Electrical Power and Energy Systems 29 (2007) 679–689

www.elsevier.com/locate/ijepes

Comparison of intelligent fuzzy based AGC coordinated

PID controlled and PSS controlled AVR system
a,* b
V. Mukherjee , S.P. Ghoshal
a
Department of Electrical Engineering, Asansol Engineering College, Asansol, West Bengal, India
b
Department of Electrical Engineering, National Institute of Technology, Durgapur, West Bengal, India

Received 15 September 2006; accepted 24 May 2007

Abstract

This paper attempts to investigate the performance of intelligent fuzzy based coordinated control of the Automatic Generation Con-
trol (AGC) loop and the excitation loop equipped with Proportional Integral Derivative (PID) controlled Automatic Voltage Regulator
(AVR) system and Power System Stabilizer (PSS) controlled AVR system. The work establishes that PSS controlled AVR system is
much more robust in dynamic performance of the system over a wide range of system operating configurations. Thus, it is revealed
that PSS equipped AVR is much more superior than PID equipped AVR in damping the oscillation resulting in improved transient
response.
The paper utilizes a novel class of Particle Swarm Optimization (PSO) termed as Craziness based Particle Swarm Optimization
(CRPSO) as optimizing tool to get optimal tuning of PSS parameters as well as the gains of PID controllers.
For on-line, off-nominal operating conditions Takagi Sugeno Fuzzy Logic (TSFL) has been applied to obtain the off-nominal optimal
gains of PID controllers and parameters of PSS. Implementation of TSFL helps to achieve very fast dynamic response.
Fourth order model of generator with AVR and high gain thyristor excitation system is considered for PSS controlled system while
normal gain exciter is considered for PID controlled system. Simulation study also reveals that with high gain exciter, PID control is not
at all effective. Transient responses are achieved by using modal analysis.
 2007 Elsevier Ltd. All rights reserved.

Keywords: AGC; AVR; CRPSO; Electromechanical local mode of oscillations; PID; PSS; TSFL

1. Introduction ACE ¼ ðDP tie þ bDf Þ ð1Þ

Zeroing of the Area Control Error (ACE) of each area The Proportional Integral Derivative (PID) controlled
of an interconnected multi-area thermal generating stations ACE over a given time interval s in Laplace domain is
is the main aim of Automatic Generation Control (AGC), defined by, (G(s)ACE(s)) where G(s) is as in (2).
so that the scheduled values of the system frequency and
Ki
tie-line error is maintained. Deviations in system frequency GðsÞ ¼ K p þ þ K dS ð2Þ
and tie-line power exchange are the two control variables, s
which are associated with AGC scheme. These two vari- In large interconnected power systems, inter-area oscil-
ables give an idea about ACE and they are related as in (1). lations (0.2–1.0 Hz) are inherent. For ensuring system sta-
bility, particularly the small signal stability phenomenon,
*
Corresponding author.
Power System Stabilizer (PSS) in Automatic Voltage Reg-
E-mail addresses: vivek_agamani@yahoo.com (V. Mukherjee), ulator (AVR) loop provides stabilization action to small
spghoshalnitdgp@yahoo.com (S.P. Ghoshal). disturbances that occur continually in the system. This

0142-0615/$ - see front matter  2007 Elsevier Ltd. All rights reserved.
doi:10.1016/j.ijepes.2007.05.002
680 V. Mukherjee, S.P. Ghoshal / Electrical Power and Energy Systems 29 (2007) 679–689

List of symbols

Dxr, DPe change in angular speed of the rotor and KA, Ka, Ta thyristor gain, amplifier gain and amplifier
accelerating power in pu, respectively time constant in s, respectively
xkþ1
i ; v kþ1
i modified position and velocity of ith particle, Ke, Te exciter gain and time constant in s
respectively H, D inertia constant in MW-s/MVA and damping
xki ; vki current position and velocity of ith particle at torque coefficient in pu torque/pu speed devia-
kth iteration, respectively tion, respectively
pBesti, gBest personal best of ith particle and group ACE area control error
best, respectively TR, Tg, Tt, Tr terminal voltage transducer, governor,
c1, c2 positive constant representing cognitive and so- non-reheat, reheat turbine time constant in s,
cial learning rate, respectively respectively
r1, r2, r3, r4 random numbers in the interval [0, 1] c, b, R reheat parameter, frequency bias coefficient and
vmin
i ; vi
max
minimum and maximum velocity of the ith governor regulation, respectively
particle, respectively Pcraz a predefined craziness probability
vcraziness
i random number chosen uniformly in the inter- DP tie ; Df change in tie line power exchange and fre-
val [vmin max
i , vi ] quency, respectively
Eb infinite bus voltage in pu Osh, Ush, Tst overshoot, undershoot and settling time of
Dv1 change in output voltage of the terminal voltage transient response, respectively
d
transducer dt ðDw r
Þ derivative of angular speed deviation
wfd field circuit flux linkage Kp, Ki, Kd proportional, integral, derivative gains of
Td1, Td2, Td3, Td4 time constants of the power system PID controller
stabilizer block DPd, DVref change in load torque and reference voltage
Ks1, Ks2 power system stabilizer gains input

stabilization action is through auxiliary corrective signals Takagi Sugeno Fuzzy Logic (TSFL) based approach can
to the reference of the AVR. Dual input PSS utilizes Dxr adjust the parameters on-line according to the dynamic
and DPe as the effective two input signals and the processed environment in which it works and can provide optimal
output of PSS (DVpss) is fed to the excitation loop. To list a damping over a wide range of operating conditions. The
few among the numerous works on PSS carried out by the coordinated PSS-CS model as compared to PID-CS model
researchers are pole-placement [1]/pole shifting [2] based as presented in this paper proves to be the most robust
PSS, PID PSS, artificial neural network PSS [3], Genetic model in damping all electromechanical local modes of
Algorithm (GA)/fuzzy logic based/rule based self tuning generator’s angular speed oscillations for all off-line and
of PSS, neurofuzzy PSS, Simulated Annealing based PSS on-line conditions, step changes of DPd, DVref and dur-
(SAPSS), design of sliding mode PSS via GA, a heuristic ing/after clearing of system faults.
dynamic programming based PSS. In PSS-CS, fourth order model of generator with AVR
In [4,5] gains of PID controller installed in AGC loop and high gain thyristor excitation system is considered.
have been optimized and thus optimal transient response For PID-CS, fourth order model of generator with AVR
profiles of system frequency deviation as well as tie-line and normal gain amplifier with normal exciter is consid-
power exchange have also been obtained. The same AGC ered. Simulation study reveals that with high gain exciter,
model has been utilized in the present work for the AGC PID control is not at all effective.
loop. PSS model utilized in [6] has been installed in excita-
tion loop. One system termed as ‘PSS Controlled System 2. Coordinated system model
(PSS-CS)’ constitutes coordinated system model including
PID controlled AGC loop assisted by PSS controlled 2.1. AGC model
AVR loop. On the other hand, the other model termed
as ‘PID Controlled System (PID-CS)’ constitutes coordi- Closed loop transfer function model of reheat PID con-
nated system model including PID controlled AGC loop trolled AGC loop is shown Fig. 1 (AGC loop subsystem of
assisted by PID controlled AVR loop. Figs. 6 and 7). Input to the PID controller is ACE. Inputs
A novel class of Particle Swarm Optimization (PSO) to the summer1 are output of PID controller (with negative
called as Craziness based Particle Swarm Optimization PID gains (Kp, Ki, Kd)) as positive and R1 f0 Dxr as negative.
(CRPSO) has been adopted in the process of optimization Inputs to the summer2 are DPd as negative, DPe as negative
for optimal tuning of the parameters of PSS-CS and PID-CS. and DPm as positive. PID controller gains are given by (2).
 V. Mukherjee, S.P. Ghoshal / Electrical Power and Energy Systems 29 (2007) 679–689 681

Fig. 1. Transfer function model of reheat AGC loop.

2.2. SMIB system and PSS controlled AVR system 2.3. SMIB system and PID controlled AVR system

Fig. 2 depicts a SMIB System [7] and is considered in Block diagram subsystem representation of SMIB sys-
this paper. Block diagram subsystem representation of tem with AVR (normal gain exciter) and synchronous gen-
SMIB system with AVR, exciter and synchronous genera- erator for PID-CS is shown in Fig. 5 (AVR input, Dd are
tor for PSS-CS is shown in Fig. 3. In Fig. 3, thyristor exci- the input ports and Dv1, DPe are output ports). In this
ter input, Dd are the input ports and Dv1, DPe are output model we remove the dual input PSS and place a PID con-
ports of AVR loop subsystem of Fig. 6. The synchronous troller in excitation loop to get the PID-CS configuration.
generator with AVR, IEEE Type ST1A thyristor excitation The mathematical model of the PID controller is the same
system along with generator and equivalent transmission as stated in (2), except the parameters being Kpp, Kii, Kdd
line reactance are represented by a two-axis, fourth order (proportional, integral, derivative gains of PID controller,
model. The block diagram subsystem representation of respectively) and the sign of all the parameters being posi-
the dual input PSS as taken for the present work [6,8] is tive. Simulation reveals that negative sign gives instability.
shown in Fig. 4. Figs. 1, 3 and 4 are subsystems of Fig. 6. Figs. 1 and 5 are subsystems of Fig. 7.

Fig. 4. MATLAB-SIMULINK based subsystem block diagram represen-

Fig. 2. Single machine infinite bus system. tation of dual input PSS.

Fig. 3. MATLAB-SIMULINK based AVR subsystem of PSS-CS.

682 V. Mukherjee, S.P. Ghoshal / Electrical Power and Energy Systems 29 (2007) 679–689

Fig. 5. MATLAB-SIMULINK based AVR subsystem of PID-CS.

Fig. 6. MATLAB-SIMULINK based coordinated excitation and AGC loop representation for PSS-CS.

Fig. 7. MATLAB-SIMULINK based coordinated excitation and AGC loop representation for PID-CS.
 V. Mukherjee, S.P. Ghoshal / Electrical Power and Energy Systems 29 (2007) 679–689 683

2.4. AGC coordinated system with PSS controlled AVR

system or PID controlled AVR system

Since there is a weak coupling between the load fre-

quency control and AVR systems, the frequency and volt-
age are controlled separately. We can study the coupling
effect of linearized AGC system and the excitation system.
Combining the AGC system of Fig. 1, AVR system of
Fig. 3 and PSS loop of Fig. 4 all together yields Fig. 6.
Fig. 6 in the present work is PSS-CS. Again, combining
the AGC system of Fig. 1, AVR system of Fig. 5 together
yields Fig. 7 and is termed as PID-CS. The SMIB system Fig. 8. D-shaped sector in the negative half of s plane.
with AVR and exciter along with PSS may be represented
by state transition matrix of appropriate order. The order reduction in imaginary parts also, thus enhancing relative
of the state transition matrix for the SMIB system along stability and increasing the damping ratio above no (0.3).
with PSS is of twelve for PSS-CS where as the same is of Finally, all closed loop system poles should lie within a
eleven for PID-CS. The state matrix representation is D-shaped sector (Fig. 8) in the negative half plane of jx
shown in Appendix. axis for which ri  6.0, ni  0.3.
Thorough computation shows that optimization of MF1
3. Design of the misfitness function is not sufficient for sharp tuning of the model parameters.
So, it has become essential to design second multi-objective
The parameters of the PSS-CS model (PID gains i.e. Kp, misfitness function, MF2 for sharp tuning of the parame-
Ki, Kd with negative sign for AGC loop and PSS parame- ters of the model considered in this paper.
ters namely Ks1, Ks2, Td1, Td2, Td3, Td4) and that of PID- So, second multi-objective misfitness function,
CS model (PID gains i.e. Kp, Ki, Kd with negative sign  2
6 2 6 2 2 d 6
for AGC loop and Kpp, Kii, Kdd with positive sign for MF2 ¼ ðOsh 10 Þ þ ðU sh 10 Þ þ ðT st Þ þ ðDwr Þ10
AVR loop) are to be so tuned that some degree of relative dt
stability and damping of electromechanical modes of oscil- ð4Þ
lations, minimized Osh, minimized Ush and lesser Tst of Minimize MF1 and MF2 in succession with the help of
oscillations are achieved. So, to satisfy all these require- CRPSO based optimization to achieve model’s optimized
ments, two multi-objective misfitness functions, MF1, parameters, which are subject to the limits as available in
MF2 which are to be minimized in succession have been literature [4–6]:
designed in Pthe following way: 0.2 6 Kp, Kpp, Kd, Kdd, Ki, Kii 6 5.0,  100.0 6 Ks1 6
2
MF11 ¼ i ðr0  ri Þ ; if ri P r0, r0 = 6.0, ri is the real 10.0, 10.0 6 Ks2 6 100.0, 0.005 6 Td1, Td2, Td3, Td4 6
part of the ith eigenvalue. The relative stability is deter- 2.0.
mined by rP 0.
2
MF12 ¼ i ðn0  ni Þ , if (imaginary part of the ith eigen- 4. Particle swarm optimization
value) > 0.0, ni is the damping ratio of the ith eigenvalue
and ni < no. Minimum damping ratio considered, 4.1. Review of particle swarm optimization algorithm
no = 0.3. Minimization of this misfitness function will min-
imize maximum
P overshoot. The PSO [9–11] is a population based optimization tech-
MF13 ¼ i (imaginary part of ith eigenvalue)2, if nique, where the population is called ‘swarm’. Each solu-
ri P 6.0. High value of imaginary part of ith eigenvalue tion candidate, called a ‘particle’, flies in the problem
to the right of vertical line r0 = 6.0 is to be prevented. space (similar to the search process for food of a bird
Zeroing of MF13 will increase the damping further. swarm) looking for the optimal position. A ‘particle’ as
MF14 = an arbitrarily chosen very high fixed value (say, time passes through, adjusts its position according to its
106), which will indicate some ri values P 0.0. This means own ‘experience’, as well as according to the ‘experience’
unstable oscillations occurs for the particular model of neighboring particles. If a particle discovers a promising
parameters. These particular model parameters will be new solution, all the other particles will move closer to it,
rejected during the selection step of the PSO cycle. exploring the region more thoroughly in the process. Based
So, first multi-objective misfitness function, on PSO concept mathematical equations for the searching
MF1 ¼ 10MF11 þ 10MF12 þ 0:01MF13 þ MF14 ð3Þ process are:
The weighting factors are being appropriately chosen to Velocity updating equation : vkþ1 ¼ vki þ c1  r1
i
make them mutually competitive during optimization. By    
optimizing MF1, closed loop system poles are thus consis-  pBesti  xki þ c2  r2  gBest  xki ð5Þ
tently pushed further left of jx axis with simultaneous Position updating equation : xkþ1
i ¼ xki þ vkþ1
i ð6Þ
684 V. Mukherjee, S.P. Ghoshal / Electrical Power and Energy Systems 29 (2007) 679–689

4.2. Craziness based particle swarm optimization (CRPSO) 4.3. Selection of CRPSO parameters

The following modifications help to enhance the global The main flavors of this novel class of PSO are
search ability of PSO algorithm.
(i) Position and velocity updating: The second and (a) balancing between over use and under use of social
the third terms on the right hand side of (5) represents and personal experiences by random number r1 as
the personal and social behavior of the particles. As the shown in (7);
two random numbers r1 and r2 are generated randomly, (b) balancing between global and local searches using
there may be the cases when these two random numbers random number r2 as given in (7);
may be too large or too small. Personal and social expe- (c) change in the direction in velocity of the particles by
riences will be over used driving the particles too far away sign(r3) as stated in (8);
from the local optimum if these two random numbers are (d) diversity in the direction by inclusion of craziness by
too large whereas, for the other case these two experi- vcraziness
i and Pcraz as noted in (10).
ences are not fully used reducing the convergence speed.
These two cases conflict the human natural behavior at The authors have investigated the impact of vcraziness
i and
work as they always try to extract the best personal Pcraz on the performance of CRPSO. It has been revealed
and group experiences accumulated so far. Two random that more the value of these parameters are, the more is
weighting parameters in (5) are not completely indepen- the fluctuation of final convergent value of MF1 and
dent. If one is too small the other should be too large MF2. If Pcraz is reduced to become less than 0.2, then
and vice versa. Thus, introduction of only one random vcraziness
i term in (10) is often be introduced which is also
number (r1) and noting that the sum of two interrelated not desirable. The definition of sign(r3) as in (9) shows that
random numbers are unity, as proposed in [12] may be the chance of sign(r3) to be as 1 is very low due to rare
stated as in (7). chance of r3 becoming less than 0.05 leading to very often
chance in change of direction of the particles. Eq. (11) deter-
vkþ1
i ¼ r2  vki þ ð1  r2Þ  c1  r1  ðpBesti  xki Þ mines whether the velocity of the particle is to be crazed or
þ ð1  r2Þ  c2  ð1  r1Þ  ðgBest  xki Þ ð7Þ not. The direction of vcraziness
i is determined by sign(r4).
Depending upon the value of sign(r4) which is on the other
Balances between local and global searches will be main- hand depends on chance of occurrence of r4 to become less
tained by random number r2 as stated in (7). Change in the than or greater than 0.5, vcraziness
i is added or subtracted in
direction in velocity [12] may be modeled as in (8). velocity updating Eq. (10). Moderate inclusion of craziness
is always recommended as the chance of occurrence of r4
vkþ1 ¼ r2  signðr3Þ  vki þ ð1  r2Þ  c1  r1 becoming less than 0.5 is usually 50%. Thus, the choice of
i
  vcraziness
i ¼ 0:1 to 0.4, Pcraz = 0.2, sign(r3) as in (9)
 pBesti  xki þ ð1  r2Þ  c2  ð1  r1Þ and sign(r4) as in (12) may be accepted values and
 
 gBest  xki ð8Þ definitions.

In (8), sign (r3) may be defined as

5. Takagi Sugeno fuzzy logic as applied to on-line tuning of

1; ðr3 6 0:05Þ parameters
signðr3Þ ¼ ð9Þ
1; ðr3 > 0:05Þ
The whole process involves three steps [4,5] as
(ii) Inclusion of craziness: Diversity in the direction of
birds flocking or fish schooling may be handled in tradi- (a) Fuzzification: Fuzification is done for input operat-
tional PSO with a predefined craziness probability [12]. ing conditions as active power (P), reactive power
The particles may be crazed in accordance with (10) before (Q), equivalent transmission line reactance (Xe) and
updating its position. Generator’s LT side bus voltage (Et) in terms of
fuzzy subsets (low, medium, high). TSFL rule base
vkþ1
i ¼ vkþ1
i þ Prðr4Þ  signðr4Þ  vcraziness
i ð10Þ table is formed, each row of the table composed
of four nominal inputs and corresponding nominal
where, Pr(r4) and sign(r4) are defined, respectively, as optimal PSS parameters and P, I, D gains as output
( determined by the optimizing algorithm CRPSO.
1; ðr4 6 P craz Þ The respective nominal central values of the input
Prðr4Þ ¼ ð11Þ subsets of P are (0.2, 0.7, 1.2), those of Q are
0; ðr4 > P craz Þ (0.2, 0.6, 1.0), those of Xe are (0.4752, 0.77, 1.08)
(
1; ðr4 P 0:5Þ and those of Et are (0.5, 0.8, 1.1), respectively, at
signðr4Þ ¼ ð12Þ which membership values are unity. These are nom-
1; ðr4 < 0:5Þ inal input conditions also. Fuzzy rule base table
 V. Mukherjee, S.P. Ghoshal / Electrical Power and Energy Systems 29 (2007) 679–689 685

consists of 34(= 81) logical input conditions or sets, of PID. Final crisp PSS parameter and PID gains
each composed of four nominal inputs. Each logical output,
input set corresponds to nominal optimal PSS P ðiÞ
parameters and PID gains as output. i lmin  Ki
K crisp ¼ P ðiÞ
ð13Þ
(b) Fuzzy inference: This is done for on-line imprecise val- i lmin
ues of input operating conditions, determining valid
input logical sets and corresponding minimum l and where i corresponds to input logical sets being satisfied
nominal PSS parameters and PID gains from TSFL among 81 input logical sets, Ki is nominal PSS param-
table for each valid logical set. eter and PID gains, Kcrisp is crisp PSS parameter and
ðiÞ
(c) Defuzzification: Defzification yields the defuzzified, PID gains. lmin is the minimum membership value
crisp output for each parameter of PSS and gains corresponding to ith input logical set being satisfied.

Fig. 9. Comparison of response profiles of Dxr (pu) applying modal analysis (with initial state deviation Dd = 5 = 0.0857 rad).
686 V. Mukherjee, S.P. Ghoshal / Electrical Power and Energy Systems 29 (2007) 679–689

6. Input parameters Q 6 1:0; 0:4752 6 X e 6 1:08; 0:5 6 Et 6 1:1: At t= 0,

it is assumed that Dd = 5 = 0.0857 rad, Dxr = 0.0,
(a) For CRPSO, maximum population size = 50, maxi- TR = 0.01 s, thyristor exciter gain, KA = 200.0.
mum number of loops = 5, maximum allowed itera-
tion cycles (generations) for optimization per Simulation results are obtained by MATLAB 6.5 soft-
loop = 200, c1 = 2.05, c2 = 2.05, Pcraz = 0.2. The ware on a 3.0 GHz P4 computer.
value of vcraziness
i lies between 0.1 and 0.4.
(b) For AGC loop, the constant input data of the ther- 7. Simulation results and discussions
mal generating station are [4,5]: b = 0.425,
R = 2.4 Hz/pu, Tg = 0.08 s, Tt = 0.3 s, Tr = 4.2 s, (i) Analytical transient response: Model parameters on
c = 0.35, DPd = 0.01 pu. sample basis for PID-CS (Kp = 2.23, Ki = 4.70,
(c) For SMIB system, H = 3.5, nominal frequency, f0 = Kd = 4.91, Kpp = 3.41, Kii = 5.00, Kdd = 0.66) and that
50.0 Hz, 0:995 6 Eb \00 6 1:0; 0:2 6 P 6 1:2; 0:2 6 for PSS-CS (Kp = 0.57, Ki = 4.54, Kd = 2.50,

Fig. 10. MATLAB-SIMULINK based transient step response profile. (a) 0.01 pu step change in DPd and no change in DVref, (b) 0.01 pu step change in
DVref and no change in DPd, (c) simultaneous 0.01 pu step changes in DPd and DVref.
 V. Mukherjee, S.P. Ghoshal / Electrical Power and Energy Systems 29 (2007) 679–689 687

Table 1
Comparison of misfitness function
Operating conditions P, Q, Type of Value of Value of
Xe, Et (all are in pu) system model MF1 MF2 · 103
0.2, 0.2, 0.4752, 0.5 PID-CS 1282.00 3.6031
PSS-CS 727.11 3.2088
0.2, 1.0, 0.4752, 0.8 PID-CS 1201.60 2.9951
PSS-CS 564.71 2.7115
0.2, 0.2, 0.4752, 1.1 PID-CS 1310.60 6.4021
PSS-CS 629.62 2.5098

Ks1 = 18.82, Ks2 = 37.56, Td1 = 1.21, Td2 = 0.35,

Td3 = 0.03, Td4 = 0.05) are computed using CRPSO, nom- Fig. 11. Comparison between transient response profiles of Dxr (pu) of
inal operating conditions being P = 0.2, Q = 0.6, both the system model under change in operating conditions with models
parameters given in Table 2 as determined by TSFL.
Xe = 0.4752, Et = 1.1 (all are in pu).
Computation and plot of responses of Dxr are
achieved with the help of modal analysis [7]. In
Fig. 9a–f, a comparison of responses of Dxr for PSS- (iv) System transient response under simulation of fault:
CS and PID-CS has been done. These optimized A fault of duration 220 ms at the L.T. bus has been simu-
response performances are achieved with the help of lated at the instant of 2.0 s and the corresponding compar-
optimum PID gains and optimum PSS parameters ative transient response profiles of Dxr for both the system
obtained through CRPSO. From Fig. 9a–f, it is clear models is displayed in Fig. 11. A very close look into this
that PSS-CS model offers lesser overshoot, undershoot figure shows that after the creation of the fault the PSS-
and settling time and thereby lesser value of MF2 which CS model recovers from this abnormal situation with much
establishes that PSS-CS model is more effective in settling lesser fluctuation in angular speed as compared to that of
down the oscillations quickly. Thus, PSS-CS model PID-CS model. Table 2 depicts the system model parame-
assists in damping system oscillations quickly ensuring ters as determined by TSFL. Thus, PSS-CS model exhibits
better small signal stabilization performance. better response having lesser amplitude of angular speed
(ii) Value of misfitness function: Trial run of the pro- deviation under fault and subsequent clearing condition
posed algorithm gives the values of MF1 and MF2 for dif- yielding better dynamic robust transient performance.
ferent operating conditions as shown in Table 1. From this
table it is clear that PSS-CS offers lesser value of MF1 and 8. Conclusion
MF2. This indicates that PSS-CS is much more effective
than PID-CS in small signal dynamic stabilization. For tuning of the PID controller gains and the parame-
(iii) System transient response under perturbation: MAT- ters of PSS, this paper represents a novel evolutionary
LAB-SIMULINK plot of step response profiles of tran- search technique Craziness based Particle Swarm Optimi-
sient responses of Dxr and Dv1 for both the system zation. Dual input PSS equipped AVR loop with high gain
models for 0.01 pu step change in DPd with no change in thyristor exciter model stabilizes the system oscillations
DVref (Fig. 10a), for 0.01 pu step change in DVref with no quickly as compared to PID controlled AVR with normal
change in DPd (Fig. 10b) and for simultaneous 0.01 pu step gain exciter. Simulation study also reveals that with high
changes in DPd and DVref (Fig. 10c) simulated to study the gain exciter, PID control is not at all effective. Individual
comparison of system responses under perturbation. These and simultaneous step changes in load torque and reference
plots (Fig. 10a–c), establish that PSS controlled system voltage reveal that PSS equipped excitation system settles
configuration exhibits better transient response than PID the oscillation with lesser overshoot, undershoot and set-
controlled one. tling time as compared to PID controlled excitation system.

Table 2
Off-nominal parameters, system model and optimal model parameters as computed by TSFL
Operating conditions P, Q, Xe, Et System Model parameters PSS-CS (Kp, Ki, Kd, Ks1, Ks2, Td1, Td2, Td3, Td4) PID-CS (Kp, Ki,
model Kd, Kpp, Kii, Kdd)
1.0, 0.6, 0.4752, 1.1 (Nominal input) PSS-CS 1.35, 2.38, 3.32, 10.00, 10.08, 0.76, 0.15, 0.006, 0.005
PID-CS 0.72, 0.59, 0.22, 0.24, 0.20, 4.97
L.T. bus fault of duration 220 ms and subsequent PSS-CS No change in parameters
clearing PID-CS
0.2, 0.2, 0.4752, 1.1 (Off-nominal input) PSS-CS 4.97, 0.2, 4.5, 10.00, 19.42, 0.21, 0.005, 0.06, 0.10
PID-CS 0.80, 0.77, 0.22, 0.20, 5.00, 4.94
688 V. Mukherjee, S.P. Ghoshal / Electrical Power and Energy Systems 29 (2007) 679–689

It is also revealed that the system response after the occur- Appendix
rence of a fault is quickly stabilized for PSS equipped exci-
tation system model. Thus, installation of intelligent fuzzy State space representation in the form X_ ¼ AX þ BU
logic tuned PSS with high gain thyristor excitation systems (A = state matrix, B = perturbation matrix, X = state-
is justified. For real time tuning of model parameters, fast, vector, U = perturbation vector) gives the state matrix
adaptive Takagi Sugeno Fuzzy Logic may be adopted. Its for PSS-CS model as in (A.1), (A.2) while that for PID-
computational burden is very low. It is easy for practical CS model as in (A.3), (A.4), (A.6). The detail parameters
implementation. are as in [7].

The A matrix for PSS-CS model:

2 D K 1 K 2 1 3
2H 2H
0 2H
0 0 0 0 0 0 0 2H
6x 0 0 0 0 0 0 0 0 0 0 0 7
6 0 7
6 7
60 K 5 1 K6
0 0 0 0 0 0 0 0 7
6 TR TR TR 7
6 K 3 K 4 K A K 3 1 K A K 3 K s2 K A K 3 K s1 7
60 0 0 0 0 0 0 7
6 T3 T3 T3 T3 T3 7
6 1 1 7
6 T d3 0 0 0 0 0 0 0 0 0 0 7
6 T d3 7
6 1 0 0 0 1 1
0 0 0 0 0 0 7
6 T d3 T d3 T d4 7
A¼6 60 K1 K2 1
7
7 ðA:1Þ
6 T d1
0 T d1
0 0 T d1
0 0 0 0 0 7
6 7
60 K1
0 K2
0 0 1 1
0 0 0 0 7
6 T d1 T d1 T d1 T d2 7
6 7
6 bf0 0 0 0 0 0 0 0 0 0 0 0 7
6 7
6A K i 1
A1012 7
6 1001 A1002 0 A1004 0 0 0 0 Tg Tg
0 7
6 7
60 0 0 0 0 0 0 0 0 1 1
0 7
4 Tt Tt 5
c 1
0 0 0 0 0 0 0 0 0 Tt
A1211 Tr

where,
       9
bf0 K d D bf0 K p bf0 K d >
A1001 ¼ 2HT g
 RTf0g  Tg
; A1002 ¼ bf2HT
0Kd K1
g
; A 1004 ¼ bf0 K d K 2
2HT g
; A 1012 ¼  2HT g
=
  ðA:2Þ
A1211 ¼ T t cT r >
;
T tT r

The A matrix for PID-CS model:

2 D K 1 1 K 2
3
2H 2H
0 0 0 2H 2H
0 0 0 0
6 7
6 x0 0 0 0 0 0 0 0 0 0 0 7
6 7
6 7
6 bf0 0 0 0 0 0 0 0 0 0 0 7
6 7
6 7
6 A0401 A0402 A0403 A0404 0 A0406 A0407 0 0 0 0 7
6 7
6 7
60 0 0 1 1
0 0 0 0 0 0 7
6 Tt Tt 7
6   7
6 7
A¼6 60 0 0 c
Tt
1
Tr
 Tct 1
Tr
0 0 0 0 0 7
7 ðA:3Þ
6 7
6 K 4 K 3 1 K3 7
60 0 0 0 0 0 0 0 7
6 T3 T3 T3 7
6 7
60 0 0 0 0 0 0 1
0 Ke
0 7
6 Te Te 7
6 7
6 A0901 A0902 0 0 0 0 A0907 A0908 A0909 0 A0911 7
6 7
6 7
60 0 0 0 0 0 0 0 Ka 1
0 7
4 Ta Ta 5
0 A1102 0 0 0 0 A1107 0 A1109 0 A1111

ðK p bf0 Rþf0 Þ

9
K i
A0401 ¼ RT g
þ DK d bf0
2HT g
; A 0402 ¼ K d bf0 K 1
2HT g
A 0403 ¼ Tg
; =
where; ðA:4Þ
A0404 ¼ 1 ; A0406 ¼ K d bf0
; A0407 ¼ K d2HT
bf0 K 2 ;
Tg 2HT g g
 V. Mukherjee, S.P. Ghoshal / Electrical Power and Energy Systems 29 (2007) 679–689 689

9
A0901 ¼ K ddT KR 5 x0 >
>
>
>
K dd K 5 ðK pp T R K dd K ii T 2R Þ ðK dd K 5 T 3 K pp T R K 5 T 3 þK dd K 6 K 4 K 3 T R Þ >
>
A0902 ¼ þ >
=
T 2R ðK dd K pp T R Þ T 3 T 2R
K dd K 6 ðK pp T R K dd K ii T 2R Þ
ðA:5Þ
A0907 ¼ þ
ðK dd K 6 T 3 K pp T R K 6 T 3 þK dd K 6 T R Þ >
>
T 2R ðK dd K pp T R Þ T 3 T 2R >
>
>
>
ðK pp T R K dd K ii T 2R Þ K ii ðK pp T R K dd K ii T R Þ >
2
;
A0908 ¼ KTddR KT 36 K 3 ; A0909 ¼ T R ðK dd K pp T R Þ
; A0911 ¼ T R ðK dd K pp T R Þ

K dd K 5 K dd K 6 T R K ii T R
A1102 ¼ ; A1107 ¼ ; A1109 ¼ ; A1111 ¼ ðA:6Þ
ðK pp T R  K dd Þ ðK pp T R  K dd Þ ðK pp T R  K dd Þ ðK pp T R  K dd Þ

References [7] Kundur P. Power system stability and control. New York: Mc-
Graw-Hill; 1994.
[1] Chow JH, Sanchez-Gasca JJ. Pole placement designs of power system [8] IEEE Digital Excitation System Subcommittee Report. Computer
stabilizers. IEEE Trans PS 1989;4 (3):271–7. models for representation of digital-based excitation systems. IEEE
[2] El-Sherbiny MK, Hasan MM, El-Saady G, Yousef AM. Optimal Trans EC 1996;11(3):607–15.
pole shifting for power system stabilization. Int J Elec Power 2003;66 [9] Kennedy J, Eberhart RC. Particle swarm optimization. IEEE Int
(3):253–8. Conf NN 1995;4:1942–8.
[3] Hsu YY, Chen CR. Tuning of power system stabilizers using an [10] Eberhart RC, Shi Y. Particle swarm optimization: develop-
artificial neural network. IEEE Trans EC 1991;6 (4):612–9. ments, applications and resources. IEEE Int Cong EC 2001;1:
[4] Ghoshal SP. Applications of GA/GASA based fuzzy automatic 81–6.
generation control of a multi-area thermal generating system. Int J [11] Naka S, Genji T, Yuru T, Fukuyama YA. Hybrid particle swarm
Elec Power 2004;70 (2):115–28. optimization for distribution state estimation. IEEE Trans PS
[5] Ghoshal SP. Optimization of PID gains by particle swarm optimi- 2003;18:60–8.
zation in fuzzy based automatic generation control. Int J Elec Power [12] Ho SL, Yang Shiyou, Ni Guangzheng, Lo Edward WC,
2004;72 (3):203–12. Wang HC. A particle swarm optimization-based method for
[6] Mukherjee V, Ghoshal SP. Velocity relaxed swarm intelligent tuning multiobjective design optimizations. IEEE Trans M 2005;41(5):
of fuzzy based power system stabilizer. IEEE PES Conference 2006: 1756–9.
Paper ID 85.