See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/3270304

A Particle Swarm Optimization Approach for Optimum Design of PID Controller

in AVR System

Article in IEEE Transactions on Energy Conversion · July 2004

DOI: 10.1109/TEC.2003.821821 · Source: IEEE Xplore

CITATIONS READS
1,701 7,090

1 author:

Zwe-Lee Gaing
Kao Yuan University
37 PUBLICATIONS 5,088 CITATIONS

SEE PROFILE

All content following this page was uploaded by Zwe-Lee Gaing on 04 February 2014.

The user has requested enhancement of the downloaded file.

384 IEEE TRANSACTIONS ON ENERGY CONVERSION, VOL. 19, NO. 2, JUNE 2004

A Particle Swarm Optimization Approach for

Optimum Design of PID Controller in AVR System
Zwe-Lee Gaing, Member, IEEE

Abstract—In this paper, a novel design method for determining Many random search methods, such as genetic algorithm
the optimal proportional-integral-derivative (PID) controller pa- (GA) and simulated annealing (SA) [2]–[9], have recently
rameters of an AVR system using the particle swarm optimization received much interest for achieving high efficiency and
(PSO) algorithm is presented. This paper demonstrated in detail
how to employ the PSO method to search efficiently the optimal searching global optimal solution in problem space. The GA
PID controller parameters of an AVR system. The proposed method is usually faster than the SA method because the GA
approach had superior features, including easy implementation, has parallel search techniques, which emulate natural genetic
stable convergence characteristic, and good computational effi- operations. Due to its high potential for global optimization,
ciency. Fast tuning of optimum PID controller parameters yields GA has received great attention in control systems such as the
high-quality solution. In order to assist estimating the perfor-
mance of the proposed PSO-PID controller, a new time-domain search of optimal PID controller parameters. Although GAs
performance criterion function was also defined. Compared with have widely been applied to many control systems, its natural
the genetic algorithm (GA), the proposed method was indeed genetic operations would still result in enormous computational
more efficient and robust in improving the step response of an efforts [5], [6]. In order to overcome the disadvantages, the
AVR system. use of real-value representation in the GA is proposed to offer
Index Terms—AVR system, optimal control, particle swarm op- a number of advantages in numerical function optimization
timization, PID controller. over binary encoding because there is no need to convert
chromosomes to binary type [3]–[5], [15].
I. INTRODUCTION Though the GA methods have been employed successfully to
solve complex optimization problems, recent research has iden-

D URING the past decades, the process control techniques in

the industry have made great advances. Numerous control
methods such as adaptive control, neural control, and fuzzy con-
tified some deficiencies in GA performance. This degradation in
efficiency is apparent in applications with highly epistatic ob-
jective functions [i.e., where the parameters being optimized are
trol have been studied [1]–[5]. Among them, the best known is highly correlated (the crossover and mutation operations cannot
the proportional-integral-derivative (PID) controller, which has ensure better fitness of offspring because chromosomes in the
been widely used in the industry because of its simple struc- population have similar structures and their average fitness is
ture and robust performance in a wide range of operating con- high toward the end of the evolutionary process)] [10], [14].
ditions. Unfortunately, it has been quite difficult to tune prop- Moreover, the premature convergence of GA degrades its per-
erly the gains of PID controllers because many industrial plants formance and reduces its search capability [10].
are often burdened with problems such as high order, time de- Particle swarm optimization (PSO), first introduced by
lays, and nonlinearities [1]–[6]. Over the years, several heuristic Kennedy and Eberhart, is one of the modern heuristic algo-
methods have been proposed for the tuning of PID controllers. rithms. It was developed through simulation of a simplified
The first method used the classical tuning rules proposed by social system, and has been found to be robust in solving
Ziegler and Nichols. In general, it is often hard to determine op- continuous nonlinear optimization problems [11]–[15]. The
timal or near optimal PID parameters with the Ziegler-Nichols PSO technique can generate a high-quality solution within
formula in many industrial plants [1]–[3]. shorter calculation time and stable convergence characteristic
For these reasons, it is highly desirable to increase the capa- than other stochastic methods [14]–[16]. Much research is still
bilities of PID controllers by adding new features. Many artifi- in progress for proving the potential of the PSO in solving com-
cial intelligence (AI) techniques have been employed to improve plex power system operation problems. Researchers including
the controller performances for a wide range of plants while re- Yoshida et al. have presented a PSO for reactive power and
taining their basic characteristics. AI techniques such as neural voltage control (VVC) considering voltage security assessment.
network, fuzzy system, and neural-fuzzy logic have been widely Their method is compared with the reactive tabu system (RTS)
applied to proper tuning of PID controller parameters [1], [2]. and enumeration method on practical power system, and has
shown promising results [16]. Naka et al. have presented
Manuscript received November 6, 2002. This work was supported by the Na- the use of a hybrid PSO method for solving efficiently the
tional Science Council of Taiwan, R.O.C. practical distribution state estimation problem [17]. Because
The author is with the Department of Electrical Engineering, Kao-Yuan the PSO method is an excellent optimization methodology and
Institute of Technology, Kaohsiung 821, Taiwan, R.O.C. (e-mail: zlgaing@
ms39.hinet.net). a promising approach for solving the optimal PID controller pa-
Digital Object Identifier 10.1109/TEC.2003.821821 rameters problem; therefore, this study develops the PSO-PID
0885-8969/04$20.00 © 2004 IEEE
GAING: A PARTICLE SWARM OPTIMIZATION APPROACH FOR OPTIMUM DESIGN OF PID CONTROLLER 385

Fig. 1. Block diagram of an AVR system with a PID controller.

controller to search optimal PID parameters. This controller is The amplifier model is represented by a gain and a
called the PSO-PID controller. time constant ; the transfer function is
The integral performance criteria in frequency domain were
often used to evaluate the controller performance, but these cri- (2)
teria have their own advantages and disadvantages [5], [6]. In
Typical values of are in the range of 10 to 400. The
this paper, a simple performance criterion in time domain is pro-
amplifier time constant is very small ranging from 0.02 to
posed for evaluating the performance of a PSO-PID controller
0.1 s.
that was applied to the complex control system.
• Exciter model.
The generator excitation system maintains generator voltage
The transfer function of a modern exciter may be rep-
and controls the reactive power flow using an automatic voltage
resented by a gain and a single time constant
regulator (AVR) [18]. The role of an AVR is to hold the terminal
voltage magnitude of a synchronous generator at a specified (3)
level. Hence, the stability of the AVR system would seriously
affect the security of the power system. In this paper, a practical Typical values of are in the range of 10 to 400. The
high-order AVR system with a PID controller is adopted to test time constant is in the range of 0.5 to 1.0 s.
the performance of the proposed PSO-PID controller. • Generator model.
In this paper, besides demonstrating how to employ the PSO In the linearized model, the transfer function relating
method to obtain the optimal PID controller parameters of an the generator terminal voltage to its field voltage can be
AVR system, many performance estimation schemes are per- represented by a gain and a time constant
formed to examine whether the proposed method has better per-
formance than the real-value GA method in solving the optimal (4)
PID controller parameters.
These constants are load dependent, may vary be-
tween 0.7 to 1.0, and . between 1.0 and 2.0 s from full
II. LINEARIZED MODEL OF AN AVR SYSTEM
load to no load.
A. PID Controller • Sensor model.
The PID controller is used to improve the dynamic response The sensor is modeled by a simple first-order transfer
as well as to reduce or eliminate the steady-state error. The function, given by
derivative controller adds a finite zero to the open-loop plant
transfer function and improves the transient response. The inte- (5)
gral controller adds a pole at the origin, thus increasing system is very small, ranging from of 0.001 to 0.06 s.
type by one and reducing the steady-state error due to a step
function to zero. The PID controller transfer function is C. AVR System With PID Controller
The above models provide an AVR system compensated with
(1)
a PID controller block diagram, which is shown in Fig. 1.

D. Performance Estimation of PID Controller

B. Linearized Model of an AVR System [16]
In general, the PID controller design method using the
The role of an AVR is to hold the terminal voltage magni- integrated absolute error (IAE), or the integral of squared-error
tude of a synchronous generator at a specified level. A simple (ISE), or the integrated of time-weighted-squared-error (ITSE)
AVR system comprises four main components, namely ampli- is often employed in control system design because it can be
fier, exciter, generator, and sensor. For mathematical modeling evaluated analytically in the frequency domain [3]–[6]. The
and transfer function of the four components, these components three integral performance criteria in the frequency domain
must be linearized, which takes into account the major time con- have their own advantages and disadvantages. For example, a
stant and ignores the saturation or other nonlinearities. The rea- disadvantage of the IAE and ISE criteria is that its minimization
sonable transfer function of these components may be repre- can result in a response with relatively small overshoot but
sented, respectively, as follows [16]. a long settling time because the ISE performance criterion
• Amplifier model. weights all errors equally independent of time. Although the
386 IEEE TRANSACTIONS ON ENERGY CONVERSION, VOL. 19, NO. 2, JUNE 2004

ITSE performance criterion can overcome the disadvantage The PSO concept consists of, at each time step, changing
of the ISE criterion, the derivation processes of the analytical the velocity of each particle toward its and loca-
formula are complex and time-consuming [6]. The IAE, ISE, tions. Acceleration is weighted by a random term, with separate
and ITSE performance criterion formulas are as follows: random numbers being generated for acceleration toward
and locations.
(6) For example, the th particle is represented as
in the -dimensional space. The best
(7) previous position of the th particle is recorded and represented
as . The index of
best particle among all of the particles in the group is repre-
(8)
sented by the . The rate of the position change (velocity)
for particle is represented as .
In this paper, a new performance criterion in the time domain
The modified velocity and position of each particle can be
is proposed for evaluating the PID controller. A set of good con-
calculated using the current velocity and the distance from
trol parameters , , and can yield a good step response
to as shown in the following formulas:
that will result in performance criteria minimization in the time
domain. These performance criteria in the time domain include
the overshoot , rise time , settling time , and steady-state
error . Therefore, a new performance criterion is de- (10)
fined as follows:

min
(9) (11)

where is , and is the weighting factor. where

The performance criterion can satisfy the designer re- number of particles in a group;
quirements using the weighting factor value. We can set to number of members in a particle;
be larger than 0.7 to reduce the overshoot and steady-state error. pointer of iterations (generations);
On the other hand, we can set to be smaller than 0.7 to reduce velocity of particle at iteration ,
the rise time and settling time. In this paper, is set in the range
;
of 0.8 to 1.5.
inertia weight factor;
, acceleration constant;
III. OVERVIEW OF PARTICLE SWARM OPTIMIZATION , random number between 0 and 1;
A. Features of Particle Swarm Algorithm [10]–[13] current position of particle at iteration ;
In 1995, Kennedy and Eberhart first introduced the particle of particle ;
swarm optimization (PSO) method. It is one of the optimization of the group.
techniques and a kind of evolutionary computation technique. In the above procedures, the parameter determined the
The method has been found to be robust in solving problems resolution, or fitness, with which regions be searched between
featuring nonlinearity and nondifferentiability, multiple optima, the present position and the target position. If is too high,
and high dimensionality through adaptation, which is derived particles might fly past good solutions. If is too small,
from the social-psychological theory. The features of the particles may not explore sufficiently beyond local solutions. In
method are as follows [14]. many experiences with PSO, was often set at 10–20% of
the dynamic range of the variable on each dimension.
• The method is developed from research on swarm such as
The constants and represent the weighting of the sto-
fish schooling and bird flocking.
chastic acceleration terms that pull each particle toward
• It can be easily implemented, and has stable convergence
and positions. Low values allow particles to roam far from
characteristic with good computational efficiency.
the target regions before being tugged back. On the other hand,
Instead of using evolutionary operators to manipulate the par-
high values result in abrupt movement toward, or past, target re-
ticle (individual), like in other evolutionary computational algo-
gions. Hence, the acceleration constants and were often
rithms, each particle in PSO flies in the search space with ve-
set to be 2.0 according to past experiences.
locity which is dynamically adjusted according to its own flying
Suitable selection of inertia weight in (12) provides a bal-
experience and its companions’ flying experience. Each particle
ance between global and local explorations, thus requiring less
is treated as a volumeless particle in -dimensional search space.
iteration on average to find a sufficiently optimal solution. As
Each particle keeps track of its coordinates in the problem
originally developed, often decreases linearly from about 0.9
space, which are associated with the best solution (evaluating
to 0.4 during a run. In general, the inertia weight is set ac-
value) it has achieved so far. This value is called . An-
cording to the following equation:
other best value that is tracked by the global version of the par-
ticle swarm optimizer is the overall best value, and its location,
obtained so far by any particle in the group, is called . (12)
GAING: A PARTICLE SWARM OPTIMIZATION APPROACH FOR OPTIMUM DESIGN OF PID CONTROLLER 387

where is the maximum number of iterations (genera- of the four performance criteria in the time domain,
tions), and is the current number of iterations. namely , , , and .
Step 3) Calculate the evaluation value of each individual in
IV. IMPLEMENTATION OF A PSO-PID CONTROLLER the population using the evaluation function given
by (13).
In this paper, a PID controller using the PSO algorithm was
Step 4) Compare each individual’s evaluation value with its
developed to improve the step transient response of AVR of a
. The best evaluation value among the is
generator. It was also called the PSO-PID controller. The PSO
denoted as .
algorithm was mainly utilized to determine three optimal con-
Step 5) Modify the member velocity of each individual
troller parameters , , and , such that the controlled system
according to (14)
could obtain a good step response output.

A. Individual String Definition

To apply the PSO method for searching the controller param-
eters, we use the “individual” to replace the “particle” and the
“population” to replace the “group” in this paper. We defined (14)
three controller parameters , , and , to compose an indi- where the value of is set by (12). When is 1,
vidual by ; hence, there are three members represents the change in velocity of controller
in an individual. These members are assigned as real values. If parameter. When is 2, represents the change in
there are individuals in a population, then the dimension of a velocity of controller parameter.
population is . The matrix representation in a population
Step 6) If , then
is as follows.
If , then .
B. Evaluation Function Definition Step 7) Modify the member position of each individual
according to (15)
In the meantime, we defined the evaluation function given
in (13) as the evaluation value of each individual in population.
The evaluation function is a reciprocal of the performance
(15)
criterion as in (9). It implies the smaller the value
of individual , the higher its evaluation value where and represent the lower and upper
bounds, respectively, of member of the individual
(13) . For example, when is 1, the lower and upper
bounds of the controller parameter are and
In order to limit the evaluation value of each individual of the , respectively.
population within a reasonable range, the Routh–Hurwitz cri- Step 8) If the number of iterations reaches the maximum,
terion must be employed to test the closed-loop system stability then go to Step 9. Otherwise, go to Step 2.
before evaluating the evaluation value of an individual. If the Step 9) The individual that generates the latest is an
individual satisfies the Routh–Hurwitz stability test applied to optimal controller parameter.
the characteristic equation of the system, then it is a feasible in-
dividual and the value of is small. In the opposite case, V. DYNAMIC BEHAVIORS ESTIMATION
the value of the individual is penalized with a very large In order to examine the dynamic behaviors and convergence
positive constant. characteristic of the proposed method, two statistical indexes,
namely the mean value and the standard deviation of
C. Proposed PSO-PID Controller evaluation values of all individuals in the population during the
This paper presents a PSO-PID controller for searching the computing processes, were used. The mean value can display
optimal or near optimal controller parameters , , and , the accuracy of the algorithm, and the standard deviation can
with the PSO algorithm. Each individual contains three mem- measure the convergence speed of the algorithm. The formulas
bers , , and . The matrix representation of the initial pop- for calculating the mean value and the standard deviation of
ulation is described in Section IV-A. Its dimension is . evaluation values are as follows, respectively:
The searching procedures of the proposed PSO-PID con-
troller were shown as below.
Step 1) Specify the lower and upper bounds of the three con- (16)
troller parameters and initialize randomly the indi-
viduals of the population including searching points,
(17)
velocities, s, and .
Step 2) For each initial individual of the population,
employ the Routh-Hurwitz criterion to test the where is the evaluation value of the individual and
closed-loop system stability and calculate the values is the population size.
388 IEEE TRANSACTIONS ON ENERGY CONVERSION, VOL. 19, NO. 2, JUNE 2004

Fig. 2. Block diagram of an AVR system with a PSO-PID controller.

TABLE I TABLE II
RANGE OF THREE CONTROLLER PARAMETERS BEST SOLUTION USING PSO-PID CONTROLLER
WITH THE DIFFERENT VALUES

Fig. 4. Convergence tendency of the PSO-PID controller.

Fig. 3. Terminal voltage step response of an AVR system without PID
controller.

VI. NUMERICAL EXAMPLES AND RESULTS

A. AVR System Parameters
To verify the efficiency of the PSO-PID controller, a practical
high-order AVR system was tested. The AVR system has the
following parameters:
The block diagram of the AVR system with a PID controller
is shown in Fig. 2. The lower and upper bounds of the three
controller parameters were as shown in Table I.
Fig. 3 shows the original terminal voltage step response of the
AVR system without a PID controller. To simulate this case, we Fig. 5. Terminal voltage step response of an AVR system with the PSO-PID
controller.
found that , , , and
. convergence characteristics of the PSO-PID controller and ter-
minal voltage step response of the AVR system at different sim-
B. Performance of the PSO-PID Controller
ulation conditions, respectively. As can be seen, through about
According to the trials, the following PSO parameters are 50 iterations (50 generations), the PSO method can prompt con-
used for verifying the performance of the PSO-PID controller vergence and obtain good evaluation value. These results show
in searching the PID controller parameters: that the PSO-PID controller can search optimal PID controller
• the member of each individual is , , and ; parameters quickly and efficiently.
• population size ;
• inertia weight factor is set by (12), where C. Comparison of Two Proposed Controllers
and ; In order to emphasize the advantages of the proposed
• the limit of change in velocity , PSO-PID controller, we also implemented the GA-PID con-
, and ; troller derived from the real-value GA method with the Elitism
• acceleration constant and . scheme [5], [6]. We have compared the characteristics of
The PSO-PID controller shown in Fig. 2 then replaced the the two controllers using the same evaluation function and
PID controller; the simulation results that showed the best so- individual definition. The following real-value GA parameters
lution were summarized in Table II. Figs. 4 and 5 showed the have been used:
GAING: A PARTICLE SWARM OPTIMIZATION APPROACH FOR OPTIMUM DESIGN OF PID CONTROLLER 389

TABLE III
SUMMARY OF SIX SIMULATION RESULTS

Fig. 8. Terminal voltage step response of an AVR system with different

controllers (Example III, = 1:0, generations = 150).

Fig. 6. Terminal voltage step response of an AVR system with different

controllers (Example I, = 1:0, generations = 50).

Fig. 9. Terminal voltage step response of an AVR system with different

controllers (Example IV, = 1:5, generations = 50).

Fig. 7. Terminal voltage step response of an AVR system with different

controllers (Example II, = 1:0, generations = 100).

• the members of each individual are , , and ;

• population size ;
• crossover rate ; Fig. 10. Terminal voltage step response of an AVR system with different
• mute rate ; controllers (Example V, = 1:5, generations = 100).
• .
Two proposed controllers and their performance evaluation Figs. 6–11 show the terminal voltage step response of the
criteria in the time domain were implemented by Matlab and AVR system of the six simulation examples. As can be seen,
control system toolbox, and executed on a Pentium III 550 per- the PSO-PID controller could create very perfect step response
sonal computer with 256-MB RAM. of the AVR system, indicating that the PSO-PID controller is
1) Terminal Voltage Step Response: There were six simula- better than the GA-PID controller.
tion examples to evaluate the performance of both the PSO-PID 2) Convergence Characteristic: Under the same conditions,
and the GA-PID controllers. In each simulation example, the we performed simulations using the two proposed controllers to
weighting factor in the performance criterion and the number compare their convergence characteristics. Fig. 12 showed their
of iterations (generations) were set as follows: convergence properties. As can be seen, the PSO-PID controller
The simulation results that showed the best solution were has better evaluation value than the GA-PID controller. The re-
summarized in Table III. As can be seen, both controllers sults showed that the PSO-PID controller could obtain higher
could give good PID controller parameters in each simulation quality solution, indicating the drawbacks of GA method men-
example, providing good terminal voltage step response of the tioned in [10] and [14].
AVR system. Table III also shows the four performance criteria We also performed 100 trials for both proposed controllers
in the time domain of each example. As revealed by the above with different random number to observe the variation in their
four performance criteria, the PSO-PID controller has better evaluation values. In addition, the maximum, minimum, and av-
performance than the GA-PID controller. erage evaluation values were obtained by the two methods. The
390 IEEE TRANSACTIONS ON ENERGY CONVERSION, VOL. 19, NO. 2, JUNE 2004

TABLE IV
COMPARISON OF THE EVALUATION VALUE BETWEEN BOTH METHODS ( = 1:5, GENERATION = 100)

Fig. 11. Terminal voltage step response of an AVR system with different
controllers (Example VI, = 1:5, generations = 150).
Fig. 14. Convergence tendency of both  and  of evaluation values using
both methods (Example IV, = 1:5, generations = 150).

TABLE V
COMPARISON OF COMPUTATION EFFICIENCY OF BOTH METHODS ( = 1:5)

3) Dynamic Convergence Behavior: In addition, during the

evolutionary processing of the two proposed methods, after each
iteration, the mean value and the standard deviation of
the evaluation values of all individuals in the population were
Fig. 12. Convergence tendency of the evaluation value of both methods. recorded for observing the dynamic convergence behavior of the
individuals in population. Fig. 14 displays the recorded data in
Example VI. As seen in the simulation, with the same number of
iterations, though both controllers can obtain stable mean evalu-
ation value under the same evaluation function and simula-
tion conditions, the GA-PID controller brings premature conver-
gence such that the evaluation value and mean value are smaller.
Conversely, the PSO-PID controller has better evaluation value
and mean value, showing that it can achieve better accuracy. Si-
multaneously, we can also find that the convergence tendency of
the standard deviation of evaluation values in the PSO-PID
controller is much faster than the GA-PID controller, because
the latter presented fluctuation resulted from the mutation in the
Fig. 13. Comparison of the statistical evaluation values of both methods GA method. This can prove that the PSO method has better con-
(100 trials).
vergence efficiency in solving the power optimization problems.
4) Computation Efficiency: The comparison of computation
results were shown in Fig. 13 and Table IV. As can be seen, the efficiency of both methods is shown in Table V. As can be seen,
evaluation values of the PSO-PID controller generated fluctua- because the PSO method does not perform the selection and
tion in a small range , thus verifying that the crossover operations in evolutionary processes, it can save some
PSO-PID controller has better convergence characteristic. computation time compared with the GA method, thus proving
GAING: A PARTICLE SWARM OPTIMIZATION APPROACH FOR OPTIMUM DESIGN OF PID CONTROLLER 391

that the PSO-PID controller is more efficient than the GA-PID [2] T. L. Seng, M. B. Khalid, and R. Yusof, “Tuning of a neuro-fuzzy con-
controller. troller by genetic algorithm,” IEEE Trans. Syst., Man, Cybern. B, vol.
29, pp. 226–236, Apr. 1999.
[3] R. A. Krohling and J. P. Rey, “Design of optimal disturbance rejection
VII. DISCUSSION AND CONCLUSION PID controllers using genetic algorithm,” IEEE Trans. Evol. Comput.,
vol. 5, pp. 78–82, Feb. 2001.
This paper presents a novel design method for determining [4] Y. Mitsukura, T. Yamamoto, and M. Kaneda, “A design of self-tuning
the PID controller parameters using the PSO method. The PID controllers using a genetic algorithm,” in Proc. Amer. Contr. Conf.,
San Diego, CA, June 1999, pp. 1361–1365.
proposed method integrates the PSO algorithm with the new [5] T. Kawabe and T. Tagami, “A real coded genetic algorithm for matrix
time-domain performance criterion into a PSO-PID controller. inequality design approach of robust PID controller with two degrees of
Through the simulation of a practical AVR system, the results freedom,” in Proc. 12th IEEE Int. Symp. Intell. Contr., Istanbul, Turkey,
July 1997, pp. 119–124.
show that the proposed controller can perform an efficient [6] R. A. Krohling, H. Jaschek, and J. P. Rey, “Designing PI/PID controller
search for the optimal PID controller parameters. for a motion control system based on genetic algorithm,” in Proc. 12th
IEEE Int. Symp. Intell. Contr., Istanbul, Turkey, July 1997, pp. 125–130.
In addition, in order to verify it being superior to the GA [7] D. P. Kwok and F. Sheng, “Genetic algorithm and simulated annealing
method, many performance estimation schemes are performed, for optimal robot arm PID control,” in Proc IEEE Conf. Evol. Comput.,
such as Orlando, FL, 1994, pp. 707–713.
[8] T. Ota and S. Omatu, “Tuning of the PID control gains by GA,” in Proc.
• multiple simulation examples for their terminal voltage IEEE Conf. Emerging Technol. Factory Automation, Kauai, HI, Nov.
step responses; 1996, pp. 272–274.
• convergence characteristic of the best evaluation value; [9] A. H. Jones and P. B. D. Oliveira, “Genetic auto-tuning of PID con-
trollers,” in Proc. Inst. Elect. Eng. Conf. Genetic Algorithm Eng. Syst.:
• dynamic convergence behavior of all individuals in popu- Innovations Applicat., Sept. 1995, pp. 141–145.
lation during the evolutionary processing; [10] D. B. Fogel, Evolutionary Computation: Toward a New Philosophy of
• computation efficiency. Machine Intelligence, 2 ed. New York: IEEE Press, 2000.
[11] J. Kennedy and R. Eberhart, “Particle swarm optimization,” in Proc.
It is clear from the results that the proposed PSO method can IEEE Int. Conf. Neural Networks, vol. IV, Perth, Australia, 1995, pp.
avoid the shortcoming of premature convergence of GA method 1942–1948.
and can obtain higher quality solution with better computation [12] Y. Shi and R. Eberhart, “A modified particle swarm optimizer,” in Proc.
IEEE Int. Conf. Evol. Comput., Anchorage, AK, May 1998, pp. 69–73.
efficiency. Therefore, the proposed method has more robust sta- [13] Y. Shi and R. C. Eberhart, “Empirical study of particle swarm optimiza-
bility and efficiency, and can solve the searching and tuning tion,” in Proc. IEEE Int. Conf. Evol. Comput., Washington, DC, July
problems of PID controller parameters more easily and quickly 1999, pp. 1945–1950.
[14] R. C. Eberhart and Y. Shi, “Comparison between genetic algorithms and
than the GA method. particle swarm optimization,” in Proc. IEEE Int. Conf. Evol. Comput.,
Different PSO optimization parameters are required for Anchorage, AK, May 1998, pp. 611–616.
solving different problems in practical application, such as [15] P. J. Angeline, “Using selection to improve particle swarm optimiza-
tion,” in Proc. IEEE Int. Conf. Evol. Comput., Anchorage, AK, May
the number of agents (individuals), weight factors and 1998, pp. 84–89.
, acceleration factors and , and the limit of change [16] H. Yoshida, K. Kawata, and Y. Fukuyama, “A particle swarm optimiza-
in velocity . Hence, how to select suit parameters for the tion for reactive power and voltage control considering voltage secu-
rity assessment,” IEEE Trans. Power Syst., vol. 15, pp. 1232–1239, Nov.
target problem, such as the sensitivity analysis of optimization 2000.
parameters for finding the best parameters, is one of our future [17] S. Naka, T. Genji, T. Yura, and Y. Fukuyama, “Practical distribution state
works. estimation using hybrid particle swarm optimization,” in Proc. IEEE
Power Eng. Soc. Winter Meeting, vol. 2, 2001, pp. 815–820.
[18] G. Cheng, Genetic Algorithms & Engineering Design. New York:
ACKNOWLEDGMENT Wiley, 1997.
[19] H. Saadat, Power System Analysis. New York: McGraw-Hill, 1999.
The authors acknowledge the technical support from Taiwan
Power Company.
Zwe-Lee Gaing (M’02) received the Ph.D. degree from National Sun Yat-Sen
University, Kaohsiung, Taiwan, R.O.C., in 1996.
REFERENCES Currently, he is an Associate Professor of Electrical Engineering at Kao-Yuan
[1] A. Visioli, “Tuning of PID controllers with fuzzy logic,” Proc. Inst. Institute of Technology, Kaohsiung, Taiwan. His research interests include the
Elect. Eng. Contr. Theory Applicat., vol. 148, no. 1, pp. 1–8, Jan. 2001. field of artificial intelligence with application to control system.

View publication stats