Sie sind auf Seite 1von 7

ELSEVIER

Electrical Power & Energy Systems, Vol. 20, No. 3, pp. 213-219, 1998
1998 Elsevier Science Ltd. All rights reserved
Printed in Great Britain
PII: S0142-0615(97)00049-5 0142-0615/98/$19.00+0.00
A s e l f - t u n i n g generalized
predictive power system
stabilizer
J A L Barrei ros
UFPA, CT, DEE, 66059 Bel6m PA, Brazil
A S e Si l va* and A J A Si m6es Costa
GSP, UFSC, CTC, EEL, 88 040-900 Florian6polis, Brazil
This paper presents a new self-tuning stabilizer f or enhance-
ment of the dynamic stability in power systems. An adaptive
control algorithm tracks the system operating conditions by
using a least-squares identification technique and the con-
troller is synthesized via a generalized predictive control
scheme. The stabilizer proposed is evaluated by computer
simulation of a single-machine system and the influence of its
main design parameters is analysed. The same scheme is
then applied to a muhi-machine equivalent of the Southern
Brazil power system. The results obtained in the simulation
tests f or both systems indicate that the proposed scheme can
be applied to synthesize power system stabilizers. 1998
Elsevier Science Ltd
Keywords: adaptive control, power system stabilizer,
generalized predictive control
I. I nt roduct i on
The function of a power system stabilizer (PSS) is to
improve the dynamic stability of power systems by provid-
ing damping torque to the generators. The PSS is designed by
using a linearized model of the power system around an
operating point [ 1,2]. However, the large variations that may
occur in a power system, because of load drift or switchings,
can affect its performance. An adaptive stabilizer, using
parameter estimation, can cope with these variations by
synthesizing a controller updated for the actual operating
condition.
Several adaptive techniques have been proposed for the
synthesis of PSSs, most of' them based on pole shifting (PS),
Recei ved 19 No ve mbe r 1996; r evi sed 20 Mar ch 1997; accept ed 29
Ma y 1997
* Cor r espondi ng aut hor
generalized minimum variance (GMV) of linear quadratic
Gaussian (LQG) control [3-8].
These techniques have been successfully tested by
simulation, and field tests were reported in [9,10]. However,
other adaptive control techniques found in the literature can
deal advantageously with some specifities of the application
to power systems.
In this work, the technique employed for synthesis of the
stabilizer is the generalized predictive control (GPC) [11],
which has been applied very successfully to the control of
other industrial processes [12]. The proposed technique is
capable of stabilizing unstable and inverse unstable plants
[11], a commonly encountered situation in applications of
digital power system stabilizers, since the plant transfer
function usually has a pair of low damped or unstable
poles and the relative degree of this function is larger than
2 [13]. This can cause the existence of zeros outside the unit
circle on the z plane, due to the short sampling periods used
in this application [14].
The self-tuning generalized predictive power system
stabilizer (STGPS) proposed in this paper uses a predictive
model of the system, the parameters of which are estimated
by an algorithm based on the recursive least-squares method.
The controller is synthesized via the optimization of a
function including the plant input u(t) and output y(t) over
specified future horizons. The scheme is of an indirect type
[15] with the stabilizer signal being applied to the voltage
regulation loop of the synchronous machine as represented
in Figure 1.
The algebraic development of the proposed method is
presented in Section II. Section III evaluates its performance
when applied to single- and multi-machine power systems.
The multi-machine system considered is a seven-bus, five-
machine equivalent of the Southern Brazil network. A
comparison with self-tuning power system stabilizers
213
214
Sel f-tuni ng general i zed PSS: J. A. L. Barreiros et al
_P . . . . sy,te~. 77]
I I
. . . . . S_T_GP S_ . . . .
Fi gure 1. Sc he me of t he pr opos e d sel f -t uni ng power
s y s t em st abi l i zer
based on pole shifting is carried out for this system. The
conclusions are presented in Section IV.
I I . A l g e b r a i c f o r m u l a t i o n
As indicated in Section I, the stabilizer proposed in this paper
is of an indirect type and composed of two parts: system
identification and controller synthesis. They are developed
in this section with emphasis on the synthesis of the con-
troller by a generalized predictive control technique.
I1.1 Identification
The technique used for the system identification is based on a
recursive least-squares parameter estimation algorithm,
which is a well known method [16]. Therefore, only the
results are presented, without demonstrations.
The system is represented by the following CARMA
(controlled auto-regressive and moving average) model
A ( q - 1)y(t) = B( q - 1 )u(t - 1) + T(q - ])~(t) (1)
with
A( q - 1) = 1 +a l q -1 + . . . +a~oq - " (2)
B ( q - l ) = b 0 + b l q - 1 + ... + b, bq -rib (3)
T( q- 1) = 1 + t l q - 1 + ... + t , , q - " ' (4)
where the order nb has to be large enough to take into account
possible zero first terms due to non-zero dead time. The
polynomial T may either represent the external noise com-
ponents affecting the output (in which case its coefficients
have to be estimated) or a design polynomial interpreted as a
fixed observer for the prediction of future outputs [11]. The
signals u(t) and y(t) are the input and output of the system,
respectively, and e(t) is a discrete white-noise sequence.
If Tis a fixed observer, it is used as a filter for the input and
output signals and a least-squares algorithm may be applied
to obtain only the ai ' s and bi' s coefficients. If T is to be
estimated, the residuals of the estimation can be used as a
measure of the unobservable white-noise sequence and a
generalized least-squares method [17] may be used. A
general formulation is presented in the sequel.
Given the vector 0 of parameter estimates by
O(t) = [al . . . . . an,, bo . . . . . b,b] T (5)
and x(t), the measurement vector, by
x(t) = [ - y ( t - 1) . . . - y ( t - na) u ( t - 1) . . . u ( t - n b - 1)] T
(6)
the update of the estimates is obtained by [15]
0(t) = 0(t - 1) + K(t)[y(t) - xT(t)0(t -- 1)] (7)
P(t - 1)x(t)
K(t) = P(t)x(t) = [/3 + xT(t)P(t -- 1)x(t)] (8)
P(t) = [I - K(t)xT(t)] P( t ~ 1) (9)
where P(t) is the covariance matrix, I is an identity matrix,
K(t) is the vector of adjustment gains and/3 is the forgetting
factor, used to progressively reduce the effects of old
measurements to the benefit of the most recent ones,
maintaining the capacity to follow variations in the system
[15].
11.2 Generalized predi cti ve cont rol l er
The generalized predictive controller is a long-range
predictive controller proposed to retain the design flexibility
and performance of earlier methods such as generalized
minimum variance and pole assignment, but which also
exhibits improved robustness against incorrect a priori
choices of the design parameters [11]. From a parametrized
system model, GPC predicts future outputs, which depend on
both currently available input/output data and on present and
future control values. The latter are determined in order to
minimize a function including predicted values of output and
input variations over pre-specified horizons.
The GPC is usually developed for a CARIMA (controlled
auto-regressive integrated moving average) model; that is,
the incremental approach which is used in [11]. In this
approach the cost function to be minimized is expressed as
a function of the control increment z~u(t) and not of the
control signal u(t). In the application to the problem of power
system stabilization, after the occurrence of a perturbation in
the system, a constant signal u(t) would result as the output of
the PSS, interfering with the voltage regulation. The results
of a test in a single-machine infinite busbar system illustrates
this point. Figure 2 shows the system output and the PSS
output following a short circuit at the terminal of the
generator. The PSS is based on an incremental GPC with
design parameters as defined in the next section. Although
the post-fault oscillations are highly damped, as shown in
Figure 2, the control signal has a non-zero steady-state value.
This off-set is not allowed in PSS applications.
In order to avoid interference with the voltage regulation
loop, the application to the design of PSSs requires the
adaptation of this formulation to the positional control
approach, so that a CARMA model can be utilized [18].
The cost function to be minimized is therefore defined by
J ~ pc( U, t ) =E Y 2 ( t + i ) + Z k ( t + i - 1 ) u Z ( t + i - l )
i = 1
(10)
where Ny is the prediction horizon, Nu is the control horizon
and k(j) is a weighting sequence over future control. The set-
point pre-specified trajectory [11] is not considered for PSS
synthesis.
The weighting sequence is usually defined as a constant
value, k(]) = k, f or j = t ..... t + N. - 1, and takes an infinite
value for j ~ t + N.. Therefore, future control values for
j -- t + N. will be zero; that is, u(t + j ) = O,j >-- N. , as desired
for the PSS implementation given in Figure 1.
The cost function given by equation (10) includes a sum
Sel f-tuni ng generalized PSS: J A L Barreiros et a l 215
y( t) ~.o,
DD
- 2 . 5 2
- 5 . D i i i ! I i i ! !
u ( t) o.a 1
'
Q[] ' .
:! :I! ! ' . . . . . . . .
10.0 15.0 2110
t i m e (sec)
Fi gur e 2. Test wi t h PSS usi ng GPC wi t h i ncr ement al
appr oach: (a) syst em o ut p ut y(t); b) PSS out put
si gnal u(t). (N v = 5, No = 2, k = 0 and speed as t he
PSS i nput )
of expected values of y (predictions of y). These predictions
can be separated into "f ree response" predictions, which
are independent of the future control inputs, and "f orced
response" predictions, dependent on the future control
inputs. This separation of the predictions y facilitates
minimization of the cost function, as shown in the sequel.
Define the vector f, with dimension Ny, formed by the
"f ree response" predictions,
f = [.9(t + llt),~(t + 210 . . . . . ~ ( t + N yl t ) ] T (11)
i.e. the predictions of ~( t + k), k = 1 ..... Ny, given { u( s - 1),
y( s) ; s <- t} , assuming that { u( t + k) = O, k >- 0}. Next define
the vector of future control values fi, with dimension N,
(recall that u( t + j ) = 0; j -- Nu),
0 = [u(t), u( t + 1) . . . . . u(t + N, - 1)] T (12)
and define the vector of predicted controlled plant outputs #,
with dimension Ny,
30 [.9(t + 1),9(t + 2), . . . , ~ ( t + N y ) ] x (13)
By using these vectors, the predictions can be represented by
the following matrix equation [11]
# = Gf i +f (14)
where
G=
go 0 " " 0
gl go "'" 0
gN. - 1 gN. - 2 " ' " go
gu y - 1 guy - 2 "'" gUy - N,
(15)
The term GQ in equation (14) corresponds to the "f orced
response" predictions The elements gi of the matrix G, with
dimensions Ny N , , are points of the plant' s step response
and can be computed recursively from the model, equation
(1), assuming zero noise and a constant unit control input.
Also the "f ree response" predictions f i t + i), which are
components of the vector f, can be computed for all i, by
iterating the plant model (1), assuming e(t + 1) = 0 and that
future control values will be zero.
The quadratic minimization of equation (1) now becomes
a direct problem of linear algebra, with
JGPC = ~T~ ~kl~Tl~ (16)
and the optimal solution of the future control vector ti is [ 11 ]
fi* = - (GTG + M) - 1GTf (17)
The matrix inversion involved in equation (17) is of
dimension Nu N , , with N, being usually a low value
Because GPC is a receding-horizon control policy, only the
first element of ti* is actually applied. Therefore it is only
necessary to calculate the first row of (GTG + M) - I G T, at
each sampling interval.
I I I . S i m u l a t i o n r e s u l t s
Two test systems are considered for the evaluation of the
STGPS synthesized through the method developed in the
previous section: the first is a single-machine infinite bus
system; the second is a multi-machine system representing
an equivalent of the Southern Brazil power system [13].
The first example serves to determine adequate values for
the main design parameters of the proposed stabilizer; that is,
the prediction horizon Ny, the control horizon Nu, the
weighting sequence ~,(]) and the sampling period ts. The
second one shows the potential of the method for larger
systems. In both examples, the deviations of the machines'
electric power are used as input signals y( t ) to the stabilizers.
A pseudo-random binary sequence (PRBS) of magnitude
0.001 pu is added to the PSS signal u(t) to improve the
estimation. This injected noise caused variations of less than
1% in the machine terminal variables. The estimation is
" f rozen" for a few seconds from the onset of the faults to
avoid estimating parameters during the post-fault fast
transients. All the simulations are carried out using the non-
linear models of the test systems, the data of which are found
in Appendix A and Appendix B. To avoid excessive inter-
ference of the PSSs on the voltage control, limits of _ 0.1 pu
are considered on the outputs of all PSSs in the tests
presented in the sequel. Two filters were added, as shown
in Figure 1: a "wash-out" filter to avoid a voltage off-set
caused by a permanent deviation of the PSS input signal and
an anti-aliasing filter. The latter is a second-order Bessel
filter with a cut-off frequency tuned to ensure at least a 10 dB
attenuation at the Nyquist frequency.
I I I . 1 Single-machine i nfi ni te busbar system
The main design parameters of the method were evaluated
with this system. All data for the one-machine infinite busbar
are found in Appendix A. Tests were carried out using the
simulations following the occurrence of faults. These faults
are short circuits at the generator terminal bus with zero
impedance and clearing time of 100 ms.
For the estimation, the values chosen were na = 3, nb = 4
and nt = 2 (if Tis to be estimated) or T = (1 - q q - 1)2, with
tl between 0.0 and 0.5 (if T is specified as an observer
polynomial). In the latter case, tl could be adjusted to filter y
216 Self-tuning generalized PSS: J. A. L. Barreiros et al
Y
o. * * * Id , 1 5= . L . , , . . . , . . . . . . . . . . . , . ,
, , , ' ' ' , , 0
O . 0 0 ' ~ . - ] b )
I ' ' ' ' 1 ' ' ' ' 1 ' ' ' ' l ' ' ' ' l
0 . 0 2 . 5 . 5 . 0 7 . 5 t O . O
U O . 0 0 2 -
O. OOt -
O. 0 0 0
c)
I J , , . . . . . . . . . . . . . . .
. . . . I . . . . I . . . . I . . . . I
o . o = . ~ o . o 7 . ~ s o . o
f r e q . ( H z )
Figure 3. Spectra of the PSS i nput and output
s i g n a l s : (a) signal y before parameter adjustment;
(b) signal yaf t er parameter adj ustment; (c) signal u
a f t e r parameter adj ustment
and u in order to improve the robustness of the system [19]
although, for the positional control approach considered in
the paper, this filtering is not as critical as it would be for an
incremental control [20]. By using these values, good
estimations were obtained, with low variance for all signals.
Figure 3 shows the spectrum of the output signal y at the
beginning of the estimation when the model parameters,
given by equation (5), are still in the process of adjustment
and after they are adjusted (curves a and b, respectively).
These curves emphasize the damping added by the proposed
stabilizer. The figure also shows the spectrum of u after the
adjustment, when the stabilizer is already tuned. The spec-
trum of the tuned PSS signal u (curve c) exhibits suitable
tuning for the attenuation of the plant' s dominant mode of
oscillation (with a frequency around 1 Hz). The signal
energy decays quickly for higher frequencies, indicating
that there is no interference with high-frequency modes.
This contrasts with the spectrum of y(t), before the adjust-
ment, which shows components of high magnitude around
1 Hz, the frequency of the system' s dominant mode.
Concerning the controller parameters, the prediction
horizon Ny wa s varied from 2 to 10, using N, = 2 as the
base value for the control. This latter value is suggested in
[11], whenever there is a pair of dominant low damped plant
poles. The values of the remaining parameters adopted were
X = 0 and ts, the sampling period, equal to 100 ms. A
progressive improved performance has been noticed up to
ivy ----- 5. No perceptible additional improvement is verified
for larger values, so that Ny ---- 5 has been adopted for
simplicity. Then, maintaining the values of the other para-
meters, N, was varied from 1 to 5, having the best results
been observed for Nu = 1 and 2. The value N, = 2 has been
t - y ( t ) ( o o e n - l o o o , I n o u )
o - - - ~ ( ~ - , ~ - - - -
- i i i i i I i 1 ~ l
i -
y ( t ) ( a n p u )
b ) o - ~
- I L I i i I ~ i ~ I
0 , 2 "
c ) - . , ~ u ( t ) ( i n l a u )
0 . 0 - "~1/~ "
-,3..2 , L , I I = ' '
1 0 1 5 2 0
T, lme ( l n s e c , }
F i g u r e 4 . S i m u l a t i o n r e s u l t s o f t h e o n e - m a c h i n e
i n f i n i t e b u s b a r s y s t e m w i t h t h e p r o p o s e d s t a b i l i z e r :
( a ) o u t p u t s i g n a l y ( t ) ( i n o p e n - l o o p ) ; ( b ) y ( t ) w i t h
S T G P S ; ( c ) c o n t r o l s i g n a l u ( t )
selected. With Ny ~--- 5 and N, = 2, X was varied for the values
0.001, 0.01, 0.1, 1.0 and 10.0. A deterioration beginning
from the value X -- 1 was noticed, owing to the increased
weighting over the control signal. A positive value near zero
(X = 0.001) was adopted to avoid possible problems with the
matrix inversion in equation (17). Finally, the sampling
period ts was increased and the performance began to
deteriorate. For low values, for instance ts < 50 ms, even
when increasing Ny the performance also tends to degrade,
since for fast sampling, more parameters would have to be
used in the estimated model. Since the estimator and the
controller in an adaptive scheme should work together in a
synergistic way, some of the design parameters which have
greater influence on both parts of the scheme have to be
chosen jointly. Therefore, ts = 100 ms was maintained as an
adequate base value. Figure 4 presents the results of a short-
circuit simulation, with the following base values assumed
for the design parameters: Ny = 5 , N, = 2, X = 0.001, ts =
100 ms , n a ---- 3, nb = 4 and nt = 2. Good performance is
shown with fast damping of the electromechanical oscilla-
tions. It is also shown that, by using the positional approach,
there is no permanent deviation in the control signal.
Next, tests were carded out by considering the con-
secutive application of three short circuits in the power
system. The first one is cleared after 150 ms without any
switching and therefore without variation of the operating
point. The second fault is cleared after 100 ms, by the outage
of one of two parallel lines, leading to a new operating point.
The third fault has a duration of 80 ms and is cleared without
any switching.
Figure 5 shows the system performance without PSS
(curve a), with a continuous fixed-parameter PSS (curve b)
and with the proposed digital adaptive stabilizer (curve c).
The fixed-parameter PSS was designed using the linearized
system equations before the first fault is applied. This
classical design ensures a minimum damping of 20% for
the dominant poles while keeping the larger damping of
the remaining poles [2]. The proposed digital adaptive
self-tuning stabilizer (STGPS) was designed by using the
base values previously determined.
The performance for both PSSs, presented in curves b and
c, is similar to that for the first fault, when no variation of the
operating point takes place. For the second fault, the adaptive
PSS estimator, after a "f reezi ng" period of 3 s, adjusts the
system parameter and a good damping is again obtained.
However, the fixed-parameter PSS is unable to obtain a
Sel f-tuni ng generalized PSS: J A L Barreiros et a l 217
t o -
1
t | I "!
e , 5 0 1
r l 0 , , ,
gT
le 100' 7
S O i ' ~ , "
i |
0
n I ' J ' ' t ' ' ' ' ' I g ' ' ' I ' ' ' ' I ' ' '
meter E S P
' I ' ' ' ' I ' " " ' ' ' I ' ' ' ' I . . . . I ' ' '
, E S P
1 0 2 0 3 0 4 0 S O 6 0
t i m e ( s e c ]
Figure 5. Simulation results considering variation of the operating point: (a) wi thout stabilizer; (b) wi th fixed-
parameter stabilizer; (c) wi th the proposed adaptive stablizer
similar performance. After the occurrence of the third fault,
the advantage of the adaptive stabilizer, with its parameters
completely adjusted, is still more evident.
These results corroborate the fact that, although a fixed-
parameter PSS can improve the damping for several
operating points of the system, it cannot ensure the same
performance for all these conditions. Self-tuning stabilizers,
owing to the self-adjustment capability of their estimators,
can maintain a high pertormance, despite variations in the
system conditions.
111.2 Mu l t i - mac h i ne s y s t e m
The multi-machine system considered is shown in
Figure 6. This system is a simplified model of the Southern
Brazil power system and was proposed as a benchmark
system for small-signal electromechanical stability studies
[13,21].
The system is a slightly modified seven-bus equivalent of
the model used in the initial studies related to the Itaipd
power plant and the corresponding AC transmission
network. The Itaipd ;generator is connected to the
Brazilian Southeastern ,;ystem (which is represented by
a static load together with a large synchronous motor)
through a series-compensated 756 kV line. An intermediate
765 kV bus is connected to a 500 kV transmission ring
containing three other hydro stations: Salto Santiago,
Salto Segredo and Foz do Areia. Simple first-order
models were used for the AVRs (automatic voltage
regulators) of all machines. The loads are of the constant-
impedance type. All s;ystem data and the operating
point considered (a heavily loaded condition presenting
difficulties for stabilization of the system) are given in [13]
and repeated in Appendb: B.
The system open-loop poles associated with the electro-
mechanical modes are: - 1.801 ___j9.176, -2. 013 - j9.168,
0.646 ---j5.392 and - 0.222 - j 5. 877. These values indicate
an unstable mode and a low damped mode. Through a
computer package based on modal analysis techniques
[22], the generators at buses 4 and 3 (Itaipd and Salto
Segredo, respectively) have been selected as adequate sites
to locate the stabilizers. The parameter estimation obtained
for both self-tuning stabilizers resulted in a model which is
unstable and inverse unstable for a sampling period of
100 ms. For instance, for the estimator of the PSS located
at the generator of Itaip6, choosing n a = 3 and nb = 4, the
model obtained is:
--=Y(t) (O' 025-0"703q-l-O' 158q-2+O' 612q-3+O' 353q-4)q ' (18)
u(t) 1 - 1. 102q - 1 _ 0. 036 q - 2 + 0. 5 7 2q 3
which has, in the z plane, poles 0.85 --- j0.55, - 0.59, 0 and 0,
and zeros - 0.63 - j0.26, 1.07 and 28.31.
Next, simulations were carded out for a short circuit at bus
3, with a fault impedance of 0.01 + j0.05 pu and successful
recovery after 87 ms. Figure 7 presents the results of
the simulations for the cases: (a) without stabilizers; (b)
with the STGPS proposed in the paper applied at the
specified generators, with the base parameter values of the
previous example; and (c) with self-tuning stabilizers based
on pole shifting [3] applied to the same generators. All the
parameter values used in case (c) are the same as those used
in case (b), with the exception of the controller parameters,
where the rule used was to shift radially all the poles to the
origin of the z plane in order to obtain a minimum of 20%
damping for the plant' s dominant poles. From these tests it
can be concluded that the proposed stabilizer is able
4 G 7

S A L T O S C G n E D O
Figure 6. Single-line diagram of the reduced
equivalent of the Southern Brazil power system
218
Sel f - t uni ng gener al i zed PSS: d. A. L. Barrei ros et al
141P
81
g , l
4 1 1 , , , , , , , , , , , I , , , , , i
O I ~ 3 6
t i me ( s e e )
t l 6
tO|
I t
U
Q
M I N I
D~
m
mN
II
q
O
6P
U
U
J
L 3
D e
l e
i | o
IM
l e o
Q
i0
e~
I
b)
I I I I I
I I I 1 I I I t l l
t t m e l o m e l
c)
2
i I l ! I I
~o oa . oo as ~o
t i m e ( h e )
Fi gur e 7. S i mu l a t i o n r esul t s of t h e S o u t h e r n Brazil
s y s t e m ( bus 7 as r ef er ence) : (a) w i t h o u t st abi l i zer s;
(b) wi t h t he p r o p o s e d st abi l i zer ; (c) wi t h st abi l i zer s
based on pol e shi f t i ng
to achieve results comparable to those obtained with
self-tuning pole-shifting stabilizers, without risk of over-
parametrization, which could be a problem for this method
[111.
Tests were also carried out for other operating conditions
and f or several pertu rbations su ch as vari ati ons o f the vol t ag e
reference and loss of loads. In all tests the proposed stabilizer
quickly damped out the electromechanical oscillations of the
system.
I V . C o n c l u s i o n s
A new adaptive power system stabilizer based on
generalized predictive control was evaluated in this paper.
This approach presents several advantages. Few parameters
must be adjusted, leading to simplicity of the design. The
technique can stabilize unstable and inverse unstable
systems, which suits the problem of power system stabiliza-
tion. Adequate values for the design parameters, obtained
from simulations in a single-machine infinite bus system, are
suggested. The proposed stabilizer was successfully tested
by simulation on an equivalent of a real power system, using
those parameters.
V . A c k n o w l e d g e m e n t s
The authors gratefully acknowledge the financial support of
the Brazilian Government agencies CNPq and CAPES. In
addition, Dr Barreiros is grateful for the support of the
Federal University of Pardi (Brazil).
V I . R e f e r e n c e s
1. de Mello, F. P. and Concordia, C., Concepts of synchronous machine
stability as affected by excitation control. IEEE Trans. on Power
Apparatus and Systems, 1969, 88(4), 316-327.
2. Larsen, E. V. and Swann, D. A., Applying power system stabilizers--
Parts 1/3. IEEE Trans. on Power Apparatus and Systems, 1981, 100,
3017-3046.
3. Cheng, S., Chow, Y. S., Malik, O. P. and Hope, G. S., An adaptive
synchronous machine stabilizer. 1EEE Trans. on Power Systems, 1986,
1(3), 101-109.
4. Seifi, H. and Hughes, F. M., Adaptive power system stabilizer using
a bang-bang pole placement strategy. Int. J. Control, 1990, 51(1),
33-50.
5. Barreiros, J. A. L., Silva, A. S. e. and Sim6es Costa, A. J. A., A self-
tuning power system stabilizer based on an implicit pole-placement
method. In Ninth Brazilian Congress on Control and Automation, July
1992, pp. 263-267 (in Portuguese).
6. Chandra, A., Malik, O. P. and Hope, G. S., A self-tuning controller for
the control of multimachine power systems. IEEE Trans. on Power
Systems, 1988, 3(3), 1065-1071.
7. Gu, W. and Bollinger, K. E., A self-tuning power system stabilizer for
wide-range synchronous generator operation. IEEE Trans. on Power
Systems, 1989, 4(3), 1191-1199.
8. Mao, C., Malik, O. P. and Hope, G. S., An adaptive generator excitation
controller based on linear optimal control. 1EEE Trans. on Energy
Conversion, 1990, 5(4), 673-678.
9. Malik, O. P., Mao, C. X., Prakash, K. S., Hope, G. S. and Hancock, G. C.,
Tests with a microcomputer based adaptive synchronous machine
stabilizer on a 400 MW thermal unit. 1EEE Trans. on Energy Conver-
sion, 1993, 8(1), 6-12.
10. Norum, W. E. and Bollinger, K. E., Lab and field tests of a self-tuning
power system stabilizer. IEEE Trans. on Energy Conversion, 1993,
8(3), 1017-1023.
11. Clark, D. W., Mohtadi, C. and Tufts, P. S., Generalized predictive
control--Parts 1/2. Automatica, 1987, 23(2), 137-160.
12. Clarke, D. W., Application of generalized predictive control to
industrial processes. IEEE Control System Magazine, 1988, 8, 49-55.
13. Martins, N. and Lima, L. T. G., Eigenvalue and frequency domain
analysis of small-signal electromechanical stability problems. In IEEE
Symposium on Applications of Eigenanalysis and Frequency Domain
Methods, Special Publication 90TH0292 PWR, IEEE, Piscataway, NJ,
t~P. 17-33, 1989.
14. Astr6m, K.J. and Steinby, S., Zeros of sampled systems. Automatica,
1984, 20(1), 31-38.
15. Astrom, K. J. and Wittenmark, B., Adaptive Control. Addison-Wesley
Company, Reading, Massachusetts, USA, 1989.
16. Young, P. C., Recursive Estimation and Time-Series Analysis. Springer
Verlag, Berlin, 1984.
17. Solo, V., Some aspects of recursive parameter estimation. Int. J.
Control, 1980, 32, 395-410.
18. Barreiros, J. A. L., Sim6es Costa, A. J. A. and Silva, A. S. e., GPC power
system stabilizer using a positional control approach. Internal Report,
PG/EEL/UFSC, 1994.
19. Robinson, B. D. and Clarke, D. W., Robustness effects of prefilter in
generalized predictive control. Proc. lEE, Pt. D, 1991, 138(1), 2-8.
Sel f-tuni ng generalized PSS: J A L Barreiros et a l 219
20. Clarke, D. W. and Mohtadi, C., Properties of generalized predictive
control. Automatica, 1989, 25(6), 859-875.
21. Martins, N., Macedo, N. J. P., Bianco, A., Pinto, H. J. C. P. and Lima,
L. T. G., Proposal for a benchmark system for power system oscillation
analysis and control. In C1GRE Colloquium on Power System Dynamic
Performance, Florian6polis, Brazil, 1993, pp. 1 - 16.
22. Martins, N. and Lima, L. T. G., Determination of suitable locations for
power system stabilizers and static vat compensators for damping
electromechanical oscillatios in large scale power systems. IEEEE
Trans. on Power Systems, 1990, 5, 1455-1469.
23. Arrillaga, J., Arnold, C. P and Harker, B. J., Computer Modelling of
Electrical Power Systems. John Wiley and Sons, New York, 1983.
Appendi x A Data of t he one-machi ne i nfi ni te
busbar system
This system consists of one machine connected by a double-
circuit tie-line to an infinite busbar. The total tie-line
resistance, reactance and shunt susceptance are given by:
R e : 0.02 pu, X e : 0.415 pu and Be = 0.0 pu, respectively.
The system operating conditions are given by V~ =
1.0 PU/-0 , Pe = 0.75 pu and Qe = 0.10 pu, where V~, Pe
and Qe are the infinite busbar voltage and the active and
reactive powers at the generator terminal bus, respectively.
The machine non-linear model represents a salient-pole
generator including damping windings [23]. The parameters
are shown in Table A1. The reactances are in pu and the other
parameters are in seconds.
The excitation system is represented by a first-order model
with a gain Ks = 200, and a time constant T~ = 0.05 s. The
limits on the excitation voltage are given by Vi i m : -4- 6.0 pu.
Table A1 . One-machine system: generator parameters
H 4.27 s Xd 1.445
Xd 0.316 Xq 0.959
X~' 0.179 X~' 0.162
Rs 0.001 Tdo 5.256
T~ 0. 0282s T~ 0.157
Appendi x B Data of 'the mul t i - machi ne system
The multi-machine system data are presented in this
appendix. The frequency is 60 Hz and the power base is
1000 MVA. The bus data and the line data are presented in
Tables A2 and A3, respec~fively. The bus shunt reactors and
the line susceptances were combined and presented as bus
data in Table A2.
The generator data are presented in Table A4. The
reactances are in pu anti the remaining parameters are
in seconds. The excitation system for all generators
is represented by a first-order transfer function with gain
Ka = 30 and time constant Fa = 0.05 s. The limits on the field
voltage ar e Vii m : ~ 5. 0 pu.
Table A2. Mu lti-machine system: bu s data
Bus Voltage
Generation
MW MVAr
1 1.03 24.5
2 1.03 27.2
3 1.029 26.6
4 1.039 48.5
5 0.998 21.2
6 0.989 21.4
7 0.966 0.0
1658.0 - 412. 0
1332.0 -200. 1
1540.0 -446. 5
6500.0 1958.6
-3164. 0 952.7
Load
Bus MW MVAr Shunt (pu)
1 2405.0 - 467. 0 0.1792
2 692.3 - 184. 0 0.1792
3 688.2 - 235. 0 0.1142
4 62.6 24.3 0.0368
5 845.8 - 9. 2 0.0330
6 - 4 . 9 79.8 2.142
7 2884.0 - 196.0 0.0420
Table A3. Mu lti-machine system: line data
Initial bus Final bus R (pu) X (pu)
1 3 0.003 0.038
2 3 0.005 0.076
4 6 0.0029 0.0734
5 1 0.019 0.245
5 2 0.015 0.225
6 5 0.0 0.039
6 7 0.004 0.057
Table A4. Mu lti-machine system: generator parameters
Generator bus
1 2 3 4 7
MVA 1900 1400 1944 6633 6000
H 4.5 4.5 4.5 5.07 5.0
Xd 0.85 0.85 0.88 0.90 1.00
Xq 0.70 0.70 0.69 0.68 0.70
X d' 0.30 0.30 0.30 0.30 0.30
Xd" 0.20 0.20 0.20 0.24 0.25
Xq" 0.20 0.20 0.20 0.27 0.25
Tdo' 5.0 5.0 5.0 7.6 8.0
Tdo" 0.053 0.053 0.06 0.09 0.09
Tqo" 0.123 0.123 0.09 0.19 0.2

Das könnte Ihnen auch gefallen