• Embed Doc
  • Readcast
  • Collections
  • CommentGo Back
Download
 
IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICS—PART C: APPLICATIONS AND REVIEWS, VOL. 30, NO. 1, JANUARY 2000 131
An Adaptive Fuzzy-Synchronous Machine Stabilizer
Fouad Mrad, Sami Karaki, and Bassel Copti
 Abstract—
This paper describes an adaptive fuzzy-synchronous ma-chine power system stabilizer (PSS) that behaves like a proportionalintegral derivative (PID) controller. The implemented adaptive techniquepredicts tracking-error divergence and makes on-line adjustments to thecontroller-gain parameters in order to obtain a faster regulation of theerror signal. The proposed PSS is less sensitive to the quality of expertknowledge, and thus is more robust than a nonadaptive fuzzy PSS whensubjected to large disturbances. Hence, adequate controls are derivedfor the various error cases, and smooth control surfaces are achieved.Simulations are carried out using the proposed PSS to show that itconsistently gives a stable response with acceptable overshoots and settlingtimes on the frequency, angle, and voltage errors.
 Index Terms—
Adaptive control, fuzzy logic, power systems, stability.
I. I
NTRODUCTION
Operating conditions of a power system are continually changingdue to load level, generation variations, and line switching. Suchchanges can induce transient low-frequency oscillations, which aredifficult to control. A convenient yet effective way of damping theseoscillations is through the use of power system stabilizers (PSS’s),which produce positive damping signals to cancel out negativedamping of the excitation system introduced under some fault or otheroperating conditions. A classical lag–lead PSS can be optimized towork under some narrow range of operating conditions. However,its damping performance deteriorates as conditions change or whennetwork switching takes place. So typically, an adaptive mechanism isused to tune the PSS parameters according to the new operating point.Some researchers tackled this problem by on-line pole shifting andgain settings [1] and by off-line [2] or real-time relocation [3] of the system poles. Others proposed minimum variance (MV)-based tech-niques [4] or used self-tuning regulators [5]. However, these adap- tive techniques need a linear mathematical model of the system andthe identification of its parameters in real-time, which makes thesemethods very time consuming or impossible in certain cases.Recently, controllers based on fuzzy logic were introduced to solvethe problem of intensive computations and inaccurate modeling, andyet they can cater to imprecise inputs. Hsu and Cheng [6] proposed afuzzy PSS that used the deviations in generator speed
(
and accel-eration
(1
as the input signals that were transformed to linguisticvariables using input-membership functions. However, critical expertknowledge was needed for the membership functions and rules of thefuzzy controller. Qiao and Mitzumoto [7] used the fuzzy proportionalderivative (PD) like a controller with adaptive parameters to control adynamic system. The output of this controller was integrated to obtaina fuzzy proportional integral (PI)-like fuzzy controller and then addedto the original output to obtain a fuzzy PID-like fuzzy controller. Tofurther improve the controller response, a parameter regulator basedon an error-peak observer was designed to update two control parame-ters, that would affect the derivative and integral gains. Other adaptivefuzzy logic controllers (AFLC’s) have been proposed in various appli-cations, and the following techniques are illustrative of existing AFLC
Manuscript received March 4, 1997; revised October 14, 1998 and July 6,1999.The authors are with the Department of Electrical and Computer Engi-neering, American University of Beirut, New York, NY10022 USA (e-mail:fuad@aub.edu.lb; skaraki@aub.edu.lb; bassel@cyberia.net.lb).Publisher Item Identifier S 1094-6977(00)00366-7.Fig. 1. Exciter with linear state feedback (LSF) control.
methodologies. Minh and Huang [8] havecombined the popular modelreference adaptive control(MRAC) with FLC’s tocorrect the output of the FLC in order to satisfy model reference output (MRO) specifica-tions.JihongLee[9]suggestedaresettingfactortocorrecttheoutputof the FLC based on another fuzzy model in order to satisfy some desirederror-convergence rate. Rather than correcting the output of the con-troller,LotfiandTsoi[10]tunedthemembershipfunctionsofthefuzzyoutput sets associatedwith the consequent rules of the fuzzy controller.The above three categories require the development of an additionalfuzzy model that is computationally intensive. Jang [11] deployed anadaptive network fuzzy inference system (ANFIS) in order to generatethe membership functions and control rules of a fuzzy controller. Al-though adaptive neuralnetworks (ANN’s)aresuitable in neuro-controlapplications, learning should only occur off line [12].In this paper, we discuss the development of an adaptive PID-likefuzzy PSS that predicts tracking-error divergence and corrects the in-tegral and derivative controller gains for improved regulation of theerror signal while depending less on expert knowledge. The proposedadaptive mechanism does not await the occurrence of an error peak toinitiate the correction of these controller gains. Furthermore, the adap-tive rule used to update these gains is changed progressively as peaksin the error signals are observed. This makes the developed PSS faster,more robust, and portable to different machines. The proposed PSS re-duces the error-settling time and the overshoot of the disturbed gen-eration system and eliminates steady-state error in the various casesof large-voltage disturbances. Simulations have shown that its perfor-mance is superior to that of traditional, existing, and recently intro-duced fuzzy PSS [13]. In the proposed scheme, the AVR response iscontrolled using linear state feedback (LSF) to relocate its dominantpoles according to supplied desired-transient specifications.II. E
XCITATION
C
ONTROL
Several excitation-control schemes are discussed by Ghosh
et al.
[14], and most of them are dismissed either for not being applicableto nonminimum phase systems or for being computationally intensive.Their recommended scheme is a modified version of a pole-placementtechnique called “pole shifting” [15].Given the adopted exciter model (Fig. 1), LSF is used to minimizethe effect of parameter variations on the exciter performance by relo-catingitspoles.Thisisdonetomeetdesiredsecond-order,system-tran-sient, step-response specifications of a settling time less than 0.8 s anda maximum overshoot of 0.05 pu. The desired damping factor
 
andthe natural frequency
can then be calculated assuming a dominantsecond-order system, which yields the two desired pole locations. Thethird real pole has the effect of reducing the maximum overshoot andincreasing the settling time, and it is chosen deep enough in the lefthalf of the complex plane to minimize its effect on the transient part
1094–6977/00$10.00 © 2000 IEEE
Authorized licensed use limited to: IEEE Xplore. Downloaded on April 19, 2009 at 13:34 from IEEE Xplore. Restrictions apply.
 
132 IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICS—PART C: APPLICATIONS AND REVIEWS, VOL. 30, NO. 1, JANUARY 2000
Fig. 2. Input membership functions [6].
of the response. The state feedback-control signal
F
will requirethe design of the gain vector
to force the closed-loop transfer inthe complex plane of the function poles’ locations as desired (Fig. 1),assuming the available measurable state vector
.III. PD-L
IKE
F
UZZY
PSSTheinputsofthe PD-likefuzzy PSSconsistofthedeviationinspeed
(
or error
and the acceleration
(1
or the rate of change of error
(
. Its input-membership functions and decision rules are thoseproposed in [6]. Linguistic variables LN, MN, SN, VS, SP, MP, andLP are used to express large negative, medium negative, small nega-tive, very small, small positive, medium positive, and large positive,respectively. Given numerical values of the two inputs, their member-ship functions with respect to the linguistic variables are then deter-mined using seven different values obtained from Fig. 2. The set of decision rules relating all possible combinations of inputs to outputsis based on previous experience in the field. This set is made up of 49rules expressed using the same linguistic variables as those of the in-puts and is stored in the form of a decision table shown in Table I. Ineach rule composition leading to an output state
linguistic vari-able, the “and” operator is replaced with a “min” operation, and the“or” is replaced with a “max.”The crisp output is calculated using the weighted-average formula,in which each output membership is weighted by the correspondingcoordinate of the center of gravity of each output class. These weightsare given by
=
5
43
3020
302
439
55] 
(1)These weights represent fair or “good” expert values that were se-lected after several trials at different loading conditions. They remainvalid for similar system properties and are not critical when adaptationis introduced. The input–output relationships can be seen in Fig. 3 bydrawing the three dimensional (3-D) plot of the control surface versusthe two inputs of the proposed PD-like fuzzy PSS. This 3-D plot showsa smooth control surface leading to a responsive controller with an ad-equate range of output values.From conventional linear control, it is known that the integral partin the control eliminates the steady-state error of most systems. Thederivative part increases the damping of systems, thus reducing theovershoot and oscillations of the transient response [7]. A system of type zero does not possess an integrator
( 
=
, and thus will have anonzero steady-state error in its response to a step input. Increasing thevalue of the proportional gain can reduce the steady-state error. How-ever, this causes the system to be more oscillatory. Keeping in mind
TABLE ID
ECISION
R
ULES FOR THE
F
UZZY
PSS [6]Fig. 3. Control surface of the PD-like fuzzy PSS.
that integral-action control tends to eliminate the steady-state error, itis recommended that one has a combination of proportional and in-tegral control action at the same time. On the other hand, while re-movingsteady-stateerror,integralactionmayleadtoundesirableoscil-latory-transient response of slowly decreasing amplitude. The deriva-tive control action provides the controller with high sensitivity by pro-ducing an early corrective action before the magnitude of the actuatingerrorbecomesexcessivelylarge.Hence,ittendstoincreasethestabilityof the system and permits the use of larger proportional gain by addingdamping to the system, which results in improved steady-state accu-racy. The derivative control action is seldom used alone and is usuallyused with a proportional or proportional integral action [16].The proposed fuzzy PSS behaves like a PD controller. If the systembeing controlled is of type zero, then a nonzero steady-state error maybeobtained.Theadditionofanintegralcontrolelementwouldimprovethe performance of the fuzzy PD-like PSS but would make it a con-trollerwiththreeinputs(
,
1
,and
)andoneoutput
.Thiswould increase the number of rules by a factor of seven, which wouldlead to a rather complex implementation. The two-dimensional (2-D)decision-rules table (Table I) would then become a 3-D cube assumingsimilar linguistic membership of the integral input (
). Further-more, the rules are based on the previous experience of the operator,who usually monitors the error and change of error but not the errorintegral. Therefore, to keep the task simple, we need a feasible wayto implement a PID-like controller using just the error and its rate of change as inputs. One way to do this [7] is by serially connecting anintegrator at the output of the controller, as illustrated in Fig. 4.Atthisstage,wewillintroduceadditionalparameterstotheresultingPSS inputs and outputs. Let a parameter
 
multiply the second input
1
, thereby emphasizing the derivative part of the controller. Let a
Authorized licensed use limited to: IEEE Xplore. Downloaded on April 19, 2009 at 13:34 from IEEE Xplore. Restrictions apply.
 
IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICS—PART C: APPLICATIONS AND REVIEWS, VOL. 30, NO. 1, JANUARY 2000 133
Fig. 4. Proposed PID-like fuzzy PSS.Fig. 5. Illustration of the adaptation-mechanism sampling.
second parameter
 
multiply the integral output of the PSS, stressingthe integral part of the controller. The output of the original fuzzyPD-like fuzzy PSS becomes
 
1_
(2)where
and
illustrate the proportional-like and derivative-like ef-fect of the fuzzy PSS. The output
of the PSS will then become
1_
 
1_
1_
=
)
1_
w
(3)From (3) above, we can see that when
 
=
, the PSS behaves likea PI controller, and when
 
=
, the PSS behaves like a PD controller.When both
 
and
 
are positive and different from zero, the PSS be-haves like a PID controller, with its inputs being
,
1_
.IV. P
ROPOSED
A
DAPTIVE
PID-L
IKE
F
UZZY
PSSThe relative strength of each control mode of the PID controller (3)depends strongly on the choice of 
 
and
 
. For a small
 
, the inte-gration component is weak, leading to a slow system response. On theother hand, a large
 
leads to an unstable system. The value of 
 
di-rectly affects the derivative part of the controlling signal. It emphasizesthe rate of change of error in controlling the output signal accordingto some desired specifications. The two introduced parameters
 
and
 
are to be changed adaptively in order to meet an optimum response,and a reasonable start would be a relatively larger weight for the inte-gration component. Initially, the integration error is small, and a larger
Fig. 6. Classical PSS block diagram:
1
pu,
 
=1
s,
 
=
s,
 
=
0
s,
 
=
7
s,
 
=
0
s, and
0
.Fig. 7. Proposed adaptive PID-like fuzzy PSS connected to a generator.
integrationconstantwouldcompensatefortheinitiallarge disturbance.This in turn would help reduce the steady-state error. Considering thedifferent parts in (3), it can be seen that decreasing
 
causes a decreasein the integral part. The proportional part also decreases, because it in-cludes a
 
term. This has the undesirable effect of slowing down thesystem-response control to the error signal. In the proportional part,
 
is coupled with
 
, so that if 
 
increases at the same rate as
 
decreases,the proportional part will remain unchanged, and thus, a fast responseto the error will be maintained. Increasing
 
will primarily enlarge theeffectofthederivativepartofthecontrolsignal(whichisadvantageousto the system), because the resistance against overshoot and oscillationtimewillincrease.Motivatedbythisidea,theauthorshave[7]designeda parameter-adaptive, PID-like fuzzy controller.The adaptation of the parameters
 
and
 
is done at the detectedpeaks of the output error signal. They used a peak observer of thesystem output to measure the absolute value of the peak 
 
p
measuredat the
th sample. This signal is then used to adjust
 
and
 
simul-taneously at each peak of the output-error signal. The algorithm usedfor adjusting the parameters is simple and consists of dividing the ini-tial derivative gain
 
by the peak value. Simultaneously, it multipliesthe integral gain
 
by the same peak value
 
p
. The equation of theparameter regulator can be written as follows:
 
 
 
p
and
 
 
 
p
(4)where
 
and
 
are the initial values of 
 
and
 
(designer choice),and
is the index corresponding to the error-peak sample time. Theseinitial values are not critical and will produce acceptable results due totheir continuous adaptation.The delay produced while waiting for a peak in the output-errorsignal, in order to update the elements
 
and
 
, is not suitable forpower systems and can lead to undesirable results. Instead, a modifiedmethod is proposed to evaluate the magnitude of the frequency errorat several intervals, with evaluation rates being varied according to theinterval position. So between time zero and the first error peak, thisrate is high to follow up a rapidly changing error. The evaluation rate isdecreased progressively after each peak until the third peak is reached.
Authorized licensed use limited to: IEEE Xplore. Downloaded on April 19, 2009 at 13:34 from IEEE Xplore. Restrictions apply.
of 00

Leave a Comment

You must be to leave a comment.
Submit
Characters: ...
You must be to leave a comment.
Submit
Characters: ...