09TP250
Sponsored by
and the
of the
Presented at
Tampa
Florida
1
IEEE Tutorial Course Power System Stabilization via Excitation Control June 2007
CONTENTS
CHAPTER 1 REVIEW OF FEEDBACK CONTROL CONCEPTS ........................................................................................ 7
I. INTRODUCTION ..................................................................................................................................................................... 7
II. LAPLACE TRANSFORMS .................................................................................................................................................... 7
III. TRANSFER FUNCTIONS AND BLOCK DIAGRAMS.................................................................................................................. 9
A. Transfer Function Example.......................................................................................................................................... 9
B. Block Diagrams............................................................................................................................................................ 9
C. Interconnection Of Systems........................................................................................................................................ 10
D. B.D. Manipulation Example....................................................................................................................................... 10
E. Transfer Function Example........................................................................................................................................ 10
IV. FREQUENCY RESPONSE MODELS .................................................................................................................................... 11
V. STABILITY CRITERIA FOR FEEDBACK CONTROL SYSTEMS .............................................................................................. 11
A. General Comments..................................................................................................................................................... 11
B. Nyquist Criterion (Gain And Phase Margins) ........................................................................................................... 12
C. Example...................................................................................................................................................................... 14
D. Root Locus Example................................................................................................................................................... 15
VI. STATESPACE TECHNIQUES............................................................................................................................................. 16
A. General Comments..................................................................................................................................................... 16
B. StateSpace Models .................................................................................................................................................... 17
C. State Space model Example ....................................................................................................................................... 17
VII. SYSTEM SIMULATION ...................................................................................................................................................... 18
CHAPTER 2 OVERVIEW OF POWER SYSTEM STABILITY CONCEPTS ................................................................... 19
I. INTRODUCTION ................................................................................................................................................................... 19
II. POWER SYSTEM STABILITY CLASSIFICATION .................................................................................................................. 19
A. Definition.................................................................................................................................................................... 19
B. Categories Of Stability ............................................................................................................................................... 19
1) Voltage Stability..................................................................................................................................................... 19
2) Frequency Stability ................................................................................................................................................ 19
3) Rotor Angle Stability ............................................................................................................................................. 19
III. BACKGROUND  GENERATOR CONNECTED TO INFINITE BUS ........................................................................................... 20
IV. TRANSIENT STABILITY .................................................................................................................................................... 20
V. TRANSIENT VS. OSCILLATORY STABILITY ...................................................................................................................... 21
VI. OSCILLATORY STABILITY ............................................................................................................................................... 21
A. Characteristic Dynamic Equation.............................................................................................................................. 21
B. Local Vs. Inter Area Oscillations............................................................................................................................... 22
C. Negative Damping Due To Voltage Regulator .......................................................................................................... 22
D. PSS For Improved Oscillatory Stability..................................................................................................................... 24
VII. GENERATOR MODELS ..................................................................................................................................................... 25
VIII. CONCLUSIONS ............................................................................................................................................................. 25
CHAPTER 3 PERFORMANCE CRITERIA AND TUNING TECHNIQUES..................................................................... 26
I. INTRODUCTION ................................................................................................................................................................... 26
II. PERFORMANCE OBJECTIVES ............................................................................................................................................ 26
A. Oscillatory Stability Limits......................................................................................................................................... 26
B. System Modes Of Oscillation ..................................................................................................................................... 26
C. Tuning Concepts......................................................................................................................................................... 27
D. Speed Input Stabilizers............................................................................................................................................... 28
E. Frequency Input Stabilizers ....................................................................................................................................... 28
F. Power Input Stabilizers .............................................................................................................................................. 29
III. TUNING EXAMPLE ........................................................................................................................................................... 29
A. Phase Compensation.................................................................................................................................................. 30
B. Root Locus.................................................................................................................................................................. 30
C. Step Test and Fault Simulations................................................................................................................................. 31
D. PSS Torsional Interaction ....................................................................................................................................... 32
E. Use of Modified Leadlag Compensation................................................................................................................... 33
2
IEEE Tutorial Course Power System Stabilization via Excitation Control June 2007
3
IEEE Tutorial Course Power System Stabilization via Excitation Control June 2007
4
IEEE Tutorial Course Power System Stabilization via Excitation Control June 2007
FOREWORD
In 1981 the first IEEE tutorial course on power system stabilization via excitation control was presented by
Ken Bollinger, Joe Hurley, Frederick Keay, Einar Larsen and David Lee and the notes from that tutorial became a
widely used reference for generation engineers working to improve power system stability.
Ideas for power system stabilization using excitation control originated as a result of electric power oscillations
occurring on interties between large power pools and stability problems associated with single generators, or
banks of generators connected to large power systems. Before the 1981 tutorial increasing use of high gain
excitation systems and increased use of transmission systems had been leading to decreased stability margins and
power system stability problems. A considerable amount of effort had been spent on research projects, sitework
and the development of control electronics to stabilize multimachine systems.
Since the 1981 tutorial, research work and product development has continued. Electronic control units now
utilize digital technology to provide repeatability, more features and easier use, and control algorithms have been
improved to eliminate some earlier shortcomings.
The development of the integral of accelerating power type stabilizer (PSS2B type as described in IEEE
Standard 421.5 2005) allowed stabilizers to operate successfully with minimal terminal voltage fluctuation even
during very rapid loading and unloading of generators. This type has now become the defacto standard and this
type of stabilizer is now a requirement in many parts of North America.
Increasing numbers of power system stabilizers have been installed as grid codes around the world and North
American requirements such as the WECC guidelines have demanded that generators be equipped with
stabilizers. In addition the power system disturbances in Western North America during July and August 1996
caused increased effort to be focused on testing and validating of generating units including their excitation
control systems and associated stabilizers. These factors have resulted in many more engineers being introduced
to power system stabilizers for the first time.
Whilst excellent papers are available on many aspects of power system stabilizer design, implementation and
testing, this tutorial is intended to provide engineers and technicians with a set of key insights into problems
related to power system oscillations and the currently available solutions.
It is expected that the course participants will have a basic understanding of power system analysis and control
concepts. The tutorial includes introductory material to provide a basis for understanding of the terminology used
in the latter part of the course.
This tutorial includes contributions from present day experts in the field of power system stability in addition to
material from the authors of the original tutorial. The first two chapters review feedback control and power
system stability concepts. Following chapters describe more detail on performance criteria, tuning techniques,
accelerating power type stabilizers, field testing techniques and application considerations. The last chapter
describes some future directions in stabilizer design.
The authors and presenters of this tutorial are grateful to the authors of original tutorial and to the experts in
this industry who have written many excellent papers referenced here.
Robert ThorntonJones
March 2007
5
IEEE Tutorial Course Power System Stabilization via Excitation Control June 2007
6
IEEE Tutorial Course Power System Stabilization via Excitation Control June 2007
7
IEEE Tutorial Course Power System Stabilization via Excitation Control June 2007
The roots of the numerator of the transfer function are e at sin (t )
called the zeros of the system and the roots of the + (s + a ) 2
2
8
IEEE Tutorial Course Power System Stabilization via Excitation Control June 2007
9
IEEE Tutorial Course Power System Stabilization via Excitation Control June 2007
elements of the physical system and the blocks of the Laplace D. B.D. Manipulation Example
B.D. Derive the T.F., C(s)/R(s), for the feedback system shown
The basic element of B.D. representation is the transfer below in Fig.1.5.
function. The input/output relation,
Therefore
E (s) 1
= (1.22)
R (s) 1 + G (s)H(s)
 10 
IEEE Tutorial Course Power System Stabilization via Excitation Control June 2007
IV. FREQUENCY RESPONSE MODELS of the transfer function can be obtained by summing up the
By definition, the frequency response of a system is the contribution to the Bode plot of each term as seen in Fig.1.6.
ratio of the output over the input when the input to a system It can further be noted from Fig.1.6 that the complex pair can
is a sinusoidal forcing function of radian frequency and give very sharp attenuation or magnification in the magnitude
steady state conditions have been reached. plot at frequencies in the vicinity of the "resonant humps" of
We are already familiar with sinusoidal steadystate these terms. This feature is important in the design of "pass"
methods of analysis from circuit analysis. and "stop" band filters; a concept that is used in the design of
Theoretically, the frequency response of a system can be filters to block generator shaft "torsional" frequencies from
obtained from the transfer function by setting s = j in the feeding back into the AVR through the PSS.
transfer function. The result is, Bode diagrams are also used to plot experimental
frequency response data when measurements have been made
Y( j) using a sinusoidal generator and recording equipment. In
= G (j) (1.23) these cases a transfer function can be obtained from the
X( j)
frequency response data by doing a "manual" or
"computerized" analysis to best fit the data. In some
Let us consider a general frequency response function that situations frequency response data is used directly to obtain
includes all types of terms as follows: data for assessing stability of existing systems or for
designing controllers to enhance the stability of the control
2
j loop. These concepts will be discussed later.
K(1+ TZ j) (1+ 2( )+j )
nz nz (1.24) It will be seen in the next section that transfer functions
G( j) =
j
2 and frequency response concepts form a vital basis for
( j) n (1+ Tp j)(1+ 2( )+ j ) assessing stability of certain types of control systems.
np np
 11 
IEEE Tutorial Course Power System Stabilization via Excitation Control June 2007
Figure 1.6 Graph Showing Contribution Of Each Term Of (1.24) To Bode Plot G
The rootlocus method is a graphical technique for plotting the effects of noise. The correlation between the transient and
closedloop poles as a function of some system parameter frequency responses is indirect, except for the case of second
commonly the gain K. order systems.
In contrast to the rootlocus method, the Nyquist criterion There are many other sophisticated computerized tuning
gives information on the stability of a feedback system. The techniques described in the literature that have been applied
Nyquist method can be applied not only to systems for which to simulated power system models or Scaleddown laboratory
mathematical description is available but also to systems for systems. These include optimal control, optimal output
which experimental frequency response information is feedback control [9]  [11], pole placement techniques [12],
available. The Nyquist criterion gives relative stability minimum variance methods [13], [14] and frequency domain
information in the form of gain and phase margins. The techniques [15]. The need for imposing additional design
Nyquist criterion uses a polar plot of the openloop transfer constraints (such as minimizing complex performance indices
function. related to system transient swings) on the controller does not
Another frequency response method that is widely used is appear to be of prime concern to Utilities. Presently most
the Bode diagram. In the Bode gain plot, gain in decibels  common method for tuning PSS consists of improving
dB is plotted against on logarithmic scale. In the Bode damping (i.e. moving characteristic roots).
phase plot the phase in degrees is plotted against on B. Nyquist Criterion (Gain And Phase Margins)
logarithmic scale. The stability information of Nyquist
The Nyquist criterion utilizes either the transfer function or
criterion is easily transferred to the Bode plots. In addition
experimentally obtained frequency response data and gives a
magnitude Bode curve can be easily plotted using asymptotic
measure of the relative stability of the roots of the
approximation. Since Bode plots are much more widely used
characteristic equation.
than the Nyquist plots, we will discuss Bode plots in more
Consider the closedloop system shown in Fig.1.7. If a
detail.
sinusoidal signal, R(s), is applied it will be modified and
Reader should note that Rootlocus and Bode plots
complement each other. While rootlocus gives the exact phaseshifted by G(j) and H( j) so that the feedback signal
closed loop pole locations, the Bode plots give relative Z(j) will subtract vectorially from the input R(j).
stability information in the form of gain and phase margins.
The rootlocus approach gives information about the stability E(j) = R(j) Z(j) (1.27)
and the transient characteristics of the system. The frequency
response approach can be used to design systems to minimize
 12 
IEEE Tutorial Course Power System Stabilization via Excitation Control June 2007
PM = GH 180o (1.30)
 13 
IEEE Tutorial Course Power System Stabilization via Excitation Control June 2007
 14 
IEEE Tutorial Course Power System Stabilization via Excitation Control June 2007
The characteristic equation is, Where Zi and Pj are referred to as the zeros and poles
of GH. It follows that the roots of the characteristic
1 + G(s) H(s) = 0 (1.35) equation must satisfy,
n (s + p n ) m (s + Pm jQ m ) = 0 (1.36)
( )
(s + Z i ) s + Pj = 180 q360 (1.44)
For K= 0 to
where pn are the real roots of (1.35) and Pm jQm are
the complex conjugate roots of (1.35).
By initially locating the zeros and poles on a graph with
an imaginary vertical axis and a real horizontal axis
Numerator(s)
C(s) = (1.37) (referred to as the complex splane), it is possible to
n (s + p n ) m (s + Pm jQ m ) establish some rules using (1.43), and (1.44) to sketch the
characteristic root excursions (referred to as root loci)
The natural modes of response of the closed loop system while varying K from zero to infinity.
are shown below,
For the Characteristic Equation,
pn t pmt
kne + k me sin(Q m t m ) (1.38)
n m
1 + KGH(s) = 0 (1.45)
G (s i )H(s i ) = 180 q360 (1.41) If all parameters are specified except the AVR gain Ka,
then a root locus plot for varying Ka would be
constructed as shown in Fig.1.11.
is a root of the characteristic equation. The transfer
function G(s) H(s) (commonly referred to as the "open D. Root Locus Example
loop" transfer function) can be written as, Consider the voltage regulator loop of the previous
example shown in Fig.1.9. A typical application for
K (s + Z i ) applying root locus techniques would be to study the
= 1 (1.42)
(s + P j ) effects of Ka on the roots of the characteristic equation.
 15 
IEEE Tutorial Course Power System Stabilization via Excitation Control June 2007
 16 
IEEE Tutorial Course Power System Stabilization via Excitation Control June 2007
 17 
IEEE Tutorial Course Power System Stabilization via Excitation Control June 2007
D K1
+ 2 H 2H (1.65)
1
 18 
IEEE Tutorial Course Power System Stabilization via Excitation Control June 2007
 19 
IEEE Tutorial Course Power System Stabilization via Excitation Control June 2007
EO
ET EHV
XT XS
Gen. Infinite
Bus
XE=XT+XS
Figure 2.1 SingleMachine Power System Configuration
A phasor diagram of the voltages on such a system is Figure 2.3 PowerAngle Curve
shown in Fig.2.2. These voltages are:
IV. TRANSIENT STABILITY
Eo infinite bus voltage Transient stability analysis is primarily concerned with
ET generator terminal voltage the immediate effects of transmission line disturbances on
Eq generator internal voltage behind transient reactance generator synchronism. Fig.2.4 illustrates the typical
Eq generator internal voltage behind synchronous behavior of a generator in response to a fault condition.
reactance
Starting from the initial operating condition (point 1), a
closein transmission line fault causes the generator
Eq electrical output power PE to be drastically reduced. The
resultant difference between electrical power and
mechanical turbine power causes the generator rotor to
Eq' accelerate with respect to the system, increasing the power
I.Xd
angle (point 2). When the fault is cleared, the electrical
SS ET I.Xd' power is restored to a level corresponding to the
appropriate point on the powerangle curve (point 3).
I.XE Clearing the fault necessarily removes one or more
EO
transmission elements from service and at least temporarily
I weakens the transmission system. For simplicity, this
Figure 2.2 Generator/ Infinite Bus Phasor Diagram
effect is not shown in Fig.2.4.
After clearing the fault, the electrical power out of the
The angle difference in the phasors indicates a real generator becomes greater than the turbine power. This
power flow from the generator into the infinite bus. causes the unit to decelerate (point 4), reducing the
In the analysis of classical steadystate stability, the momentum the rotor gained during the fault. If there is
voltage Eq and the power angle ss are used in the enough retarding torque after fault clearing to make up for
characteristic electrical power flow equation. However, for the acceleration during the fault, the generator will be
simple transient and dynamic stability analyses, it is more transiently stable on the first swing and will move back
appropriate to use the voltage E behind transient reactance toward its operating point in approximately 0.5 second
and the angle between this voltage and the infinite bus from the inception of the fault. If the retarding, torque is
voltage. In this way, the generator electrical power can be insufficient, the power angle will continue to increase until
expressed as: synchronism with the power system is lost.
 20 
IEEE Tutorial Course Power System Stabilization via Excitation Control June 2007
Power
add more parallel transmission lines in order to lower the
reactance between the generator and the load center. Such
PMAX 3 a solution may be quite costly as well as unfeasible to
implement. In the presence of a weak transmission system,
4
PDCEL control means, such as a power system stabilizer (PSS),
Stable acting through the voltage regulator, can provide
1 Turbine
PM significant stabilization of such oscillations if properly
Power
implemented.
PACCEL Unstable
VI. OSCILLATORY STABILITY
2 A. Characteristic Dynamic Equation
Following a system disturbance, whether it is a large
0o o
90o 180o disturbance or just a minimal load change on the system, a
Power Angle  generating unit will characteristically tend to oscillate
around its operating point until it again reaches steady
Figure 2.4 PowerAngle Curve Illustrating Transient Stability. state. The characteristics of these oscillations are
analogous to the motion of the spring mass systems.
Excitation system forcing during and following the fault For a synchronous machine under constant field
attempts to increase the electrical power output by raising excitation, an approximation of its dynamic motion is
the generator internal voltage Eq, thus increasing PMax. Fast obtained by relating the angular acceleration of the
and powerful excitation systems can improve transient generator rotor to the torques imposed on the rotor, in the
stability, although the effect is limited due mainly to the same manner that linear acceleration relates to force in a
large field inductance of the generator which prevents a spring mass system. This relationship for a synchronous
sudden change in Eq for a sudden increase in exciter machine is:
output voltage.
The steadystate stability refers to the ability of a power (Inertia) (Angular Acceleration) + (Damping Torque)
system to maintain synchronism at all points for + (Electrical Torque  Turbine Torque) = 0.
incremental slowmoving changes in power output of units
or power transmission facilities. Steadystate stability a For small changes, the behavior is described by a
small signal phenomenon is governed by the synchronizing characteristic equation having the same structure as that of
coefficient. Transient stability a large signal phenomenon the mechanical spring mass system. This is the "swing"
is also governed by the synchronizing coefficient. A fast equation:
acting, high gain AVR in general increases the
synchronizing coefficient but may decrease the damping 2H d 2 D d
coefficient. Thus a high gain AVR helps the steady state + + K1 = 0 (2.2)
s dt 2 s dt
and transient stabilities but may reduce the oscillatory
stability.
Where the parameters are defined as:
V. TRANSIENT VS. OSCILLATORY STABILITY (radians) rotor angle deviation from the
steady state operating point
In presentday systems, a machine being transiently
H (kW sec/kVA) inertia constant of the rotor of the
stable on the first swing does not guarantee that it will
generating unit (or group of units)
return to its steadystate operating point in a welldamped
D (p.u. power/p.u. freq. change) damping
manner and thus be stable in an oscillatory mode.
coefficient representing friction
Significant improvement in transient stability has been
and windage, prime mover and
achieved through very rapid fault detection and circuit
load damping, etc.
breaker operation. System effects such as sudden changes
s (radians/sec) synchronous frequency = 377 r/s
in load, short circuits, and transmission line switching not
on 60 Hz systems
only introduce transient disturbances on machines, but also
K1 (p.u. P/radian) synchronizing coefficient.
may give rise to less stable operating conditions. For
example, if a transmission line must be tripped due to a
Acting like the restoring force of the spring in the
fault, the resulting system may be much weaker than that
mechanical spring mass system, the term K1 called the
existing prior to the fault and oscillatory instability may
synchronizing power, acts to accelerate or decelerate the
result.
rotating inertia back toward the synchronous operating
One solution to improve the dynamic performance of
point. For small deviations from the operating point, the
this system and large scale systems in general could be to
synchronizing coefficient K1 is the slope of the transient
 21 
IEEE Tutorial Course Power System Stabilization via Excitation Control June 2007
power angle curve at the particular steady state operating B. Local Vs. Inter Area Oscillations
point, as shown in Fig. 2.5.
Power
Operating
If o is the steady state angle between Eo and E q , then
&
Point E'q EO
the slope is simply the derivative of the power angle PE = sin
function, or: X
' PE
PM
dP E q Eo
K1 E o = cos o (2.3)
d '
X d+ X e
Where:
' 0
E q is the internal voltage behind transient reactance in
0o 90o 180o
p.u., Power angle
Eo is the infinite bus voltage in p.u.,
'
Figure 2.5 Power Angle Curve Showing Derivation of Synchronizing
X d is the generator transient reactance in p.u., Coefficient K1.
Xe is the external reactance in p.u., and
is the angle between E and Eo. The frequency of the characteristic local mode is
generally in the 1 2 Hz range, depending mainly on the
Stronger transmission systems, with lower values of Xe, impedance of the transmission system. Stronger
have a larger value of K1 and thus provide more transmission systems generally have the higher local mode
synchronizing power to the generator. frequencies along with less of a tendency toward
The characteristic "swing" equation in (2.2) governs the spontaneous or undamped oscillations.
power system dynamic response with a damped oscillatory A second type of oscillations, known as "interarea"
behavior, having an oscillation frequency of approximately modes, is more complex because they usually involve
combinations of many machines on one part of a system
swinging against machines on another part of the system.
K1s
n radians / sec . (2.4)
2H C. Negative Damping Due To Voltage Regulator
Just as in a mechanical spring mass system, a power
The inherent modal frequency that is exhibited is system contains inherent damping effects that tend to damp
dependent mainly on such factors as unit inertia(s), out dynamic oscillations. Even when the proper conditions
machine and transmission system reactances, and load exist for dynamic instability (i.e., high network reactances,
level. line outages, high load levels, etc.), the natural damping of
There are two distinct types of dynamic oscillations that the system, represented by the positive D term in equation
have been known to present problems on power systems. (2.2), will prevent any sustained oscillations unless a
One type occurs when a generating unit (or group of units) source of negative damping is introduced.
at a station is swinging with respect to the rest of the It is generally recognized that the normal feedback
system. Such oscillations are called "local mode" control actions of voltage regulators and speed governors
oscillations. The oscillations are termed "local" because the on generating units have the potential of contributing
behavior is mainly localized at one plant, with the rest of negative damping which can cause undamped modes of
the system experiencing much less of the effects. dynamic oscillations. Direct evidence of this has been seen
Spontaneous local oscillations tend to occur when a very by the fact that sustained oscillations on power systems
weak transmission link exists between a machine and its have been stopped simply by switching voltage regulators
load center, such as for an isolated power station sending from automatic to manual control. However, removing
power across a single long transmission line. Such systems voltage regulators from service is not a realistic solution to
can usually be accurately modeled by a machine, single the problem, because the beneficial features of the voltage
transmission line, and infinite bus. regulator would be lost. The fortunate aspect of the
problem is that the same voltage regulator control that
causes negative damping can be supplied with
supplementary controls to contribute positive damping for
oscillatory stabilization.
The major function of the voltage regulator is to
continually adjust the generator excitation level in response
to changes in generator terminal voltage. The voltage
 22 
IEEE Tutorial Course Power System Stabilization via Excitation Control June 2007
 23 
IEEE Tutorial Course Power System Stabilization via Excitation Control June 2007
Positive Damping Negative Damping excitation system. The resulting signal is amplified to a
(Stable Region) (Unstable Region)
desired level and sent through a signal wash out module.
Resulting This module functions to continuously balance the
Synchronizing Axis Response
t
stabilizer output and thus prevent it from biasing the
generator voltage for prolonged frequency or power
excursions. The output limiter serves to prevent the
Et stabilizer output signal from causing excessive voltage
1
2
changes upon load rejection and to retain the beneficial
action of regulator forcing during severe system
D s E'q disturbances [22].
Damping Regulator
Axis Et Forcing Action
Term. Freq.
Signal
Shaft Speed Transducer
Washout
or Power
Figure 2.8 Phasor Relationship Of Signals And Torque Components; Unit
Under Voltage Regulator Control, Large Xe.
Output
Lead/Lag
Amplifier
Signal to Regulator
It should be made clear that the undamping component Network Limiter Sensing
Circuit
K2Eq is generally only large enough to cause growing
oscillations when the unit is under conditions that cause Figure 2.9 Major Elements Of Power System Stabilizer
large power angles, such as a weak transmission system,
large load, or low terminal voltage. Test experience has For proper damping action, PSS control settings
shown that a unit with a steady state power angle above 70 consisting of the lead, the lag, and the gain adjustments of
degrees (between the generator internal voltage and the the stabilizer have to be judiciously determined. Since the
infinite bus voltage) tends to have potential local mode dynamic response of a unit involves the machine and the
stability problems under voltage regulator control. external system, such settings may vary from unit to unit.
D. PSS For Improved Oscillatory Stability Also, particular PSS settings designed to suppress intertie
oscillations may not be effective in damping local
Since voltage regulator control can act to reduce the
machine/system oscillations. Therefore, setting procedures
damping of unit oscillations by sensing terminal voltage, it
for PSS's generally involve either a field test, a study of the
seems reasonable that a supplementary signal to the voltage
machine and system, or both.
regulator can increase damping by sensing some additional
A PSS setting procedure involving frequency response
measurable quantity. In doing so, not only can the
tests in the field has been widely accepted by utilities in the
undamping effect of voltage regulator control be cancelled,
Upper Midwest and Western United States for damping
but damping can be increased so as to allow operation even
inter area oscillations. From the measurement of terminal
beyond the steady state stability limit. This is the basic
voltage deviation in response to sinusoidal inputs to the
idea behind the power system stabilizer (PSS). The
voltage regulator reference, phase information is obtained
supplementary signal of a PSS may be derived from such
upon which PSS lead lag settings are determined. The PSS
quantities as changes in shaft speed (), generator
amplifier gain is empirically found. This is achieved by
electrical frequency (f)), or electrical power (PE).
monitoring the dynamic response of the unit with the PSS
There are a number of considerations in selecting the
in service and slowly increasing the PSS gain until small
right input quantity. The factors that play a role are
rapid oscillations appear, usually in the frequency range of
requisite gain and phase compensation, the susceptibility to
1 to 4 Hz. The PSS gain is then set to approximately one
other interactions such as torsional oscillations, and the
third this value.
noise level in transducers. The speed and frequency inputs
Since local mode stability problems often involve a
have been widely used. The trend is more towards PSS
single machine and the system, modeling procedures can
design based on integral of accelerating power. This type
be used as an aid in determining PSS settings. Such
of PSS provides satisfactory damping and reduced torsional
modeling techniques include representation of the PSS,
interactions. The basic theory of a PSS control based on
excitation system, generator, and external system. Both
integral of accelerating power input signal, its advantages
time response and frequency domain analysis can be used
and the methods of tuning the stabilizers are discussed by
in evaluating PSS settings.
Murdoch et al [21].
A block diagram showing the major elements of a
typical PSS is shown in Fig. 2.9. A special speed,
frequency, or power transducer converts the stabilizing
signal to a control voltage. The transducer output is then
phase shifted by an adjustable lead lag network, which acts
to compensate for time delays in the generator and
 24 
IEEE Tutorial Course Power System Stabilization via Excitation Control June 2007
Positive Damping Negative Damping generator as it behaves when connected through an external
(Stable Region) (Unstable Region)
reactance to an infinite bus. This model is used to illustrate
Resulting generator dynamic behavior as well as to analyze actual
Response t Synchronizing Axis
local mode dynamic stability problems. The model applies
to a DQ axis generator representation with a field circuit
PSS Forcing in the direct axis, but without subtransient amortisseur or
action
solid iron eddy current effects in either axis.
The effects of the voltage regulator can be included in
the model by adding a feedback branch between the output
4
3
ET and the input EFD. Similarly, a power system
Damping
Axis stabilizer branch can be included in Fig. 2.11 between the
PSS output signal and the summing junction of the voltage
regulator. A speedgovernor can also be represented as a
Figure 2.10 Phasor Relationship Of Signals; Unit Under Voltage
Regulator Control With PSS In Service. path between and the mechanical turbine torque Tm.
Kundur [22] shows the effect of a high gain voltage
A typical phase relationship of the signals associated regulator and the role played by a PSS.
with the PSS control sensing frequency or speed is shown Specifically the value of K5 has a significant bearing on
in Fig. 2.10. The quantity sensed by the PSS is phase the influence of the AVR on the damping of system
shifted by an amount 3 in the lead lag network. This PSS oscillations. Kundur [22] shows analytically as well as via
output signal is sent into the voltage regulator. The time examples that for high value of external system reactance
delays in the excitation system and generator result in an and high generator outputs, K5 is negative. In practice, the
actual forcing action that has a large damping component. situation where K5 is negative is commonly encountered.
By increasing the PSS amplifier gain, this damping action For such cases, a high response exciter is beneficial in
can be increased. However, higher order effects limit this increasing sysnchronizing torque. However, in so doing it
gain. The resultant dynamic forcing action due to the PSS introduces negative damping. In order to meet these
cancels the undamping effect of voltage feedback control conflicting requirements one then uses the high response
and extends the stability limit. exciter and a PSS.
Figure 2.11 Small Signal Model Of Generator, Transmission Line And Infinite Busbar
 25 
IEEE Tutorial Course Power System Stabilization via Excitation Control June 2007
 26 
IEEE Tutorial Course Power System Stabilization via Excitation Control June 2007
Therefore, a power system stabilizer applied to a single unit considered to have two components, viz. (3.2) that which is
can only contribute to the damping of an intertie mode in produced by the power system stabilizer solely by
proportion to the power generation capacity of the unit modulation of generator flux, Tep, and (3.2) that which
relative to the total capacity of the area of which it is a part. results from all other sources, including shaft motion, Teo
As a consequence, a stabilizer should be designed to The functional relationship between speed and torque is
provide adequate local mode damping under all operating shown for a stabilizer employing generator speed as an
conditions, with particular attention to conditions of heavy input signal. The contribution of torque due to the stabilizer
load and weak transmission, and simultaneously to provide path is given by:
a high contribution to damping of intertie modes. These
criteria ensure good performance for a wide range of power Tep
system contingencies. = PSS (s)GEP(s) = P(s) (3.1)
G
C. Tuning Concepts where GEP(s) = the plant through which the stabilizer
Stabilizers must be tuned to provide the desired system must operate. (Generator, Exciter, and
performance under the condition which requires Power System)
stabilization, typically weak systems with heavy power = Tep / E pss (3.2)
transfer, while at the same time being robust in that PSS(s) = speedinput stabilizer
undesirable interactions are avoided for all system = pss / G (3.3)
conditions.
To provide damping, the stabilizer must produce a Tep = component of electrical torque due
component of electrical torque on the rotor that is in phase solely to stabilizer path
with speed variations. The implementation details differ, Epss = stabilizer output signal
depending upon the stabilizer input signal employed.
However, for any input signal the transfer function of the The plant GEP(s) has the highest gain and greatest phase
stabilizer must compensate for the gain and phase lag under conditions of full load on the unit, hence a target
characteristics of the excitation system, the generator, and value of base load for the tuning process. The tuning will
the power system [31] to [33]. These elements collectively also be affected by the transmission system. Expected
determine the transfer function from the stabilizer output values from short circuit studies give starting point and
(Epss) to the component of electrical torque, which can be various contingency cases will define likely weaker
modulated via excitation control (Tep). It must be noted that transmission cases. A wide range should be covered to
Tep is only a portion of the total torque acting on the rotor, insure robustness in the tuning. The PSS may not be critical
and that it cannot be measured directly. Since it is related to for normal operation but critical if the system become
flux changes in the machine, it can be measured indirectly weaker. Choosing high AVR gains may limit operation
via terminal voltage, as will be discussed in later sections. without PSS for maintaining small signal stability margins.
The transfer function relating Tep to Epss, denoted here The de Mello Concordia model [23] in Fig 3.2 is similar
as GEP(s), is strongly influenced by voltage regulator gain, to the basic block diagram in Fig 3.1 and allows us to
generator power level, and ac system strength. illustrate the process in conceptual terms. The top part of
this figure shows the relationship between the accelerating
torque and the rotor speed and angle. These are the basic
electromechanical equations for the generator. The terms
"D" and "K1" represent the equivalent damping and
synchronizing torque coefficients. The lower part of this
figure shows the effects of the AVR and the flux dynamics
in the generator. The effect of the AVR is to increase the
effective "K1" term, but unfortunately at the same time this
reduces the damping term. To maintain or increase the
dynamic stability margin it is necessary to increase the
effective "D" term in the model. If we consider the addition
of the PSS loop which acts through the AVR and generator
field circuit, what we would like the PSS control loop to do
Figure 3.1 Stabilizer With Speed Input System Block Diagram is to provide changes in torque which are exactly in phase
with changes in rotor speed.
The block diagram in Fig. 3.1 illustrates, in terms of a
few basic smallsignal transfer functions, the relationship
between the applied torques on the turbinegenerator shaft
and the resulting generator rotor speed, G and rotor
angular displacement, . The electrical torque may be
 27 
IEEE Tutorial Course Power System Stabilization via Excitation Control June 2007
 28 
IEEE Tutorial Course Power System Stabilization via Excitation Control June 2007
F. Power Input Stabilizers looking out from the unit into the system. These are given
Other designs are based on electrical power by itself, or as a total of the generator step up transformer and the
accelerating power based on measurements of electrical Thevenin equivalent looking out from the HV switchyard.
power and speed or frequency. The next chapter has detail To consider the robustness of the parameter choices, a
of the current accelerating power PSS design. The interest range of impedances is usually chosen to cover normal and
in using accelerating power as a stabilizer input signal contingency cases. In this example 15, 30 and 45% on the
results from the inherently low level of torsional interaction unit base were used.
[33]. It must be noted that a practical powerbased PSS Tuning Studies for IEEE Tutorial
P = 0.9 pu Q = 0 pu Xtot = 15/30/45%
stabilizer must utilize some form of compensation for 40
Uncompensated (Te/Etref)+(EXCSIG/Speed)
Gain(dB)
powerbased stabilizer design to a speedinput stabilizer. 20
Phase(Degree)
The integral of accelerating power design results in 0
canceling the phase lead from power and the phase 100
Frequency(Hz)
1
10
2
10
 29 
IEEE Tutorial Course Power System Stabilization via Excitation Control June 2007
required. The torsional filters, if required, for the integral in the input signals on PSS output. The discussion on the
of accelerating power PSS are relatively modest so usually deMello Concordia block diagram illustrated how we can
only one iteration in the tuning process is required. view these components are resolved into damping and
A point to be made here is the slight difference in the synchronizing torques. The choice of leads of 0.2 and 0.18
transfer function of uncompensated transfer function, and seconds, and lags of 0.035 and 0.04 seconds are shown in
the AVR closed loop, which can be readily measured. This Fig 3.5 where the plot also reflects the choice of washout
point was mentioned in [31] to [33] and the previous time constant of 5 seconds. Compensated phase is within
tutorial material. Note the plot on Fig 3.4 compares the two 30 degrees of target (zero degrees) for most of the range of
transfer functions and we can see that they agree closely interest, and no more than 45 degrees of lag at 3 Hz, which
near local mode but differ by as much as 30 degrees in the is above the local mode frequency (about 1.6 Hz).
intertie region. In that case if one would assume the AVR
PSS Tuning Studies for IEEE Tutorial
closed loop was the target value, the performance goal in P = 0.9 pu Q = 0 pu Average Transfer Function for Xtot = 15/30/45%
Compensated (T /E )+(EXCSIG/Speed) PSS Lead/Lag : 0.18,0.2,0/0.035,0.04,0
the intertie region would lead towards overcompensation
e tref
150
one used for tuning. That validates the tuning process and 0
allows confidence in the transfer function that was used in
Phase (degrees)
tuning, but which cannot be easily measured. 50
20 Uncompensated Phase
PSS Lead/Lag
300
40 2 1 0 1 2
10 10 10 10 10
Frequency (Hz)
60
200
Vt / Vtref Fig. 3.5. Uncompensated, Compensated and PSS Phase Plot
Uncompensated Te / Speed
100
0
Choice in the phase compensation for adding more lead
lag stages or changing washout time constant can be made
Phase(Degree)
100
at this point and the effects on phase can be considered. At
200
this point the high frequency gain of the PSS can also be
300
10
2
10
1
10
0
10
1 2
10
considered. If we are concerned about the application of
Frequency(Hz) torsional filters, the characteristics of the leadlag stages in
Fig. 3.4. Comparison of Uncompensated and AVR Closed Loop this frequency range are considered. Generally speaking
the torsional frequencies are high enough compared to the
In this example, two PSS leadlag stages are used to local mode that the asymptote in gain which is the product
compensate the uncompensated transfer function. Most of all the lead time constants divided by the product of all
stabilizers offer three leadlag stages for phase the lag time constants can be used. In this case the high
compensation (reference the new PSS2B model in 421.5 frequency gain is (0.2*0.18)/(0.035*0.04) = 26.5. If two
2005), and some manufacturers offer even more than three designs are otherwise roughly the same it may be wise to
stages in their designs. The use of additional leadlag choose the one with lower high frequency gain. A further
stages gives the ability to better shape the gain and phase discussion of the PSStorsional interaction is in a later part
characteristics over a wider range of frequencies to meet of this section.
performance targets. Older equipment may be limited to
B. Root Locus
two stages for choosing compensation.
The next step in the process is to consider the root locus
A. Phase Compensation as PSS gain is varied. As shown in Appendix A of [33], the
The phase compensation is chosen to be close to zero initial direction of local mode eigenvalue migration as the
phase for the range of 0.1 to 3.0Hz. If we leave the phase stabilizer gain is increased from zero is determined by the
slightly undercompensated we have both significant compensated transfer function phase at the local mode
positive damping torque contribution, and also positive frequency. For perfect compensation, i.e., L , = 0 pure
synchronizing torque contribution. The phase at low positive damping will be obtained and the eigenvalue will
frequency is affected by choice of the washout (high pass move directly into the left half plane with no change in
filter) that is applied to eliminate the steadystate changes frequency. If phase lag exists, the frequency will increase
 30 
IEEE Tutorial Course Power System Stabilization via Excitation Control June 2007
in proportion to the amount of damping increase, factor between setting and instability point). Here we see
specifically the gain of 80 on the PSS will result in instability at about
5.5Hz, well above (18dB) the recommended PSS gain of
L = tan L L (3.4) 10.
If frequency domain tools are not available, the process
where L = local mode frequency (rad/sec) of PSS tuning is somewhat less rigorous but the damping
and oscillation frequency of local mode can be determined
L = local mode decay rate (sec1)
from time simulations to get equivalent information as
= implies change due to stabilizer contained in Figs 3.6 and 3.7. Also, that same information
For L = 45 frequency will increase at the same rate as can be validated during testing in similar fashion. It is also
damping. For L = 90 a restructuring of (3.4) will show possible in any system to measure the PSS open loop
that no change in damping will take place, but frequency transfer function (from AVR input to PSS output with PSS
will increase. This basic concept is very useful in not connected) and we have found that the measured gain
understanding the root locus. at the crossover point correlates well with the inverse of the
Root locus plots are shown in Fig. 3.6 for the three instability gain. The crossover point also provides the
different impedances. These plots represent the migration instability frequency.
of the local mode eigenvalues as stabilizer gain is increased PSS Tuning Studies for IEEE Tutorial
from zero towards infinity. P = 0.9 pu Q = 0 pu Xtot = 15/30/45%
Control Mode Locus as a Function of PSS Gain PSS Lead/Lag : 0.18,0.2,0/0.035,0.04,0
8
PSS Tuning Studies for IEEE Tutorial
P = 0.9 pu Q = 0 pu Xtot = 15/30/45%
Local Mode Locus as a Function of PSS Gain PSS Lead/Lag : 0.18,0.2,0/0.035,0.04,0 7
2
7
6
6
1.8 * Kpss 7
567
5 4 3 1 0
6 2 5
5 6
Frequency (Hz)
1 Xtot = 15% 1.6 e
2 10 od 4
lM 5
ro
* Kpss 3 20 nt 4
6 5 4 3 1.4 Co 4
1 0 2 1 Xtot = 30% 4 40 4
7
2 2 3
Frequency (Hz)
6 5 4 3 1.2 5 60
8 7 2 1 Xtot = 45%
3 4 3 3
8 6 80 3
4 6 1 7 100 2
5 8 2 2 Local Mode 2
6 10 0.8 2 1
2 1
7 15 3 3 3 2 1
0.6 1
4 44
8 20 5655
6
54
7
6657
7 3433222
45
6 7776
2334456 5767
12314
0.4 7
1 13
4
51 6
6
2 4
5
3
7
21 11 232
45
6
7
3
245
6
7456
3 7 1
61
3
4
5
7
2 3
4
5
6
71
2 3
4
5
6
7
2 72
63
5
71
54
6
2
3
5
7
6
4
7
6
7
5
6
4
4
325
6
73
3
4
1
5
6
7
21
4
5
6
7
2 0
25 20 15 10 5 0 5
0.2 Sigma (1/Seconds)
123456781234567
8 12345678
1
3
4
2
5
7
8
6 1
2
3
4
5
7
8
6 1
2
3
4
5
7
8
6 878
54
7
65
3
44
5
2
28
67
3
13
62
4
5
7
8 7
6 8
4
5
3
1
2
6 0
5 4.5 4 3.5 3 2.5 2 1.5 1 0.5 0 0.5 Fig. 3.7. Control Mode Root Locus as PSS Gain is Increased
Sigma (1/Seconds)
Fig. 3.6. Root Locus of Local Mode for Different System Strengths Some newer digital based controls have built in test
functions that allow for these frequency responses as well
Although many eigenvalues exist for the total system as step tests to be performed and have tools that plot
only the dominant ones associated with the local mode are frequency responses directly.
shown in this figure. We can see that local mode frequency C. Step Test and Fault Simulations
is increased as the system tie becomes stronger, and
After choosing phase compensation and gain, time
significant damping is introduced by the PSS without much
simulations are run to verify performance with voltage
frequency change. In this case the frequency increases
steps and fault disturbances. Voltage step tests will be used
slightly as PSS gain is increased, illustrating the comments
later for performance verification. Cases are normally run
made previously about positive synchronizing torque
for all of the system reactances to check the robustness of
contribution. Weaker systems have inherently lower
the design.
damping without PSS, but in all cases the PSS provides
For this example we show the 2% voltage step response
significant improvement in damping.
in Fig 3.8 for the strong system connection of X=15%. As
The other aspect of the PSS design comes into the PSS
this is a single machine system the response is local mode
control mode that is becoming dynamically less stable as
and the oscillation without PSS can be seen to be at about
the local mode dynamic stability is increasing. The plot in
1.6Hz which correlates with the root locus plot in Fig 3.6.
Fig 3.7 shows the control mode and also the local mode
The step is applied for 5 seconds and then removed. The
root locus on the same plot. Here we focus on ensuring the
oscillation in MW and speed clearly show the response and
chosen gain allows for adequate gain margin. From a point
the damping as provided by the stabilizer.
of view of control system design we need to ensure a
minimum of 6dB and typically better 10dB margin (a 3:1
 31 
IEEE Tutorial Course Power System Stabilization via Excitation Control June 2007
PSS Tuning Studies for IEEE Tutorial PSS Tuning Studies for IEEE Tutorial
P = 0.9 pu Q = 0 pu Xtot = 15% PSS Lead/Lag : 0.18,0.2,0/0.035,0.04,0 P = 0.9 pu Q = 0 pu Xtot(prefault) = 15% Xtot(postfault) = 20% PSS Lead/Lag : 0.18,0.2,0/0.035,0.04,0
Response to a 2% Step in Terminal Voltage Reference Blue : PSS ON Red : PSS OFF Response to a 0.1 Second Fault MidWay to Infinite Bus Followed by the Loss of Lines Blue : PSS ON Red : PSS OFF
1.04 0.2 1.5 2
1.03 0.15 1
1
0
Qe(pu)
1.02 0.1
Qe(pu)
Vt(pu)
Vt(pu)
1.01 0.05 1
0.5
1 0 2
0.99 0.05 0 3
3.5 50.04 15 51
3 50.5
50.02
10
Speed(Hz)
Speed(Hz)
Efd(pu)
2.5
Efd(pu)
50 50
2
5
49.98 49.5
1.5
1 49.96 0 49
0.04
0.92 2
PSS Output(pu)
0.005
PSS Output(pu)
0.02
Pe(pu)
0.91
Pe(pu)
0 1 0
0.9 0.02
0
0.005
0.89 0.04
1 0.06
0.88 0.01 0 2 4 6 8 10 0 2 4 6 8 10
0 2 4 6 8 10 0 2 4 6 8 10
Time (Seconds) Time (Seconds)
Time (Seconds) Time (Seconds)
Fig. 3.8 2% Voltage Step Simulation with X=15%  with & without PSS Fig. 3.9 Three Phase Fault Simulation with X=15%  with & without PSS
The fault response in Fig 3.9 shows the same local mode For each mode 40 cases are studied, for a range of
response following the fault and subsequent line clearing. system impedances and MW and MVAR loading as shown
This case illustrates the principle that the PSS cannot in the bottom of the figure as bar graphs.
provide effective damping while the PSS output is in limit. 0.35
TIV/Sigma for Torsional Mode #1 Less Than 0.1 > No Interaction Larger Than 0.1 > Interaction
0
During the first two cycles of the local mode swing the PSS Threshold
Without Filters
is alternating between positive and negative limits, and 0.3 With Filter
when the PSS comes out of limit the damping reduces the 0.25
0.15
D. PSS Torsional Interaction
For steam turbines with lightly damped low frequency 0.1
the PSS. In Fig 3.10 is a plot of the TIVs for PSS applied Fig. 3.10. TIV with and without Filters for Mode 1 Response
to a nuclear plant for mode 1 (near 10Hz), where torsional
filters were applied. As part of the dynamic setup, more Note that the interaction increases for stronger systems
detailed models of the generator and excitation system are and for operation under excited with leading power factor.
used to accurately represent their behavior at higher The TIV calculations are done for all subsynchronous
frequencies. The torsional dynamics are included in the modes and typically only the lower frequency modes show
models using modal form equations for all the sub interaction with the PSS. For any modes showing
synchronous torsional modes. Test results have validated interaction above the 10% level in TIV/Sigma, filters are
models at dozens of machines studied specified. Any band reject filter design can be used, and a
common form is the biquadratic form, as follows:
n 2 + 2 N n s + s 2 (3.5)
n 2 + 2 D n s + s 2
In some cases more than two filters are applied to a
torsional mode and stagger tuned to give a wider
bandwidth. This is common with mode 1 as the modal
frequency varies slightly as the unit operating point
changes.
 32 
IEEE Tutorial Course Power System Stabilization via Excitation Control June 2007
Frequency (Hz)
3 4
we could also retain the washout time constant at 2.0 Sigma (1/Seconds)
seconds, to effectively decouple the low frequency power Fig. 3.12. Local Mode Root Locus for Aero Turbine Example
oscillations from affecting the PSS. Figs 3.11 and 3.12
show the plots of phase compensation and local mode root X=0.5 X=0.6
locus. 1 3
F. Interarea Mode Damping Pg=0.9 X=1.2 Pg=0.4
0
always be acting correctly within its limits to aid damping
of the resulting oscillations. In a multimachine
50 environment, however, more than the first swing may be
100
critical, and the nonlinear performance of the stabilizer
becomes important. This will be apparent from the large
150 Compensated Phase signal performance analysis that follows:
Uncompensated Phase
PSS Lead/Lag The response of the fourmachine system to a three
200
10
2
10
1
10
0
Frequency (Hz)
10
1 2
10 phase fault on the right side bus (see Fig 3.13) is plotted in
Fig. 3.14 for the case of no stabilizer on any machines as
Fig. 3.11. Phase Compensation for Aero Turbine Example compared to all machines having stabilizers. The plot of the
speed signals from the simulation clearly shows the
improvement that can be obtained in both local mode and
interarea mode damping. The plot in Fig 3.15 shows the
root locus that has the local modes and interarea mode
shown graphically in frequency domain.
 33 
IEEE Tutorial Course Power System Stabilization via Excitation Control June 2007
PSS Tuning Studies for IEEE Tutorial areas. Stabilizer performance must therefore be measured
Machine Speed Response to a 4Cycle Fault on Bus 4 4Machine System
Red: All Machines Without PSS Blue: All Machines With PSS in terms of enhancing damping under these weak system
conditions. This measure must include not only the small
61 61
3 (Hz)
60 60
following large disturbances, when all modes of the system
59.5 59.5 are excited simultaneously. Based upon this measure, it is
shown that the most appropriate stabilizer tuning criteria is
59 59
0 5 10 15 20 0 5 10 15 20
to provide good damping to local modes of oscillation and
61 61
also the interarea modes of oscillation.
60.5 60.5 The first step before beginning the PSS tuning process is
to determine the settings for the AVR control based on the
2 (Hz)
4 (Hz)
60 60
equipment and the grid interconnection requirement that
59.5 59.5 might exist. Once the AVR tuning is determined, we use
phase compensation, root locus, and time domain analyses
to tune and evaluate the performance of the PSS. Phase
59 59
0 5 10 15 20 0 5 10 15 20
Time (Seconds) Time (Seconds)
Fig. 3.14. Plot of Speed Signals from 4 Machine System compensation lets us select the PSS lead/lag settings to
compensate for phase lags introduced by the generator,
The above observations are intended to illustrate a excitation system, and power system. Root locus analysis is
general relationship between smallsignal damping and used to select the PSS gain, and to determine the PSS
largedisturbance performance, rather than a comparison of instability gain and ensure adequate gain margin in the PSS
performance between particular input signals. Most new control loop. Finally, time domain analysis demonstrates
designs use the integral of accelerating power type PSS but the performance of the PSS in the power system
many existing plants have speed, frequency and power environment. The choice of washout time constants and
input PSS designs. Stabilizer output limits and dynamic the use of additional leadlag stages were illustrated by
limiting circuits (third stage switchable washout) [39] also example. The need to insure adequate margins for low
can be part of the tuning process. In the interest of space frequency torsional oscillations that might interact with the
these concepts will remain documented in the literature. PSS was also mentioned as impacts the tuning process.
The tuning concepts and performance criteria developed
PSS Tuning Studies for IEEE Tutorial
4Machine Study System
in this chapter, including the relationship of performance to
Root Locus as a Function of PSS Gain PSS Gain on All Machine Varied Simultaneously
phase compensation characteristics, provides the basis for
2.5
field tuning procedures. The references [44], [45], [46],
[47], [48], [49], [50] and [51] are to be recommended for
6 5 4 3 2 1
2 study and review. Further background and practical
* Kpss
application information is in Sections 4, 5, and 6 of this
1 0 tutorial which are replete with practical experience.
Frequency (Hz)
1.5
2 10 6 5 4 3 2 1
3 20
4 30
5 40 1
6 50
6
5
4
6 5 3
4 2 1
3 0.5
2
1
1 2 3 4
5 1
654
3 26
123456 12
3
4
56
54
63
2
1
4
5
3
6
2
5
4
3
6 23456 0
5 4.5 4 3.5 3 2.5 2 1.5 1 0.5 0 0.5
Sigma (1/Seconds)
 34 
IEEE Tutorial Course Power System Stabilization via Excitation Control June 2007
 35 
IEEE Tutorial Course Power System Stabilization via Excitation Control June 2007
Output
M
1 + s Tw1 1 + s Tw2 1 + s T6 M 1 + s T2 1 + s T4
A +
(1+s T9)

Vstmin
Ks3
HighPass Filters
 36 
IEEE Tutorial Course Power System Stabilization via Excitation Control June 2007
The earliest systems combined an electrical power with the selection of phase compensation, gain and output
measurement with a derived mechanical power limit settings and their effect on the overall performance of
measurement to produce the required quantity. On the PSS. This will not be repeated here. Instead, this
hydroelectric units this involved processing a gate position section will focus on the derivation of the accelerating
measurement through a simulator that represented turbine power signal and its use in deriving an equivalent speed
and water column dynamics [54]. For thermal units a signal. Specifically, this section will describe the impact of
complex system that measured the contribution of the speed measurement issues and mechanical power variations
various turbine sections was necessary [58]. on the operation of units equipped with this style of PSS
Due to the complexity of the design, and the need for and how this has influenced the design of PSS2A
customization at each location, a new method of indirectly stabilizers.
deriving the accelerating power was developed. The With a large base of installed units, and long history of
operation of this design of stabilizer is described in usage, experience has been acquired with many different
references [56], [57]. The IEEE standard PSS2A model vintages of hardware. Early designs suffered from failures
used to represent this design is shown as Fig. 4.1 [36]. due to mechanical components such as speed pickups.
The principle of this stabilizer is illustrated by rewriting Replacement of the measured speed signal with a derived
equation (4.1) in terms of the integral of power. frequency signal has greatly improved reliability at many
facilities. The early analogelectronic designs also suffered
1 from reliability problems due to failures of components
= ( Pm Pe ) t (4.2) used to implement the adjustable settings (e.g. switches,
2H potentiometers). Digital designs have eliminated these
components and improved reliability and ease of use.
The integral of mechanical power is related to shaft
Further gains in reliability are achieved when the PSS is
speed and electrical power as follows:
implemented as additional software code in a complete
digital excitation system, since this eliminates any
Pmt = 2H + Pet (4.3) additional hardware.
A. Signal Mixing
The P stabilizer makes use of the above relationship Referring to the block diagram of Fig.4.1, the two input
to simulate a signal proportional to the integral of signals to the P stabilizer are speed (A) and active
mechanical power change by adding signals proportional to power (B). Although the P design has many advantages
shaftspeed change and integral of electrical power change. over stabilizers that employ only one of these inputs it is
On horizontalshaft units, this signal will contain torsional sensitive to the relationship between these two inputs. For
oscillations unless a filter is used. Because mechanical optimum performance it is critical that the two signal paths
power changes are relatively slow, the derived integral of (AC and BF) are matched in terms of gain and filter time
mechanical power signal can be conditioned with a low constants.
pass filter to attenuate torsional frequencies. The power path employs two highpass filter stages and
The overall transfer function for deriving the integralof an integration to derive the integralofelectrical power
accelerating power signal from shaft speed and electrical
change signal, Pe:
power measurements is given by:
2
P sT 1
P Pe (s) Pe (s) 2He 1 + sTW W s2H Pe
a t + G(s) + (s) (4.4)
2H 2Hs
2Hs (4.5)
sTW3 K S2
Pe
where G(s) is the transfer function of the lowpass filter. 1 + sTW3 1 + sT7
The major advantage of a P stabilizer is that there is a The second part of Equation 5 is based on the notation of
greatly reduced requirement for a torsional filters with this Fig.4.1 and the following settings:
design. This alleviates the exciter mode stability problem, TW3 = T7 = TW
thereby permitting a higher stabilizer gain that results in TW4 = 0 (i.e. this block is bypassed)
better damping of system oscillations. A conventional end KS2 = TW / (2H)
ofshaft speed measurement or compensated frequency KS3 = 1
signal can be used with this design. In order for the speed signal path to match the power
path it must employ two stages of highpass filtering as
III. PRACTICAL APPLICATION ISSUES well, and its equivalent filter time constant must be kept as
Many excellent papers have been written dealing with small as possible:
the tuning of PSS [31], [32], [33], [50]. These authors dealt TW1 = TW2 = TW
 37 
IEEE Tutorial Course Power System Stabilization via Excitation Control June 2007
T6 0 not the case and a large error signal can propagate to the
PSS output, thereby changing terminal voltage and reactive
With these settings the signal appearing at point D is power on the unit. This problem forced the selection of
proportional to changes in the integralofmechanical low PSS gains or output limits, severely limiting the
power, Pm. When recombined with the Pe signal at effectiveness of the PSS.
point G, the integralofaccelerating power, Pa, is formed. The transfer function between the power input, PE, and
This signal is then treated as equivalent speed and the the integralofaccelerating power signal, PA, (points B and
phase lead blocks that follow are set to compensate in order G in Fig.4.1) may be written as follows:
to maximize the contribution of the stabilizer to damping
torque. PA (s) sTW3 K S2
= ( G(s) 1) (4.6)
B. Mechanical Power Variations PE (s) 1 + sTW3 1 + sT7
Although the original requirement for the PSS units was
based on a need to provide damping for the local plant The original design of mechanical power lowpass filter
modes of oscillation, many new installations and retrofits consisted of a simple multipole filter of the form:
have been applied to improve damping of interarea modes
of oscillation [50] as is common in western U.S. utilities. 1
In order to be effective at damping these modes of G(s) = (4.7)
oscillation, the highpass filters, parameters Tw1 to Tw4 in
(1 + sT9 ) M
Fig.4.1, must be set to admit frequencies as low as 0.1 Hz
without significant attenuation or the addition of excessive which is achieved in the model by setting the following
phase lead. values:
Early attempts at retuning PSS for these frequencies
identified some side effects related to mechanical power T8 = 0
variations on the units. Tests on the original P design N=1
on thermal units included fast intercept valve closures that
produced a step change in power of approximately 5%, The filter order, M, and time constant, T9, can be
followed by a ramp of 0.55%/s [55]. The maximum selected to provide adequate attenuation of the lowest
generator terminal voltage change produced by a PSS torsional frequency for horizontalshaft applications.
configured with short washout time constants was below Researchers [60] discovered that they could reduce the
2%, for the normal inservice gain. On the first tests of this sensitivity to mechanical power variations by redesigning
design on hydraulic units, mechanical power ramprates in the mechanical power lowpass filter to utilize a transfer
excess of 10%/s were achieved under gate limit control. function of the form:
The introduction of long highpass filter time constants
M
produced excessive terminal voltage and reactive power 2
deviations. In response to this problem, researchers 1+ s
o
identified the root cause of the variations and modified the G(s) = 2 (4.8)
designs accordingly. s 2
When mechanical power is changed rapidly, electrical 2 + s+1
o o
power follows quickly but there is a limited change in the
rotor speed. Although this depends on the strength of the Further analysis and tests on actual hardware
system interconnection, the speed changes will always be implementations confirmed that the complexpole
relatively small and are considered to be negligible in the implementation was not optimal and that the following
following analysis. transfer function could be used to reduce mechanical power
Referring to Fig.4.1, when electrical power (B) is effects on the PSS output.
ramped, the integralofelectrical power signal (F) will
change with a rate and magnitude determined by the N
selected washout time constants and unit inertia. From this (1 + sT8 )
point forward, the signal follows two paths to the output.
G(s) = M
(4.9)
The lower path is a direct connection to the derivation of
(1 + sT9 )
the equivalent speed signal at point G. The signal
produced at point F also travels through the mechanical The filter of equation (4.9) is frequently identified as a
power lowpass filter (E) before appearing at the output. "ramptracking filter based on its properties when the
Ideally these signals would exactly cancel each other, since coefficients, T8, T9, M and N are selected correctly.
the PSS was not intended to produce an output for this The criteria used to analyze the merits of different
condition. With long washouts and high ramp rates, this is mechanical power filter designs are the following:
Attenuate highfrequency components in the input
 38 
IEEE Tutorial Course Power System Stabilization via Excitation Control June 2007
Active Power
this purpose, the integralof mechanical power changes are 0.8
(pu)
characterized as combinations of the following time 0.6
domain inputs: 0.4
step, A*u(t) 0.2
ramp, B*t
0
parabola, C*t2 0 5 10 15 20
ramptracking
where t is time in units of seconds and A, B and C are 0.20 lowpass
the magnitudes of the associated components in per unit. PSS Output 0.15
The steadystate PA signal for each of these inputs can be
calculated using the final value theorem by evaluating the 0.10
(pu)
following: 0.05
0
lim t p A (t) = lims0 (s * Input *(G(s) 1)) (4.10)
0.05
0 5 10 15 20
Appendix A provides details of the evaluation of
Time (seconds
equation (4.10) for a conventional lowpass filter (equation
Figure 4.2 Simulated Ramp Response
4.7) and the ramptracking filter (equation 4.9). The result
for each type of input is summarized in Table 4.1.
Different coefficients and time constants can be used to
TABLE 4.1: improve the tracking of power ramps or to provide greater
STEADY STATE RESPONSE TO POWER VARIATIONS attenuation of lowfrequency torsional components.
Input SteadyState Output Increasing the denominator order or the denominator time
LowPass RampTracking constant is a viable alternative to introducing notch filters
step input 0 0 at torsional frequencies since it does not interfere with the
ramp input B*M*T9 0 selected phase compensation of the resulting accelerating
parabolic input infinite C*F(M,T9) power signal. This will increase the sensitivity of the
stabilizer to power changes however this is normally
The key result in this table is that the ramptracking filter acceptable on large horizontal shaft units with their slow
produces a zero steadystate output for a ramp input and a loading rates.
bounded output for a parabolic input. This is only true if The performance of this filter may also be critical to the
the coefficients are selected to satisfy behaviour of the unit, in the event of inadvertent islanded
operation resulting in large frequency and mechanical
T8 = M *T9 (4.11) power variations.
C. Input Signals
The derivation of the results provided in Table 1, Electrical power is readily available as an input. In
including the relationship of equation (4.11) is included as analog implementations it can be measured using a three
Appendix A. phase Halleffect watt transducer or equivalent device that
The most commonly used ramptracking filter produces an instantaneous output proportional to the
coefficients are N=1 and M=5 since this provides four net generator active power. Selective filtering is required to
poles with the minimum number of numerator and remove the characteristic harmonics present in the output
 39 
IEEE Tutorial Course Power System Stabilization via Excitation Control June 2007
 40 
IEEE Tutorial Course Power System Stabilization via Excitation Control June 2007
quadrature reactance that applies to the frequency range of operating conditions (Fig 4.6)
interest. For roundrotor machines this normally requires
an impedance value close to the transient quadrature 120
reactance. stabilizer phase compensation
Each generator will be somewhat different, and the closedloop exciter phase lag
100
compensating reactance should be selected based on
Phase (degrees)
knowledge of the machine reactances and time constants.
80
IV. HARDWARE CONSIDERATIONS washout&
laglead selection
The hardware should be designed so as to allow setting 60
of the PSS parameters over a sufficiently wide range. The
design should also ensure a high degree of functional 40
reliability and allow sufficient flexibility for maintenance. leadlag
selection
These requirements are often overlooked, resulting in 20
unreliable and unsatisfactory performance of the PSS,
much to the frustration of operators. There have been
many instances of operators turning off the PSS because of 0
0.1 0.2 0.5 1 2 5
poor performance resulting from inadequate hardware
design and improper selection of control parameters. Frequency (Hz)
The requirement for high reliability and maintainability
Figure 4.4 ClosedLoop Exciter Phase Compensation
of PSS and other elements of the excitation system may be
in part satisfied by component redundancy. Duplicate
voltage regulators and PSS [30], [57] have been used on
Active Power
1.00
critical generating units. One voltage regulator with its
(pu)
0.95
PSS would be in service at any one time with the other
tracking it. In the event of a PSS malfunction, various 0.90
protective features would initiate transfer to the alternate 1.04
regulator and PSS. In addition to improving the detection
Terminal V
1.03
PSS ON
consequences of such failures. The improved reliability PSS OFF
and reduced parts count of newer digital exciters, with 1.02
builtin PSS, have mitigated the need for such complex 0.0010
delta speed
systems. 0.0005
(pu)
V. PSS COMMISSIONING & FIELD VERIFICATION As noted in the previous section, the tests usually consist
During field commissioning, the actual response of the of injecting small step changes to the voltage regulator
generating unit with the PSS is measured and used to verify terminal voltage reference and monitoring a number of
some of the analytical results. Typical tests performed generator variables. If there are discrepancies between
during commissioning include: computed and measured responses, the models are
measurement of the online closedloop excitation appropriately modified; if necessary, revised PSS settings
system phase compensation requirements (Fig. 4.4) are determined and implemented. This "closed loop" design
step response tests to measure damping and commissioning process is very effective [61].
improvement at local mode frequencies (Fig. 4.5) Initially, the PSS gain should be increased slowly, with
loadramping tests to ensure that the PSS does not transient testing at each setting. To insure sufficient
produce undesirable modulation of the units stability margin, a good practice is to check the
terminal voltage under normal or emergency performance of the PSS with the gain increased up to twice
 41 
IEEE Tutorial Course Power System Stabilization via Excitation Control June 2007
0.15
0.10
(pu)
0.05
1
0
0.05
G(s) 1 = M
1
0.03
a (sT )i 9
i
PSS Output
0.02
i=0
(pu)
0.01
M (A5.4)
sT9 a i (sT9 )i 1
0
0.01
1.0 = i =1
Active Power
M
1 + a i (sT9 )i
0.8
(pu)
0.6
0.4 i =1
0.2
where the fact that a0=1 has been used to reduce the
0.0006
numerator and expand the denominator.
delta speed
0.0002
(pu)
0.0002
0.0006 Step input: U(s) = A/s
0.0010
450
M
i 1
A 9 a i (sT9 )
300
150 sT
(Vdc)
Field
0
150
lim y(t) = lim s i =1
M
s 1 + a (sT )i (A4.5)
300
t s0
filter mech power
0.15 i 9
0
0.15
i =1
(pu)
0.30
0.45 =0
0.60
0 5 10 15 20
Time (seconds) Ramp input: U(s)=B/s2
Figure 4.6 Fast Load Ramp
M
i 1
VI. APPENDIX  DERIVATION OF FILTER RESPONSES B 9 a i (sT9 )
sT
lim y(t) = lim s 2 i =1
M
A. Background t s 0
s 1 + a (sT )i
The conventional lowpass filter and ramptracking filter
i =1
i 9
are both based on the general form of a filter:
M
i 1
(1 + sT8 ) BT9 a1 + a i (sT9 )
G(s) = (A4.1) = lim i=2 (A4.6)
(1 + sT9 ) M s 0 M
1 + a i (sT9 )i
i =1
The steadystate response of the output, y, to various
inputs, u, is calculated from the final value theorem. = B*T9 * M
lim y(t) = lim(s * U(s) *(G(s) 1)) (A4.2) Parabolic input: C/s3
t s 0
lim y(t) =
t
B. Conventional Low Pass Filter
The conventional lowpass filter is obtained from A.1 by
setting T8 = 0. The denominator of A.1 can be expanded as C. RampTracking Filter
follows:
 42 
IEEE Tutorial Course Power System Stabilization via Excitation Control June 2007
1 + sT8
G(s) 1 = M
1
a (sT )
i=0
i 9
i
M (A4.7)
s(T8 MT9 ) a i (sT9 )i
= M
i=2
a (sT )
i =0
i 9
i
M
A s(T8 MT 9 ) a i (sT9 )i
lim y(t) = lim s M
i=2
t s0
s
i=0
a i (sT9 ) i
=0
M
B s(T8 MT 9 ) a i (sT9 )i
lim y(t) = lim s 2 M
i=2
t s0
s
i =0
a i (sT9 )i
= T8 MT9
M
i
C a i (sT9 )
lim y(t) = lim s 3 iM= 2
t s0
s i
i =0
a i (sT9 )
M
i
a 2 T 9 + a isi 2 T9
= lim C M
i =3
(A4.11)
s0
1 + a i (sT9 ) i
i =1
M 1
= CT9 i
i =0
 43 
IEEE Tutorial Course Power System Stabilization via Excitation Control June 2007
 44 
IEEE Tutorial Course Power System Stabilization via Excitation Control June 2007
for terminal voltage transducers. Notch rejection filters or Speed should be monitored directly from the turbine
a double order low pass filter with break points at 60 Hz or generator shaft. A gear wheel mounted directly on the
lower will normally be sufficient if good signal shielding turbinegenerator shaft with stationary electromagnetic
practices are followed. pickups (or the equivalent realized through electrooptics)
is normally employed. This produces a signal whose
D. Field Voltage
frequency or pulse repetition rate is proportional to the
Field voltage is already a direct quantity although it may speed of the shaft. (Speed transducers driven by couplings
contain noise components (360 Hz and higher) several are normally unsatisfactory for observation of the
times larger than its direct voltage (particularly in thyristor torsionals as they tend to produce additional modes
controlled systems). In addition, it is necessary to provide representative of the coupling in addition to, or instead of,
a transducer which will isolate the instrumentation from the those of the shaft system.) The speed measurement
high power floating field circuit, while maintaining the technique is usually based upon a conventional tachometric
desired 10 Hz or higher signal bandwidth. Transducers circuit, although other methods have also been used
which will provide an output voltage proportional to the successfully.
generator field voltage for frequencies up to about 1 kHz In all machines, the generator shaft moves around
while maintaining the required (2000 V) level of isolation somewhat in the bearings. This can lead to "noise"
are commercially available. As field voltage varies widely components at 60 Hz for a 3600 rpm machine and 30 Hz
with input to the voltage regulator, suppression of the for an 1800 rpm machine. Such components can be
steadystate value may not be necessary. partially removed by the use of either multiple pickups to
For the measurement of excitation response times on cancel this motion (diametrically opposed speed sensors
high initial response systems, maximum bandwidth have been shown to be quite effective) or sharply tuned
(minimum filtering) should be used, but for stabilizer notch filters to attenuate them. Slowly rotating hydro
tuning, a reduced bandwidth will suffice. [62] Techniques generators have noise components below 10 Hz (e.g. 3 Hz
similar to those used for terminal voltage will allow the for a 180 rpm machine), so this signal is seldom used in
specification of adequate filtering for stabilizer setup. A stabilizers for hydro applications.
simple second order low pass filter with break points at The first (lowest frequency) torsional mode is lightly
about 50 Hz will normally be adequate for the damped and normally subjected to random excitation by
measurement of field voltage of most excitation systems. the steam supply system. Hence it can frequently be seen to
E. Generator Speed come and go in a more or less random fashion in a
The measurement of shaft speed is extremely useful for sufficiently sensitive recording. The amplitude of such
assessing power system damping (although electrical torsional oscillations under normal operation might
power can also be used for this purpose). In addition, the typically be of the order of 0.005% of rated speed when
PSS transfer function has historically been defined with monitored at the end of the shaft. They might be higher or
shaft speed as its input. Thus, measurement of shaft speed much lower depending upon the loading condition and the
has been critical to the development of PSS. However, nature of the system load.
implementation of dualinput stabilizers using electrical Normally, the filtering used for speed measurement is
power and internal frequency has lessened the more complex than that used with other quantities, as
importance of shaft speed measurement. dictated by the requirements for high gain, wide bandwidth
Shaft speed measurement is still important in turbo and high noise attenuation. Because one is interested only
generators because the PSS, if improperly designed or in very small changes about the steadystate value, the
adjusted, has the capability of exciting shaft torsional steady state component can be subtracted, easing the
oscillations through excitation control as discussed in instrumentation requirements. Such direct signal
Section V. Therefore, in these types of machines it is suppression must be extremely stable. Alternatively, high
important to monitor these mechanical modes and ensure pass filtering can be used in which case the signal is "rolled
that they are not affected by the stabilizer. The first off" below some low (e.g. 0.01 Hz) cutoff frequency.
torsional mode is the one most likely to be excited through The combination of transducing and filtering should then
the stabilizer and is also the one most readily monitored at allow the desired signal amplification with a minimum of
the ends of the generator shaft. Typical torsional mode phase shift and attenuation in a band from about 0.1 Hz or
profiles are shown in Figure 5.7. The extent to which lower to about 50 Hz. For fourpole steam turbine
various torsional modes are observable at any shaft location generators, the high frequency end of the band can be
can be assessed from such profiles. further restricted to about 25 Hz as torsional components
Speed is the most difficult signal of those described to are proportionately lower in frequency. This allows the
monitor successfully. This is particularly due to the fact possibility of filtering of components at 30 Hz.
that the speed changes to be observed are of the order of To monitor this signal adequately, notch rejection filters
0.05% of rated and to "see" the torsional components, wide at 30 Hz (for 4 pole machines only), 60 Hz and higher
bandwidth is required, (50 Hz for a 3600 RPM machine). harmonics, combined with low pass filters would normally
 45 
IEEE Tutorial Course Power System Stabilization via Excitation Control June 2007
be required. For example, consider a tachometric circuit revolution to produce adequate dynamic torque angle
which produces 2000 pulses per second at rated speed. results.
Assume 100% of 2 kHz ripple and that to observe the
I. Signal Recording
torsionals adequately this must be attenuated to 0.001% of
rated. In other words 2 kHz must be attenuated by a factor Many varieties of signal conditioning, recording, and
of 100,000 or 100 db. If a 4th order filter is chosen so that data acquisition equipment are suitable for PSS and
100 db attenuation occurs at 2 kHz, then a 4th order break excitation system testing. Basically, a multichannel
point at 110 Hz would be possible. A torsional component recording system with built in signal amplification and a
at 40 Hz would then experience an attenuation of only 28% bandwidth of about 100 Hz is required. Increased
or 2.2 db, but would be phase shifted by 80 in the filter. flexibility is provided by a recording system with high
The effects of any notch filters used would have to be impedance differential inputs, high common mode
added to this. capability or isolation, internal signal bias, built in
amplification and signal conditioning and a wide range of
F. Terminal or Internal Frequency recording speeds. There are a multitude of recorders and
In lieu of shaft speed, terminal or internal frequency is data acquisition systems that will meet these requirements.
often used. Internal frequency is defined as the frequency Some excitation systems may have builtin data recording
of the internal voltage phasor obtained by adding the capabilities that can be used for this purpose as well.
voltage drop across the quadrature axis impedance to the
generator terminal voltage. The voltage drop across the III. TESTING TECHNIQUES
quadrature axis impedance is determined by multiplying Once the desired signals are in a form suitable for
the generator terminal current phasor by the quadrature axis observation or measurement with the desired accuracy,
impedance, Xq. This measurement has proven to be easier gain and bandwidth, the following test procedures can be
to obtain and more accurate for salient pole hydro performed.
generators than for round rotor machines that are subject to
quadrature axis saturation; however, it has been used A. Step and Impulse Response Testing
successfully with all types of machines. While various types of perturbations can be considered
To transduce frequency from either the terminal voltage for the time domain evaluation of a system, the ones used
or internal voltage phasor a method much like the one most often in PSS testing are step and impulse disturbances
associated with shaft speed is used. In this case the pulses added to the terminal voltage reference signal. If the
are generated by zero crossings or other points on the desired worst case power system conditions can be set up,
waveform. Care must be taken to filter out the effects of they permit quick and simple checks on the stability of the
harmonics on the input waveforms. A signal with 10 Hz overall system. Without extensive testing or comparative
bandwidth much like the terminal voltage signal is desired. model test results, they do not allow much insight into the
way parameters should be modified for improved tuning;
G. Power System Stabilizer Output
however, a series of time domain tests with various PSS
This quantity is normally available in a form suitable for adjustments can be useful in verifying model accuracy.
direct observation or recording. It may be necessary to 15.35 655
provide appropriate isolation, particularly if the voltage
15.30 650
regulator is not grounded. Instrumentation must present a
15.25 645
high input impedance so that it does not overload the
stabilizer output. 15.20 640
Terminal Voltage, Vt (kV)
 46 
IEEE Tutorial Course Power System Stabilization via Excitation Control June 2007
typical step response with the stabilizer out of service and PSS(s) frequency response in Figure 5.4.
in service is shown in Figures 5.1a and 5.1b
15.40 640 The combination of characteristics 1 and 3 determine the
15.35 635 ability of the stabilizer to contribute damping at the various
15.30 630
machinesystem and interarea modes. The combination of
15.25 625
characteristics 2 and 3 (the stabilizer open loop response)
Terminal Voltage, Vt (kV)
Gain (dB)
10
14.95 595 15
20
14.90 590
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 5.5 6 6.5 7 7.5 8 25 Without PSS
With PSS
Time (s) 30
0.1 1.0 10
Vt MW Frequency (Hz)
0
Figure 5.1b Typical system step response with PSS on
50
Phase (deg)
100
This test is easily performed with the transducers and 150
data acquisition discussed in section II and a simple circuit 200
and switch (or software tools) to generate the input signal. 250
validation of the model and hence increased confidence in Figure 5.2 Frequency Response of Et(s)/EtRef(s)
20
30
the PSS and excitation system. 40
50
Pe/Vref
B. Frequency Response Testing 60 Freq/Vref
70
Frequency response characteristics permit much greater 0.1 1.0 10
Frequency (Hz)
insight into the small signal operation and tuning of a 200
control system than time responses. There are three 100
100
the tuning of the PSS: 200
300
60
1.0 10
2. The transfer function ((s)/PSS(s)) describing the Frequency (Hz)
0
stabilizers. The transfer function can also be deduced from
90
the more readily measurable change in electrical power vs. 180 Speed/Vref
360
PSS(s)/(s); PSS(s)/Pe(s); PSS(s)/Pacc(s); or Figure 5.4 Overall PSS Frequency Response (Open Loop Responses)
equivalent, as shown superimposed on the Pe(s) /
 47 
IEEE Tutorial Course Power System Stabilization via Excitation Control June 2007
C. Equipment and Techniques for Frequency Domain In addition, because of the random frequency distribution,
Analysis there is a lower probability of over exciting sharply tuned
While frequency response characteristics could be resonances.
measured using a signal generator and recorder, Most randomnoise analyzers have features that allow
commercially available signal analyzers greatly facilitate them to be used in a manner similar to sinewave analyzers
such measurements, automate the procedure and allow by replacing the random noise input signal with a
much higher transfer function resolution with lower levels sinusoidal input signal.
of disturbance input. These commercially available D. General Comments
analyzers include both frequency response analyzers that
It is not always convenient or expedient to perform
take sinewave measurements at discrete frequencies and
frequency response tests over the desired range of possible
analyzers that use random noise techniques.
machine and system conditions. If the field test results can
Sinewave analyzers allow the measurement of the
be used to confirm computer models, the models can then
magnitude and phase of the ratio of output to input of a
be used to investigate the desired range of system
control block at the distinct frequency at which the system
conditions. When the frequency response characteristics
is excited. Because correlation techniques are employed,
have been measured, the stabilizer phase lead
accurate measurements can be obtained even when the
characteristics can be adjusted as described in other
measured signal at the test frequency is deeply "buried" in
sections.
noise.
After the stabilizer phase compensation has been
To determine the transfer characteristic of an arbitrary
suitably adjusted, the stabilizer frequency response can be
control system block, the instrument actually measures two
checked. The combination of the appropriate
characteristics and performs the following calculation:
machine/system transfer function and the stabilizer transfer
function provides the open loop characteristic which can be
V 2 ( j ) V 0 ( j ) V 2 ( j )
G ( j ) = = (5.1) used for the determination of control loop stability.
V 0 ( j ) V1 ( j ) V1 ( j )
IV. ON SITE TUNING AND STABILITY ASSESSMENT
The instrument generates the sinusoidal signal V0 which
drives the control system, generally through addition to the A. The Excitation System
terminal voltage reference signal to obtain the system The generator excitation system is normally designed to
transfer functions Et(s) / PSS(s) and (s) / PSS(s). meet a host of cost and performance criteria described by
With this instrument, care must be exercised in the the purchaser. It is then set up to meet specified steady
choice of signal frequency and amplitude, particularly near state regulation requirements and to respond at the rate and
system resonance. The machinesystem resonant local with a forcing capability specified for the desired transient
mode frequency (0.82 Hz) should be approached with stability enhancement through excitation control. The
caution. voltage regulator should be adequately damped on open
circuit. Typical open circuit voltage responses to step
Due to possible excitation of torsional modes, tests change in voltage reference are shown in Figure 5.5.
should not be conducted at frequencies higher than 7 Hz on 1.025
turbogenerators without special precautions which are
beyond the scope of this work.
1.020
Terminal Voltage (pu)
increased resolution and the use of a coherence function Figure 5.5 Offline voltage step responses
allows an indication of the quality of the measurement. After the excitation system has been properly adjusted,
As with sinewave analyzers, useful results can be the power system stabilizer should be able to contribute
obtained even when the response to the test signal is sufficient damping to prevent oscillatory instability over
smaller than the uncorrelated system noise. Random noise the range of possible system configurations and conditions
analyzers can obtain the frequency response over the entire envisaged for any machine. In other words, the stabilizer
spectrum simultaneously, thereby reducing the testing time. should have the effect of moving the smallsignal stability
 48 
IEEE Tutorial Course Power System Stabilization via Excitation Control June 2007
limit beyond other power system stability limits. excitation system response requirements are low, it will be
High speed excitation systems, while contributing to relatively easy to provide a satisfactory stabilizer  in fact it
better transient stability and allowing larger operating may not be necessary to add damping at all for the local
angles than slower excitation systems, can unfortunately machine/system mode of oscillation. For interarea mode
produce an additional negative component of damping. stability, the unit must provide an appropriate damping
Fortunately, a welltuned PSS can overcome this negative contribution relative to its share of generation in the area.
damping and provide substantial additional positive
C. PSS Testing
damping for both interarea and local machinesystem
modes. While providing the required damping, action of The first critical step in PSS testing is measuring the
the PSS should not detract significantly from the high transfer function from the PSS output to terminal voltage as
speed excitation system contribution to improvement of shown in Figure 5.2. This is normally performed with the
system transient stability. unit online at light load, so the effect of rotor swings on
Slower excitation systems do not contribute as much to the response is minimal. Then, the PSS compensation
the improvement of transient stability, but at the same time transfer function is selected and documented as in Figure
do not contribute significantly to negative damping of the 5.4. [30] Finally, PSS gain is chosen and a benchmark step
local mode. With slow systems it is difficult to add much response is obtained as in Figure 5.1.
damping to the machinesystem mode via a power system
stabilizer, but in turn little is required. Depending upon the V. SHAFT TORSIONAL OSCILLATION
system, it still may be possible to provide a significant Consider first a simple model of a spring and mass in
contribution to interarea mode stability. torsion as shown in Figure 5.6. The equation for rotational
With modern excitation systems, high steadystate gain motion of the mass can be written as:
(200) or integral gain (in many digital systems) is
normally employed. Depending upon power system d 2
requirements and design philosophy, transient gain J + K + T = 0 (5.2)
reduction may also be used. Systems without transient gain dt 2
reduction can contribute more to transient stability, but at where J is the inertia
the expense of an additional negative contribution to local K is the spring constant of the shaft
mode damping. However, this can be overcome by using T is an external torque input.
slightly higher stabilizer gains. For this reason, care should
be exercised in the comparison of stabilizer gains for If the external torque is a constant or zero (T=0), the
systems with and without transient gain reduction. A factor equation is one of an undamped system which would
of approximately the value of the transient gain reduction oscillate continuously if set in motion. Assume now that
separates the two types of systems for an equivalent the external torque input T can be manipulated, say as a
contribution to system damping (when the stabilizing function of the speed of the mass as shown in Fig 5.6.
signal is injected before the TGR circuit) [30]. Also, gain
for interarea modes and the machinesystem local mode
will be different for these two types of systems. In either
case, it should be possible to tune the stabilizer to entirely
remove the power system oscillatory instability.
B. Tuning Criteria
The most important PSS tuning criterion is that, after
choosing and setting up the excitation system, the power T
system stabilizer should move any oscillatory stability G
limitations beyond all other power system limitations. This
criterion should hold for all possible machine and system Figure 5.6  Simplified torsional oscillation model
operating conditions. In some power systems, providing
damping of interarea modes may be of primary In a turbinegenerator, torque changes could be produced
importance. If so, adequate gain and phase relationships in this manner through the field of the generator, if shaft
must be maintained for all known oscillatory modes. Many speed were used as an input to the excitation control.
times these modes are not visible during field testing, so If this feedback path constitutes a straight gain G,
computer models and analytical techniques must be used to equation 5.2 becomes:
verify proper tuning.
The degree of difficulty experienced in providing d 2 d
damping for all oscillatory modes is dependent on the J 2 +G + K = 0 (5.3)
performance required from the excitation system. If dt dt
 49 
IEEE Tutorial Course Power System Stabilization via Excitation Control June 2007
in which case damping of the torsional oscillation is it would be necessary to ensure that these torsional modes
produced as a function of the amplitude of G. If, however, could not be excited.
the transducer produces not just gain but phase lag of In practice, a combination of selective speed location
greater than 90 at the oscillating frequency, a negative and/or sharply tuned rejection filters at the torsional
damping component of torque is produced and oscillation frequencies is employed to preclude torsional excitation.
will build spontaneously. Because the lowest torsional frequency is only about a
In the frequency domain a transfer function of the form decade away from frequencies at which stabilization is
required, the torsional filter does add some small
T ( s ) G undesirable phase lag to the stabilizing loop, and has the
G( s) = = (5.4) effect of limiting the maximum usable stabilizer gain.
( s ) (1 + As )(1 + Bs )
In a stabilizing system which uses terminal frequency as
an input, torsional frequencies are attenuated somewhat
could produce the necessary phase lag to provide a
compared to a stabilizer which uses an end of shaft speed
negative component to undamp this mode.
signal [31], [32], [33]. In stabilizers using electrical or
The rotating components of a large steam turbine form a
accelerating power as input, the torsionals are inherently
more complex torsional system with many shaft sections.
highly attenuated [31], [32], [33], [58].
The torsional components are all very lightly damped with
decay time constants measured in seconds. The frequencies
of the torsionals are functions of turbine generator
configuration and design. Some large units may have first
torsional modes as low as 7 Hz, while other units may have
a highest torsional mode near 55 Hz.
The rotating system is connected to the outside world
through the rotating air gap torque. The strength of this tie
can be represented by a spring connected to a fixed
reference. The steady state rotation of the shaft is now
effectively removed from the problem. A dashpot in
parallel with this spring represents machinesystem
damping. The total mass, spring and dashpot define the
machinesystem mode of oscillation to be damped by the
PSS.
The objective of the PSS is to enhance this machine
system damping by generating a torque in phase with speed
changes at the frequency of this oscillation. If the signal is
Figure.5.7 Typical Shaft System And Torsional Mode Shapes
generated directly from shaft speed measurement, it is
difficult to find a place on the shaft where the transducer
responds only to the mode in which the shaft acts as a rigid
body oscillating against the power system. If speed is
measured at the generator end of the shaft and if the speed
to torque transfer function constitutes strictly a gain and no
phase lag, damping would be provided for both the system
mode and the first torsional mode. Analysis of damping of
this first torsional mode is directly related to the simple
modal analysis in the first part of this section.
Unfortunately, the phase lags through the stabilizer and
generator (Equation 5.4) are sufficient to cause instability
of the first torsional mode when the stabilizer is adjusted
for appropriate damping of the system mode. This
possibility of torsional excitation through the power system
stabilizer and excitation system has been well documented.
[55][41]
Placing the speed sensor at the node of the first torsional
mode would make this mode invisible to the control
system, but this location would normally not be equipped
for speed measurement and might be inaccessible. Even if
speed could be measured at this node, other torsional signal
components would still be detected by the transducer, and
 50 
IEEE Tutorial Course Power System Stabilization via Excitation Control June 2007
80
APPLICATION CONSIDERATIONS 20
0
I. INTRODUCTION 7.0
applications. 0.9
0.8
0.6
rotating main exciters and lowpower (less than 10A) pilot 1.2
1.1
Pow er
0.9
0.10
undesirable in a utility application when a PSS is also used. 0.05
0.00
0.10
is a generator overvoltage. 0.15
 51 
IEEE Tutorial Course Power System Stabilization via Excitation Control June 2007
1000.00 140
Measured
Nominal 120
100
100.00
Phase (degrees)
Magnitude (V/V)
80
60
10.00
Magnitude
40
20
Phase
1.00 0
0.10 1.00 10.00
Frequency (Hz)
 52 
IEEE Tutorial Course Power System Stabilization via Excitation Control June 2007
u ( t ) = f [ s ( t ), y( t ), u ( t T )] (7.1)
 53 
IEEE Tutorial Course Power System Stabilization via Excitation Control June 2007
AVR
+ yr &
uc y_
u Exciter
Controller System
e Transmission Lines
yc
Adaptation Mechanism
u yr +  y
Fiure.7.1 MRAC Structure Reference +
Model
e 
Reference model must be selected such that the actual Z1
FLC
system is capable of matching its performance
characteristics. The ideal reference model is 1. However, e
when this is not achievable due to system limitations, delay Figure 7.2 System Configuration
of inputs will occur even for a first order model. This
might cause the response of the reference model to be Results of a number of studies show that this APSS
substantially different from that of the actual system and provides good damping over a wide operating range and
will be interpreted as system fault even under normal improves the performance of the system. An illustrative
conditions. example showing the system response to a three phase to
The most important feature in ensuring the success of ground fault with the selflearning MRAC based FLC and a
MRAC is the selection of a proper reference model and its fixed center FLC is given in Fig.7.3.
parameters. The selected parameters must be such that the
system is capable of following the reference model output
and that the control signal remains within the physical
control limits.
A systematic method to determine a proper reference
model for the plant is described in [70]. In this approach
time domain performance of the controlled system with an
analytical poleshift APSS [71] was studied by simulation
studies under various operating conditions and
disturbances. The best closedloop pole locations that do
not violate the control constraints were obtained. A set of
average zeros and closedloop pole locations of the poles
obtained from these studies was used as the parameters of a
discrete third order reference model.
Application of an APSS based on the MRAC principle is
shown in Fig.7.2. An FLC with selflearning capability is
used to adapt the system performance to track the reference Figure 7.3 ThreePhase To Ground Fault At The Middle Of One
model. Two inputs, generator speed deviation and its Transmission Line And Successful Reclosure With APSS (MRAFC) And
Fixed FLC PSS (P= 0.95 pu, 0.9 PF Lag)
derivative, and the supplementary control output, each have
seven membership functions. The FLC uses the Mamdani B. Indirect Adaptive Control
type fuzzy PD rule base [72]. Updating the center points of A general configuration of the indirect adaptive control
the controller input membership functions, i.e. the weights as a selftuning controller is shown in Fig.7.4. At each
of the fuzzy controller, using the steepest descent algorithm sampling instant, the input and output of the generating
provides it with a selflearning capability. It can thus adapt unit are sampled and a plant model is obtained by some on
the system performance to track the reference model. line identification algorithm to represent the dynamic
behavior of the generating unit at that instant in time. It is
expected that the model obtained at each sampling instant
can track the system operating conditions.
 54 
IEEE Tutorial Course Power System Stabilization via Excitation Control June 2007
 55 
IEEE Tutorial Course Power System Stabilization via Excitation Control June 2007
 56 
IEEE Tutorial Course Power System Stabilization via Excitation Control June 2007
J = E[ y(t + 1)  y r (t + 1)] 2 (7.12)
1 1
<> (7.11)
 57 
IEEE Tutorial Course Power System Stabilization via Excitation Control June 2007
 58 
IEEE Tutorial Course Power System Stabilization via Excitation Control June 2007
 59 
IEEE Tutorial Course Power System Stabilization via Excitation Control June 2007
corresponds to one specific function with the node are used in a manner similar to Fig.7.11, one acting as the
functions in each layer being of the same type. With this controller and the other acting as the predictor. The plant
network representation of the fuzzy logic system, it is identifier can compute the derivative of the plants output
straightforward to apply the backpropagation or a similar with respect to the plants input by means of the back
method to adjust the parameters of the membership propagation process illustrated by the line passing through
functions and inference rules. the forward identifier and continuing back through the
neurofuzzy controller that uses it to learn the control rule.
The selflearning ANF PSS was initially trained offline
on a power system simulation model over a wide range of
operating conditions and disturbances. Electric power
deviation and its integral were used as the input to the
stabilizer. The ANF PSS, with the parameters, membership
functions and inference rules obtained from the offline
training procedure, was implemented on a DSP mounted on
a PC and its performance was evaluated on a physical
model of a power system in the laboratory. A digital CPSS
Figure 7.13 Basic Structure Of Fuzzy Logic Controller was also implemented in the same environment on the DSP
board for comparative studies.
In this network, the links between the nodes from one Out of the various tests, results for a 0.25 pu step
layer to the next layer only indicate the direction of flow of decrease in the input torque reference applied at 1s and
signals and part or all of the nodes contain the adjustable removed at 9s with the generator operating at 0.9 pu power,
parameters. These parameters are specified by the learning 0.85 pf lag and 1.10 pu Vt are shown in Fig.7.14. The
algorithm and should be updated according to the given ANF PSS provides a consistently good performance for
training data and a gradientbased learning procedure to either of the two disturbances.
achieve a desired input/output mapping. It can be used as Simulation studies on a single machine connected to a
an identifier for nonlinear dynamic systems or as a non constant voltage bus and on a multimachine power system
linear controller with adjustable parameters. [80] and experimental studies on a physical model of a
D. Training And Performance power system have demonstrated the effectiveness of the
ANN PSS in improving the performance of a power system
Because the neurofuzzy controller has the property of
over a wide operating range and a broad spectrum of
learning, fuzzy rules and membership functions of the
disturbances.
controller can be tuned automatically by the learning
algorithm. Learning is based on the error in the controller
output. Thus it is necessary to know the error that can be
evaluated by comparing the output of the neurofuzzy
controller and a desired controller.
To train this controller as an adaptive network based
fuzzy PSS (ANF PSS), training data was obtained from a
selfoptimizing poleshifting APSS. Training was
performed over a wide range of operating conditions of the
generating unit including various types of disturbances.
Based on earlier experience, seven linguistic variables for
each input variable were used to get the desired
performance.
Extensive simulation [86] and experimental studies with
the ANF PSS show that it can provide good performance Figure 7.14. Comparison Of ANF PSS and CPSS Responses to a
over a wide operating range and can significantly improve 0.25 p.u. Step Torque Disturbance (P= 0.9 p.u., 0.85 PF Lag)
the dynamic performance of the system over that with a
F. NeuroFuzzy Controller Architecture Optimization
fixed parameter CPSS.
Adaptive fuzzy systems offer a potential solution to the
E. SelfLearning ANF PSS knowledge elicitation problem. The controller structure,
In the above case the ANF PSS was trained by data expressed in terms of the number of membership functions
obtained from a desired controller. However, in a general and the number of inference rules, is usually derived by
situation, the desired controller may not be available. In trial and error. The number of inference rules has to be
that case, the neurofuzzy controller can be trained using a determined from the standpoint of overall learning
selflearning approach [87]. capability and generalization capability.
In the selflearning approach two neurofuzzy systems The above problem can be resolved by employing a
 60 
IEEE Tutorial Course Power System Stabilization via Excitation Control June 2007
genetic algorithm to determine the structure of the adaptive center (active neuron). The other neurons with centers
fuzzy controller. By employing both genetic algorithm and quite different from the input vector will have outputs near
adaptive fuzzy controller, the inference rules parameters zero (nonactive neurons).
can be tuned and the number of membership functions can The connections between the hidden neurons and the
be optimized at the same time. output node are linear weighted sums as described by the
equation:
VII. AMALGAMATED ANALYTICAL AND AI BASED PSS
nh pc 2
A. Adaptive PSS With NN Identifier And PoleShift
Control
y= exp .
2
i
(7.16)
i =1
A selftuning APSS described above can improve the
dynamic performance of the synchronous generator by where ci, , and nh are the centers, widths, weights
allowing the parameters of the PSS to adjust as the and the number of hidden layer neurons, respectively.
operating conditions change. However, proper care needs To make the proposed RBF identifier faster for online
to be taken in the design of the RLS algorithm for applications, the hidden layer is created as a competitive
identification to make it stable, especially under large layer wherein the center closest to the input vector becomes
disturbances. the winner and all the other nonactive centers are
It is possible to make the identification more robust by deactivated. Also, the scalar weights are modified as a
using a NN for identifying the system model parameters.
vector whose size equals the size of the input vector.
An analytical technique, such as the poleshift control, can
The weight vector is given by:
be retained to compute the control signal. One approach,
using a radial basis function (RBF) network for model
parameter identification, is described below. The PSS [
( t ) = a 1' a 2' a 3' b1' b 2' b 3' ] (7.17)
shown in Fig.7.4 now consists of an ANN identifier and the
poleshifting control algorithm described above. Linearizing the output of the RBF, y(t) = f [y(t1), u(t
The RBF network, Fig.7.15, is used to identify the 1)], by Taylor series expansion at each sampling instant, a
system model parameters, ai, bi, (7.13) and (7.14). The onetoone relationship between the weight vector and
network consists of three layers: the input, hidden and the system model parameters ( t ) (equation 7.6) can be
m
output layers. The input vector is:
obtained. These parameters are then used in computing the
control signal.
V ( t ) = [ Pe ( t T ), Pe ( t 2 T ), The RBF identifier was first trained offline to choose
Pe ( t 3 T ), u ( t T ), (7.15) appropriate centers using data collected at a number of
u ( t 2 T ), u ( t 3 T )] operating points for various disturbances. The nmeans
clustering algorithm used for training yielded 15 centers for
the RBF model. After the offline training, the weights
(system parameters) were updated online to obtain the
appropriate control signal using the poleshifting controller.
A 100ms sampling period was chosen for digital
implementation.
Results of an experimental study for a 0.10 pu decrease
in torque reference applied at 10s and removed at 20s, with
the generator operating at 0.6 pu power, 0.92 pf lead and Vt
of 0.99 pu are shown in Fig.7.16. It can be seen that the
APSS can provide a welldamped response.
Figure 7.15 Radial Basis Function Network Model
 61 
IEEE Tutorial Course Power System Stabilization via Excitation Control June 2007
Figure 7.16 Pe Response For 0.1 p.u. Input Torque Reference Step
Change With APSS Figure 7.17 Three Phase To Ground Fault At The Middle Of One
Transmission Line And Successful Reclosure (P=0.95 pu, 0.9 PF Lag)
B. Adaptive PSS with Fuzzy Logic Identifier and Pole
Shift controller C. Adaptive PSS With RLS Identifier And Fuzzy Logic
TakagiSugeno (TS) fuzzy systems have been Control
successfully employed in the design of stabilization control Fuzzy logic controllers (FLCs) have attracted
of nonlinear systems. considerable attention as candidates for novel
A nonlinear plant can be represented by a set of linear computational systems because of the advantages they offer
models interpolated by membership functions of a TS over the conventional computational systems. They have
fuzzy model. Although the TS system identifier is a been successfully applied to the control of nonlinear
NARMAX model, at each sample an average linear dynamic systems, especially in the field of adaptive
discrete ARMA model can be determined to identify the control, by making use of online training.
controlled plant according to the current active rules. This A selflearning adaptive fuzzy logic controller has been
ARMA model can be used to determine the control signal developed. Only the inputs and outputs of the plant are
by the poleshifting control strategy. Using this approach, measured and there is no need to determine the states of the
a selftuning adaptive controller has been developed and plant. Using online training by the steepest descent
applied as a PSS [88]. method and the identified system model, the adaptive FLC
The proposed singleinput singleoutput TS model used is able to track the plant variations as they occur and
for the identification of dynamic systems is composed of compute the control.
fuzzy rules , the consequent part of which provides the rule In the proposed controller, a discrete model of the plant
output at time k based on the past inputs and past outputs is first identified using the RLS parameter identification
with fuzzy sets designed in universe of discourse. The method. This allows a continuous tracking of the system
consequent part of the rule then identifies the parameters of behavior.
a desired order discrete model of the plant. Two parallel The control learning is based on the prediction of the
online learning procedures, one each for the identification identified model. The identified model output is used as
of premise and consequent parameters, are used to track the input to the Mamdanitype PD controller [72]. The center
plant in realtime [89]. points of the controller inputs are updated [89] by treating
In the proposed TS system for generating unit them exactly the same as the weights of an NN and by
identification, two input signals, the past control input, u(k using the steepest descent algorithm with chain rule.
1), and the past generator speed output, y(k1), are used to The proposed adaptive FLC has been applied as an
identify a 3rd. order model of the plant. The output at adaptive fuzzy PSS (AFPSS) [90]. For the AFPSS, the
sample k is the estimated generator speed output, (k). The generating unit is identified as a 3rd. order model. The
TS system is trained by using the steepest descent controller has two input signals, the generator speed
algorithm for the premise parameters and RLS algorithm deviation and its derivative, with an initial set of seven
for the consequent parameters using the error of the system equally spaced membership functions over the normalized
output and the estimated TS output. Initially a set of three universe of discourse. The output, the supplementary
equally spaced membership functions, over the normalized control signal, also having seven membership functions, is
universe, are used for the inputs of the system. added to the AVR summing junction. A number of
The response of the system with the TS system based simulation studies have been performed for various
identifier and poleshift controller based APSS has been disturbances at different operating conditions. An
studied for various disturbances at different operating illustrative result for a 0.05 pu increase in torque and return
conditions. One illustrative result for a three phase to to initial condition, shown in Fig.7.18, demonstrates the
ground fault is shown in Fig.7.17. performance of this AFPSS.
 62 
IEEE Tutorial Course Power System Stabilization via Excitation Control June 2007
0.69
APSS VLMAX
CPSS LI
0.68
No PSS KL
0.67 FL VLMIN
VSTMAX
Power Angle [rad]
0.66 VIMAX
KI
+
VST
+
0.65
+
FI VIMIN
0.64
Speed VSTMIN
Transducers
KH
0.63 VH MAX
0.62
0 1 2 3 4 5 6 7
H
time [s] FH
Figure 7.18 Response To A 0.05 p.u. Step Increase In Torque And Return VH MIN
To Initial Condition (P= 0.95 p.u., 0.9 PF Lag)
Figure 7.19 Multiband PSS Algorithm Structure
 63 
IEEE Tutorial Course Power System Stabilization via Excitation Control June 2007
MW Deviation
10
ranges. The resulting stabilizing signal is formed by the 0.5 1 1.5 2 2.5
PSS Signal(%)
0
Speed Deviation(Hz)
Each branch of a differential filter was designed to
0
the figure 7.21 as base and concentrating just in the high 0.04
hybrid block that either can be used as washingout when Figure 7.23a MVAr step response in LG2 HPSsimulation
KH11= K17=0 or for leadlagging when KH11=KH17=1. With
such an arrangement, one can use a single branch (the
+50MVARS STEP
positive or negative) to tune the PSS similar to the usual
PSS.
MW Deviation
10
level parameters and the whole leadlag compensation 1 PSS4BNormal Gain (Test #7d16)
PSS4BHigh Gain(Test #17d16)
circuit is defined by six parameters. They are the three filter 2
Existing PSS (Test #5d16)
central frequencies FL, FI, FH and gains KL, KI, KH. 0.5 1 1.5 2 2.5
The time constants and gains are derived from simple 0.02
Speed Deviation(Hz)
0.04
Beside the several laborious theoretical investigations, 0.5 1 1.5 2 2.5
Hydro Qubec also is carrying out extensive field testing of Time in seconds
 64 
IEEE Tutorial Course Power System Stabilization via Excitation Control June 2007
M W S TE P  B b lu e : A c tu a l L G 2 P S S ; R e d : M B P S S ( N o r m a l G ai n )
0 M B  P S S i n c lo s e d lo o p ( T e s t # 1 0 d 1 6 )
C on ve n ti o n a l P S S i n c lo s e d  l o o p ( T e s t # 4 d 1 6 )
MW Deviation 1 0
2 0
3 0
4 0
5 0
6 0
0 5 10 15
1
0 .5
PSS Signal(%)
0
 0 .5
1
 1 .5
2
0 5 10 15
0
Speed Deviation(Hz)
 0 .0 2
 0 .0 4
 0 .0 6
 0 .0 8
 0 .1
 0 .1 2
0 5 10 15
20
2 0
4 0
6 0
G e n tilly 2 e ve nt r e c o r d e d o n J ul y 1 8 , 2 0 0 0 1 2 : 1 8 d ur i n g o p en  lo o p te s ti n g o f th e M B  P S S
0 5 10 15
sec R a w S p e e d E s tim a te s
0 .2
0 .15
Deviation (Hz)
0 .1
Figure 7.24 MW step response test in closed loop at LG2
0 .05
carried out at power level of 200MW. Figure 7.23a shows 0 .15 T un e d N o tc h F ilte r e d S p e e d
f n 1= 9 .9 5 Hz a nd f n 2 = 17 . 8 2 H z
0 .04
oscillation modes, the filtered speed signal after the notch 0 .03
PSS (p.u.)
0 .02
filters and the MBPSS output signal compared with the 0 .01
 65 
IEEE Tutorial Course Power System Stabilization via Excitation Control June 2007
 66 
IEEE Tutorial Course Power System Stabilization via Excitation Control June 2007
 67 
IEEE Tutorial Course Power System Stabilization via Excitation Control June 2007
[50] M. Klein, G.J. Rogers, S. Moorty, P. Kundur, Analytical [72] M. Mamdani, Application Of Fuzzy Algorithm For Control Of
Investigation of Factors Influencing Power System Stabilizers Simple Dynamic Plant," Proc. IEE, Vol. 121(12), 1974, pp. 1585
Performance, IEEE Transactions on Energy Conversion, Vol. 7, 1588.
September 1992, pp. 382390. [73] P. Eykhoff, "System Identification," Wiley London, 1974.
[51] Analysis and Control of Power System Oscillations, CIGRE Task [74] S.J. Cheng, Y.S. Chow, O.P. Malik, and G.S. Hope, "An Adaptive
Force 38.01.07 Final Report, December 1996. Synchronous Machine Stabilizer," IEEE Trans. on Power Systems,
[52] P.L. Dandeno, A.N. Karas, K.R. McClymont, and W. Watson, Vol. PWRS1(3), 1986, pp. 101109.
"Effect Of HighSpeed Rectifier Excitation Systems On Generator [75] B.D.O. Anderson, J.B. and Moore, "Linear Optimal Control,"
Stability Limits," IEEE Trans., Vol. PAS87, pp. 190201, January PrenticeHall, New Jersey, 1971.
1968. [76] K.J. Astrom, U. Borisson, L. Ljung, and B. Wittenmark, "Theory
[53] P. Kundur, M. Klein, G.J. Rogers, and M.S. Zywno, "Application Of And Application Of SelfTuning Regulators," Automatica, Vol. 13,
Power System Stabilizers For Enhancement Of Overall System 1977, pp. 457476.
Stability," IEEE Trans., Vol. PWRS4, pp. 614626, May 1989. [77] P.E. Wellstead, J.M. Edmunds, D. Prager, and P. Zanka, "Self
[54] W. Watson and G. Manchur, "Experience With Supplementary Tuning Pole/Zero Assignment Regulators," Intl. J. of Control, Vol.
Damping Signals For Generator Static Excitation Systems," IEEE 30, 1979, pp. 126.
Trans., Vol. PAS92, pp. 199203, January/February 1973. [78] P.E. Wellstead, D. Prager, and P. Zanka, "PoleAssignment Self
[55] W. Watson and M.E. Coultes, "Static Exciter Stabilizing Signals On Tuning Regulator," Proc. IEE, Vol. 126, 1979, pp. 781787.
Large Generators Mechanical Problems," IEEE Trans., PAS92, [79] S.J. Cheng, O.P. Malik, and G.S. Hope, "Damping Of MultiModal
pp. 204212, January/ February 1973. Oscillations In Power Systems Using A DualRate Adaptive
[56] F.P. deMello, L.N. Hannett, and J.M. Undrill, Practical Approaches Stabilizer," IEEE Trans. on Power Systems, Vol. PWRS3(1), 1988,
To Supplementary Stabilizing from Accelerating Power," IEEE pp. 101108.
Trans., Vol. PAS97, pp. 15151522, September/October 1978. [80] G.P. Chen, O.P. Malik,. and G.C. Hancock, "Implementation and
[57] D.C. Lee, R.E., Beaulieu, and J.R.R. Service, "A Power System Experimental Studies Of An Adaptive SelfOptimizing Power
Stabilizer Using Speed And Electrical Power Inputs Design And System Stabilizer," Control Engineering Practice, Vol. 2(6), 1994,
Field Experience," IEEE Trans., Vol. PAS100, pp. 41514167, pp. 969977.
September 1981. [81] O.P. Malik, V.A. Stroev, V.A. Strobel, G.C. Hancock, and R.S.
[58] J.P. Bayne, D.C. Lee, W. Watson, "A Power System Stabilizer For Beim, "Experimental Studies With Power System Stabilizers On A
Thermal Units Based On Derivation Of Accelerating Power," IEEE Physical Model Of A MultiMachine Power System," IEEE Trans.
Trans., Vol. PAS96, Nov/Dec 1977, pp 17771783. on Power Systems, Vol. 11(2), 1996, pp. 807812.
[59] F.W. Keay, W.H. South, "Design Of A Power System Stabilizer [82] O.P. Malik, C.X. Mao, K.S. Prakash, G.S. Hope and G.C. Hancock,
Sensing Frequency Deviation," IEEE Trans., Vol. PAS90, Mar/Apr "Tests With A Microcomputer Based Adaptive Synchronous
1971, pp 707713. Machine Stabilizer On A 400MW Thermal Unit," IEEE Trans. on
[60] J.M.C. Soares, F.H. Pons, F.Reichert, D.O. Reis, "Modelling And Energy Conversion, Vol. 8(1), 1993, pp. 612.
Field Tests Of A Supplementary Stabilizing Signal For General [83] P. Shamsollahi and O.P. Malik, "An Adaptive Power System
Use," National Seminar on the Generation and Transmission of Stabilizer Using OnLine Trained Neutral Networks," IEEE Trans.
Electrical Energy, Brazil, 1987. on Energy Conversion, Vol. 12(4), 1997, pp. 382387.
[61] P. Kundur, G.R. Berube, L.M. Hajagos, R.E. Beaulieu, Practical [84] A. Hariri. and O.P. Malik, "A SelfLearning AdaptiveNetwork
Utility Experience with and Effective Use of Power System Based Fuzzy Logic Power System Stabilizer In A MultiMachine
Stabilizers, IEEE PES Meeting July 2003, Vol 3, pp 17771785 Power System," Engineering Intelligent Systems, Vol. 9(3),
[62] "IEEE Guide for Identification, Testing, And Evaluation Of The September 2001, pp. 129136.
Dynamic Performance Of Excitation Control Systems" IEEE [85] J.G. Jang, "AdaptiveNetworkBased Fuzzy Inference System,"
Standard 421.21990. IEEE Trans. on Systems, Man and Cybernetics, Vol. 23(3), 1993, pp.
[63] M.M. ElMetwally, N.D. Rao, and O.P. Malik, "Experimental 665684.
Results On The Implementation Of An Optimal Control Of [86] A. Hariri, A. and Malik, O.P., "A Fuzzy Logic Based Power System
Synchronous Machines," IEEE Trans. on Power Apparatus and Stabilizer With Learning Ability," IEEE Trans. on Energy
Systems, Vol. PAS94(4), 1975, pp. 11921200. Conversion, Vol. 11(4), 1996, pp. 721727.
[64] S. Chen. and O.P. Malik, "An H Optimization Based Power System [87] J.G. Jang, "SelfLearning Fuzzy Controllers Based On Temporal
Stabilizer Design," IEE ProceedingsGeneration, Transmission and Backpropagation," IEEE Trans on Neural Networks, Vol. 3(5), 1992,
Distribution, Vol. 142(2), 1995, pp. 179184. pp. 714723.
[65] W.C. Chan, and Y.Y. Hsu, "An Optimal Variable Structure Stabilizer [88] Abdelazim, Tamer and O.P. Malik, "Fuzzy Logic Based Identifier
For Power System Stabilizer," IEEE Trans. on Power Apparatus and And PoleShifting Controller For PSS Application," Proceedings,
Systems, Vol. PAS102, 1983, pp. 17381746. IEEE Power Engineering Society 2003 General Meeting, July 1317,
[66] T. Hiyama, "Application Of RuleBased Stabilizer Controller To 2003, Toronto, Canada.
Electric Power System," IEE Proceedings C, Vol. 136(3), 1989, pp. [89] J.M. Adams and K.S. Rattan, "Backpropagation Learning For A
175181. Fuzzy Controller With Partitioned Membership Functions,"
[67] L.H. Zadeh, et. al., "Calculus Of Fuzzy Restriction In Fuzzy Sets Proceedings, NAFIPS 2002 Annual Meeting of the North American
And Their Application To Cognitive And Decision Process," Fuzzy Information Processing Society, pp. 172177.
Academic Press, 1975, pp. 140. [90] Abdelazim, Tamer and O.P. Malik, "An Adaptive Power System
[68] K.A. ElMetwally, G.C. Hancock,. and O.P. Malik, "Implementation Stabilizer Using OnLine SelfLearning Fuzzy System,"
Of A Fuzzy Logic PSS Using A MicroController And Experimental Proceedings, IEEE Power Engineering Society 2003 General
Test Results," IEEE Trans. on Energy Conversion, Vol. 11(1), 1996, Meeting, July 1317, 2003, Toronto, Canada.
pp. 9196. [91] R. Grondin, I. Kamwa, G.Trudel, J. Taborda, R. Lenstroem, L.
[69] Y. Zhang, O.P. Malik, and G.P. Chen, "Artificial Neutral Network GrinLajoie, J.P. Gingras, M. Racine, and H. Baumberger, "The
Power System Stabilizer In MultiMachine Power System MultiBand PSS: A flexible Technology Designed to Meet Opening
Environment," IEEE Trans. on Energy Conversion, Vol. 10(1), 1995, Markets," CIGRE 2000 in Paris, Paper 39201.
pp. 147155. [92] R. Grondin, I. Kamwa, G.Trudel, J. Taborda. "Modeling and closed
[70] Abdelazim, Tamer and O.P. Malik, "Power System Stabilizer Based loop validation of a new PSS concept, the MultiBand PSS" IEEE
On Model Reference Adaptive Fuzzy Control," Electric Power Power Engineering Society Summer Meeting, 2003, panel section.
Components and Systems, Vol. 33(9), 2005, pp. 985998. [93] I. Kamwa, R. Grondin, G.Trudel, " IEEE PSS2B versus PSS4B: The
[71] O.P. Malik, G.P. Chen, G.S. Hope, Y.H. Qin, and G.Y. Yu, Limits of Performance of Modern Power System Stabilizers", IEEE
"Adaptive SelfOptimizing PoleShifting Control Algorithm," IEE Trans. on Power Systems, Vol. 20. No.2 May 2005, pp 903915
ProceedingsD, Vol. 139(5), 1992, pp. 429438. [94] L.GrinLajoie, D. Lefebvre, M. Racine, L. Soulires, I. Kamwa,
"HydroQubec Experience with PSS tuning", Panel on System
 68 
IEEE Tutorial Course Power System Stabilization via Excitation Control June 2007
 69 
IEEE Tutorial Course Power System Stabilization via Excitation Control June 2007
Roger Beaulieu graduated from the Robert Grondin received his B.Sc.A.
University of Waterloo in Ontario in Electrical Engineering from
Canada, with a B.A.Sc, in 1967. After University of Sherbrooke, Canada in
a 26year career at Ontario Hydro, in 1976 and his M.Sc.from INRS Energie,
the areas of power system protection, Varennes, Canada in 1979. He then
stabilizers and power system modelling joined HydroQubec research
and testing, he retired to the life of an institute, IREQ. Currently a Senior
adviser to utilities and electrical Research Engineer in the Power
equipment manufacturers. He is a senior engineer with System Analysis, Operation and Control Department, he is
Goldfinch Power Engineering. leading research activities in the field of power system
dynamics and defense plans. Member of IEEE Power
Roger Brub received his Bachelor Engineering society and of CIGR, he is also a registered
and Masters of Engineering degrees professional engineer in province of Qubec, Canada
from McGill University in Montreal
Canada 1981 and 1982 respectively. Professor Arjun Godhwani is an
He worked for Ontario Hydro between emeritus professor of Electrical
1982 and 2000 in various roles Engineering at Southern Illinois
involving the design, testing and University Edwardsville. He has been
simulation of generator control systems. a member of IEEE Excitation Systems
Since 2000 he has been working as Senior Engineer with Subcommittee for over 10 years. He
Kestrel Power Engineering Ltd. a leading firm in the area actively consults in the field of
of generator control consulting and regulator compliance excitation systems.
testing. He is a member of the IEEE Power Engineering
Society and a participant in the Excitation Systems Sub Les Hajagos received his B.A.Sc. in
Committee and associated Working Groups. 1985 and his M.A.Sc. in 1987 from
the University of Toronto. Since 1988
Dr. George E. Boukarim received the he has worked mainly in the analysis,
B.S., M.E., and Ph.D. degrees in design, testing and modeling of
electric power engineering from generator, turbine and power system
Rensselaer Polytechnic Institute, Troy, control equipment and power system
NY, in 1987, 1988, and 1998, loads, first at Ontario Hydro, and since 2000 as one of the
respectively. From 1988 to 1994 and principals at Kestrel Power Engineering. He is a registered
from 1998 to the present he was with Professional Engineer in the Province of Ontario and an
the Power Systems Energy Consulting active member of the IEEE Power Engineering Society as
(PSEC) Department of the General Electric Company, chair of the Generator Model Validation and Excitation
Schenectady, NY, working in the area of power system System Modeling task forces.
 70 
IEEE Tutorial Course Power System Stabilization via Excitation Control June 2007
 71 
IEEE Tutorial Course Power System Stabilization via Excitation Control June 2007
 72 