in our system we do not have a conflict between the requirements of local and inter-area modes. Our investigations
show that the use of transient gain reduction (TGR) does not contribute
to any significant improvements to the damping of local, exciter and
torsional modes [A]. Reduction of transient exciter gain is usually unacceptable due to transient stability considerations. We allow for a
stabilizer out of service condition by designing the system to
automatically transfer to either a manual control or an alternate
regulator in case of stabilizer failure. Even without this feature, loss of
stabilizer on one unit at a plant with several units would not lead to

dynamic instability.
While we are not in favour of universal use of TGR, we have justified
its application in specific cases. We have used TGR on a hydraulic unit
equipped with a thyristor exciter. The other units in the station have
slow rotating exciters and we found it necessary to use TGR with the
thyristor exciter to ensure dynamic stability in the event of loss of the
stabilizer signal. High transient performance was not a requirement at
this station and there was no provision for regulator transfer in the
event of loss of stabilizing signal. We have also recommended the use of
TGR in one situation where a unit must operate into both very high and
extremely low impedance systems [B].
With a low regulator transient gain in the neighborhood of 20 pu
Efd/pu Etref, stabilizer output limits of 0.1 pu Etref used by the
authors could be restrictive for high frequency excursions where Et can
not respond. In our applications with nearly 10 times higher regulator
gain, we use a positive limit of about 0.2 pu to ensure maximum contribution of the stabilizer. This is complemented by a terminal voltage
limiter which prevents the terminal voltage from exceeding a set level.
On the negative side, a stabilizer output limit of 0.05 to 0.1 pu is used.
This allows sufficient control range while reducing the probability of a
unit trip for failure of a stabilizer component driving the output signal
to the negative limit.
We would suggest that the power based stabilizer configuration used
for the example in this paper is not directly comparable with the speed
based stabilizer considered. The power based stabilizer transfer function, given by equation 6 of part II, can be manipulated to a speed based stabilizer of the form:

0.5s
(1+0.25s)
(1+0.5s)
(1+0.5s) (1+0.05s)
(1+0. 06s)
Of particular significance is the effective 0.5s "washout". The low
value of this time constant may be a contributing factor to the stability
of the low frequency voltage regulator mode and the cause of the poor
showing of this system in the transient comparison tests.
The effect of mechanical power variation in the system shown in
Figure 2 of part III can be mitigated to the extent desired by suitable
choice of Em(s). In the limiting case, this could be a full blown torsional
K

filter. In practice, universal filters can be chosen which will allow very

3044

fast mechanical power change with minimal terminal voltage disturbance and no governor interaction.
Could the authors elaborate on their concern for torsional interaction
with power input stabilizer when applied to four-pole units. We have a
similar stabilizer in operation on our 750 MW, 1800 RPM nuclear units
equipped with thyristor exciters. For these units the lowest torsional frequency is 9 Hz and the local mode frequency is about 0.8 Hz. Our experience shows that with thyristor exciters there is no need to use torsional filters for stabilizers using accelerating power as input even for
four-pole units. Is there anything inherent with the GENERREX excitation system that causes this concern?

REFERENCES
A. P. Kundur, D. C. Lee, H. M. Zein El-Din, Closure of "Power
System Stabilizers for Thermal Units: Analytical Techniques and
On-site Validation", paper no. F80227-9, presented at the IEEE
PES Winter Meeting, New York, NY, Feb. 3-8, 1980.
B. D. C. Lee, P. Kundur, H. M. Zein El-Din, discussion of M.
Mobarak, D. H. Thorne and E. Hill, "Contrast of Power System
Stabilizer Performance on Hydro and Thermal Units", paper no. F
79 659-4, IEEE PES Summer Meeting, Vancouver, Canada, July
15-20, 1979
Manuscript received July 29, 1980.

P. Bonanomi and R. Bertschi (Brown Boveri & Co. Ltd., Baden


Switzerland): The authors have presented an interesting and useful
analysis of the many aspects of applying power system stablizers. The
discussors wish to add a few comments on the choice of the input
signals. A question is also raised as to a practical aspect in tuning the

stabilizer.
The paper contains a collection of interesting arguments concerning
the adequacy of the different input signals for the stabilizer. Each
choice of one such signal is shown to have its own advantages and
drawbacks. Considering the same arguments, the discussors feel that a
combined feedback of power and frequency or power and speed provides better results [A]. This case has not been considered in the paper.
One of the main disadvantages in using power feedback alone is that the
required integrating characteristic (lag/lead) causes adverse effects in
case of mechanical power variations. This troublesome lag/lead term is
not necessary when an additional frequency signal is used, thus reducing the detrimental effect to a great extent. The combined use of power
and frequency also eliminates the need for differential terms (lead/lag)
in the frequency or speed feedback, which attenuates the noise problems due to high gains at high frequencies. The authors of the paper did
not deal with the combined use of two feedback signals in their paper,
and may have some comments which would be most appreciated.
When power feedback is used alone, the detrimental influence of the
PSS during mechanical power variations may be reduced by an alternative signal connection. The improvement is obtained by inserting the
PSS signal after the AVR compensator instead of the input summing
point. The integrating characteristic of the AVR compensator is thus
bypassed. Damping values are the same when the transfer function is
adjusted properly. Did the authors make any tests with this circuit arrangement?
As to the tuning of the PSS, we see some advantage in partitioning
the transfer function into two parts with individually adjustable gain
values, permitting adjustment of gain and phase. This form of transfer
function lends itself to the use of the so-called "Domain Separation"
method of optimization, which yields the optimum gain values, the attainable damping, the sensitivity of gain variations upon damping and a
check on the stability of the so-called exciter mode at very moderate
computer costs. Do the authors have any comments to this tuning
method?
Optimal damping of electromechanical oscillations is a well accepted
criterion for tuning a stabilizer. This performance goal may result in
high stabilizer gains, however, which is regarded as a drawback by some
utilities. The high gains may indeed produce unnecessary deviations in
the excitation voltage which give rise to fluctuations in the terminal
voltage. In steady state situations where stability is not really critical the
gains may be reduced for smoother voltage control. This suggests a kind
of adaptive tuning of the stabilizer gains. The question remains open,
however, as to how much effort is justified in trying to avoid these
voltage fluctuations. The discussors would be pleased to know the
authors opinion on this matter.

REFERENCE

[A] F. Blaser, "Improved Stabilizer Signal for Smooth Active Power


Transmission", Brown Boveri Review, Vol. 67, No. 3, 1980.
Manuscript received August 7, 1980.

J. A. Pinnello and S. T. Naumann, (Commonwealth Edison Company,


Chicago, IL): The authors are to be congratulated on this paper and the
two companion papers which give a most comprehensive presentation
of the theory and application of power system stabilizers. The series of
papers is a valuable reference for anyone working in this area. The
following discussion gives some of our experience in applying stabilizers
to several large units on our system.
The papers give a clear explanation of the tradeoffs involved in obtaining satisfactory performance for local mode, interarea and large
disturbance considerations. In conducting transient stability studies
with the stabilizer simulated, we have noticed some degradation due to
the stabilizer as the authors caution. In addition, actual line dropping
tests showed a slower rise in the generator field voltage when the
stabilizer was in service.
The author's detailed comparison of three alternative stabilizer input
signals is a most welcome contribution. Some years ago in our first
stabilizer application, we encountered severe noise problems using ac
frequency from a zero crossing transducer and settled for a speed
signal. In our most recent application, which has given trouble free performance for about three years, we also used a speed signal, since a
suitable frequency transducer was not available at the time.
The authors make an interesting statement in Part I of this series of
papers, when discussing the characteristics of the speed input signal.
Since the stabilizer loop is least stable under strong system conditions,
this limits the gain of the stabilizer. However, higher gains are desirable
under weak transmission system conditions. As the authors stated, this
problem could be solved with adaptive gain control. At our Powerton
Station, we have installed an adaptive control system on the underexcited reactive ampere limiter (URAL) to fully utilize the reactive
capability of the machine under strong system conditions [1]. Could the
authors comment on the feasibility of such a system applied to a speed
input signal power system stabilizer having two gain settings (one for
the strong system and another for the weak system).
REFERENCE

[1] G. L. Landgren, "Extended Use of Generator Reactive Capability

By A. Dual Underexcitation Limiter", Paper F79-725-3, IEEE


Transactions on Power Apparatus and Systems, Vol. PAS-99, pp.
1381-1385, July/Aug. 1980

Manuscript received July 21, 1980.

E. V. Larsen and D. A. Swann: We thank the discussors for their kind


remarks and for raising several technical points, which we will attempt
to clarify in our closure.
To the comments of Messrs. de Mello and Undrill regarding voltage
regulator crossover frequency, our experience has been that the voltage
regulator eigenvalue lies between I and 2 rad/sec with the machine connected to the transmission network. With a transient gain of 20 per unit,
this is consistent with the comments made by the discussors. We might
also note that in Ontario Hydro's discussion of de Mello and
Condordia's paper (Reference 2 of Part 1), their experience with very
high regulator gains on a fast excitation system gave no voltage
regulator loop instabilities. The criterion for determining voltage
regulator gain on fast-acting excitation systems has not necessarily been
the damping of the voltage regulator loop under open-circuit conditions, but rather to minimize the negative damping introduced by the
voltage regulator action with the machine operating at full load into a
weak transmission system, without a stabilizer. The gain of 20 per unit
commonly used on today's excitation systems originated from the
voltage regulator criterion proposed by de Mello and Concordia, but
has been retained because of the impact of voltage regulator action on
oscillatory mode stability limits.
We question the need to disconnect the stabilizer below 30 to 50%
load as proposed by de Mello and Undrill, since, even though it may not
be needed for system stability, it will not do any harm since GEP(s), and
hence the stabilizer loop gain, is reduced under this condition. The
variation of GEP(s) with system strength is a crucial point in designing

3045
stabilizers. For oscillation frequencies beyond the voltage regulator
bandwidth, the variation in GEP is due only to variations in K2, which
increases with system strength. Below the crossover frequency of the
voltage regulator, it is further amplified by a reduction in K6 as the
system becomes stronger. As was emhasized in the paper, there exists a
"tuning condition" for the stabilizer under which one must be sure no
undesirable interactions exist and which therefore determines the maximum gain of the stabilizer. The "performance condition" under which
the stabilizer must provide damping is not necessarily the same as the
tuning condition. For speed and power input stabilizers, the tuning condition is with a strong ac transmission system since this gives the highest
loop gain, whereas the performance condition is with a weak transmission system where the impact of the stabilizer is attenuated due to a
reduction in GEP(s). This is illustrated in Figures 4 and 6 of Part 2 of
the paper.
We did consider the use of a compensated frequency signal synthesized from terminal conditions, as proposed by de Mello and Undrill.
While this certainly gives more flexibility in designing a stabilizer to add
damping to a specific mode, we found that the use of terminal frequency or frequency from the high side of the generator step-up transformer
gives a reasonably good compromise between the disadvantages of
generator shaft speed resulting from compensating with X = XQ, and
the reduction in performance arising from sensing a signal too far out in
the system. We disagree with de Mello and Undrill's comments regarding the "filtering" of torsional modes when using frequency input. It is
true that a frequency-based stabilizer requires less torsional filtering
than a speed-based device, but for the torsional interaction
phenomenon, ac bus frequency can be considered as primarily proportional to the speed of the generator shaft. The proportionality constant
used for filter design depends on the worst case situation; for terminal
frequency, the worst case would be with a weak external transmission
system since this yields the strongest coupling to generator shaft speed,
with Af 0.8 ACOG. For a synthesized signal obtained from terminal
conditions, compensating for the subtransient reactance yields Af
A(OG in the torsional frequency range. Compensation for the
quadrature-axis reactance XQ will yield a coupling factor greater than
unity for strong transmission systems and hence more torsi6nal interaction than using generator shaft speed. The benefits gained with respect
to torsional filtering requirements by utilizing a frequency signal is twofold: 1) The generator shaft speed generally has less torsional motion
than either the front or rear standard positions, and 2) because of the
self-compensating nature with respect to transmission strength, tuning
of the frequency input stabilizer yields a lower bandwidth than a speed
input stabilizer having equivalent response in the performance condition.
Both Concordia and de Mello commented on our discussion of power
input stabilizers in Part 1. In this discussion, we were trying to provide a
theory for stabilizer application based on an equivalent damping path
from speed to torque. Such a theory allows for analysis of both electromechanical oscillations and torsional interaction. In hindsight,
however, this is probably not the best way to present the concepts. We
offer the following comments in clarification. For analysis of the interaction of the stabilizer with system modes of oscillation below a few
Hz, Equations Cl and C2 are applicable. The stabilizer design follows
that of the speed input stabilizer design, taking advantage of the inherent 900 phase lead indicated by Equation C2, once a measure of
mechanical power has been obtained to satisfy Equation Cl.
accel

Pn

(C1)

(C2)
s 2HwG
For analysis of torsional interaction, however, the correct accelerating
power experienced by the generator mass must be used, as per Equation
C3.

paccel

@G[KTG()T

KGE(EG

(C3)
Pe
for
this
The assumptions behind Equations Cl and C2 are invalid
situation, and the more comprehensive theory must be utilized to
analyze the effective damping path from speed back through torque
caused by the stabilizer. This analysis method requires that one consider
a speed change being the initiating event (even though it is caused by a
disturbance elsewhere which finally causes the speed change), and a torque change caused by the stabilizer resulting from the speed change.
We agree with de Mello and Undrill that a Te! a is not a scalar, and
point that they missed the indicated functional dependency of Kle(s) on
frequency. This is intended to be a transfer function having a similar

eG)

meaning as K, in the Heffron-Philips model, but evaluated with a


generator and transmission model valid for the phenomenon being investigated. With an appropriate model, the value of Kle(s) in the torsional frequency range is slightly different than the steady-state value,
but it is still close to being a scalar and has roughly the same functional
dependence on loading and system reactance as does the steady-state
value. It is therefore an effective tool for presenting concepts, even
though evaluation of torsional filtering requirements for particular application is best done with eignevalue analysis with a program such as
MANSTAB [Cl] having a system representation valid in the torsional
frequency range.
Messrs. Lee, Kundur, and Zein-EI-Din make a good point by showing the equivalent speed-based stabilizer for the power-based stabilizer
design used in Part 2 of the paper. The first lag term of 0.5 seconds in
the power-based stabilizer design turns into a 0.5 second washout in a
speed-based design. The discussors suggest that this low washout time
constant could give enough desynchronizing effects to cause a poor performance of the power-based stabilizer in a large-signal application. We
would add that our choice of time constants at .25 and .5 seconds was
based on optimizing the performance for the power-based stabilizer for
a weak transmission system. Thus, our point regarding the potential
detrimental impact of excessive damping contribution remains valid.
Nearly all the discussors commented on the use of additional signals
to compensate for mechanical power. The basis for the observations of
Messrs. Bonanomi and Bertschi lie in the concept developed by de
Mello (Reference 11 of Part 3). As we indicated in Part 3 of the paper,
an optimum value of filtering of the mechanical power signal exists,
beyond which the torsional interaction due to the power path itself
becomes dominant. In addition, the mechanical power filter design
must take into account fast valve action. Messrs. Lee, et all. indicate
that they have achieved such a filter design for their units. Dr.
Concordia also points out the need for a washout to mitigate the impact
of an inperfect steady-state compensation for mechanical power, which
we also addressed in Part 3 as a practical consideration. As previously
noted, this would be equivalent to adding a second washout to a speedbased stabilizer and hence, the time constant must be greater than about
three seconds so that it does not cause additional desynchronizing effects from the stabilizer.
Messrs. Lee, et al. raised a question about torsional interaction with
four-pole turbine-generator units having a GENERREX* excitation
system. For the low frequency torsional modes of concern with the
four-pole unit, i.e., from about 6 to 8 Hz, the torsional interaction with
a GENERREX system, either compound power source or potential
power source, is roughly the same as that due to a bus-fed excitation
system having the same wide-bandwidth control. Our conclusions are
based on a conservative estimate of inherent mechanical damping and
assuming generator shaft speed was used as an input signal. Ontario
Hydro's practice of utilizing a speed pickup at a node of the first torsional mode would significantly reduce or- eliminate this filtering requirement. We would caution, however, that use of a combined front
and read standard signal to provide mode 1 cancellation, as opposed to
a mechanical measurement at the torsional node, carries with it a risk of
torsional instability should one of the speed signals be lost.
Dr. Concordia is obviously not in favor of high voltage regulator
gains, even though the experience of Ontario Hydro appears to have
been quite satisfactory. We would only comment that the Research
Department of Ontario Hydro has been extremely active in all aspects
of stabilizer applications, and have implemented controls which utilize
their excitation systems to the utmost for stability enhancement. This
has been done in conjunction with extensive studies specific to their
system. This is somewhat unique in the electric utility industry, and for
general application moderate gain in both the voltage regulator and
power system stabilizer is perhaps the best philosophy, leading to Dr.
Concordia's hope for a "standard" power system stabilizer.
As indicated by Messrs. Lee, et al. power system stabilizer output
limits can have a significant impact on performance following a major
disturbance. Setting the negative limit to a smaller value than the
positive limit is an excellent idea. As long as there is some takeover
regulator and/or protection for excessive generator flux, we see no
practical need to place an upper bound on the value of the positive
limit.
Messrs. Bonanomi and Bertschi raised the issue of steady-state
voltage fluctuations due to stabilizer action, which arise from their
design criterion of optimal damping for electromechanical oscillations.
We feel a more appropriate design criterion is to provide adequate
*

Trademark of General Electric Company

3046

damping so that oscillatory instabilities are prevented, recognizing that


the primary function of the excitation system is to regulate terminal
voltage and not to damp electromechanical oscillations. Our experience
has been that this criterion does not lead to excessive voltage fluctuations. We are aware of one situation, however, reported in Reference 11
of Part 2, where a single unit was utilized to damp tie-line power for a
large area. In such a situation, voltage fluctuations were of concern and
a tradeoff between desired damping and voltage regulation was required. The adverse impact of voltage fluctuations is primarily subjective, unless load tap changers of voltage regulating distribution
transformers are exercised, in which case it could have a significant impact on equipment. Our experience has been that, as long as the
stabilizer does not try to do more than its fair share of system damping,
voltage fluctuations are not a significant problem.
Messrs. Bonanomi and Bertschi also discuss use of individual gain
and phase adjustments in tuning the stabilizer on the computer. Utilizing lead/lag circuits to provide phase shifting, the gain and phase can be
specified only at one particular frequency, and we presume this is what
they refer to. The tuning procedure indicated by the results presented in
Part 2 consist of first adjusting the lead/lag settings to obtain an appropriate phase shift versus frequency characteristic, then to determine
gain based upon a root locus calculation or a gain margin test in the
field. This method provides insight into the characteristics of the particular application, and computer costs are negligible, particularly in
comparison to setting up the equipment in the field. With respect to
connecting a power system stabilizer downstream of the voltage
regulator summing junction, we feel that there would be absolutely no
difference in transient performance providing the excitation system and
stabilizer remain within the limits. Such a connection would have different performance for large disturbances where nonlinear effects come
in and there may be some advantage for this situation.
We have already discussed Dr. Concordia's points 2 and 3. To his
first point, we chose a value of 2.5 Hz assuming that some of the local
modes in the area may have been increased in frequency by addition of
stabilizers on other units. Such would be the case for a plant having a
strong intermachine mode and a speed-based stabilizer on one of the
units. In addition, introduction of a stabilizer even on a single unit increases the frequency of the local mode oscillation or introduces a
higher frequency exciter mode. For speed- and frequency-based
stabilizers, it is always the highest frequency mode which limits the obtainable gain. His point number 4 arises from differing interpretations
of the term "transient" stability. Our use was with respect to the first
swing, whether it be of the local mode or of an area mode, whereas Dr.
Concordia interprets the transient to consist of everything between the
two steady-state conditions. Thus, we should have distinguished between "first swing" stability and "oscillatory" stability. To his fifth
point, we agree that only a 4 db gain margin for the optimum damping
with a frequency-based stabilizer may be less than desirable for conservative control system design practices, to which we commented in Part 3
of the paper. However, utilizing the same 10 db gain margin with frequency as would be used with speed, will result in identical performance
for the weak transmission system condition; a higher gain will give bet-

ter performance than would be possible with a speed-based stabilizer.


An additional comment is warranted on the use of gain margin as a
design criterion. With a power-based stabilizer, or with speed- or
frequency-based stabilizers having no torsional filtering requirements,
optimum damping occurs with gain margins in excess of 14 db, and
varies considerably between different applications. For these situations,
it is best tocalculate the gain to achieve a desired damping contribution
as per Equation A5 of Part 1, rather than determining the instability
gain and backing off a specified amount.
Dr. Concordia feels that either frequency or power offer the best
choices for a power system stabilizer input signal. While Messrs.
Pinnello and Naumann point out that adaptive gain control can be used
to make a speed-based stabilizer be sensitive only to the modes of concern, this function exists inherently when using frequency as an input
signal. Although the power-based stabilizer is still more sensitive to the
local modes relative to system modes than the frequency-based
stabilizer, it has the advantages of requiring little torsional filtering and
a high gain margin.
Frequency input has an advantage in reliability since it requires only
the terminal voltage signal, which is always available in the voltage
regulator cubicle and is required for automatic voltage regulation
anyway. Power input requires a measure of stator current and some
measure of shaft speed, either directly via a transducer or by synthesis
from current and voltage. In addition, to make a power-based stabilizer
sensitive only to the system modes of concern, it would be required to
obtain power signals from several other locations to sum them. Design
of the filtering (Fm(s)) required on the synthesized mechanical power
signal requires consideration for fast valving action, which would involve a large number of nonlinear simulations including a valid
representation of the boiler and turbine controls.
The potential problems which may occur with a frequency input
stabilizer are: 1) insufficient gain margin leading to an instability in the
2 to 3 Hz region, or 2) insufficient torsional filtering leading to a torsional instability. Either of these can be tested readily when the equipment is commissioned. On the other hand, the potential problems
associated with using a power-based stabilizer include excessive damping contribution for a weak transmission system and excessive filtering
in the synthesized mechanical signal path. The potential adverse consequences would show up only following a major system disturbance and
it would be difficult or impossible to verify the design adequacy based
on simple field tests.
The publication of this three-part paper culminates many years experience. Our intent was to thoroughly document the current state-ofart on power system stabilizer applications to provide a basis for further
advancements. Judging from the discussor's comments, this objective
appears to have been achieved.

REFE1RENCE

[Cl]

E. V. Larsen, W. W. Price, "MANSTAB/POSSIM Power System Dynamic Analysis Programs...", IEEE PICA Proceedings,
1977, pp. 350-359.
Manuscript received February 17, 1980.

You might also like