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

net/publication/224560544

Real-coded genetic algorithm and fuzzy logic approach for real-time tuning of
proportional-integral - derivative controller in automatic voltage regulator
system

Article in IET Generation, Transmission and Distribution · August 2009

DOI: 10.1049/iet-gtd.2008.0287 · Source: IEEE Xplore

CITATIONS READS

101 10,153

2 authors, including:

D. Devaraj
Kalasalingam University
318 PUBLICATIONS 5,227 CITATIONS

SEE PROFILE

All content following this page was uploaded by D. Devaraj on 08 May 2014.

The user has requested enhancement of the downloaded file.

 www.ietdl.org

Published in IET Generation, Transmission & Distribution

Received on 19th June 2008
Revised on 10th February 2009
doi: 10.1049/iet-gtd.2008.0287

ISSN 1751-8687

Real-coded genetic algorithm and fuzzy

logic approach for real-time tuning of
proportional –integral – derivative controller
in automatic voltage regulator system
D. Devaraj B. Selvabala
Power System Automation Laboratory, Kalasalingam University, Krishnankoil 626190, Tamilnadu, India
E-mail: [email protected]

Abstract: Optimal tuning of proportional – integral – derivative (PID) controller parameters is necessary for the
satisfactory operation of automatic voltage regulator (AVR) system. This study presents a combined genetic
algorithm (GA) and fuzzy logic approach to determine the optimal PID controller parameters in AVR system.
The problem of obtaining the optimal PID controller parameters is formulated as an optimisation problem and
a real-coded genetic algorithm (RGA) is applied to solve the optimisation problem. In the proposed RGA, the
optimisation variables are represented as floating point numbers in the genetic population. Further, for
effective genetic operation, the crossover and mutation operators which can deal directly with the floating
point numbers are used. The proposed approach has resulted in PID controller with good transient response.
The optimal PID gains obtained by the proposed GA for various operating conditions are used to develop the
rule base of the Sugeno fuzzy system. The developed fuzzy system can give the PID parameters on-line for
different operating conditions. The suitability of the proposed approach for PID controller tuning has been
demonstrated through computer simulations in an AVR system.

Osh overshoot
Nomenclature Ess steady-state error
Kp proportional gain of PID controller Ts settling time
Ki integral gain of PID controller Tr rising time
Kd derivative gain of PID controller
Ka amplifier gain
Ke exciter gain 1 Introduction
Kg generator gain
The main objective of the automatic voltage regulator
Ks sensor gain
(AVR) is to control the terminal voltage by adjusting the
ta amplifier time constant generator exciter voltage. The AVR must keep track of the
te exciter time constant generator terminal voltage all the time and under any load
tg generator time constant condition, working in order to keep the voltage within
pre-established limits. Despite significant studies in the
ts sensor time constant
development of advanced control schemes, the classical
DVt incremental change in terminal voltage proportional – integral – derivative (PID) controllers [1 – 5]
DVref incremental change in reference voltage remain the controllers of choice to control the AVR

IET Gener. Transm. Distrib., 2009, Vol. 3, Iss. 7, pp. 641 – 649 641
doi: 10.1049/iet-gtd.2008.0287 & The Institution of Engineering and Technology 2009

Authorized licensed use limited to: Arulmigu Kalasalingam College of Engineering. Downloaded on August 20, 2009 at 05:01 from IEEE Xplore. Restrictions apply.
www.ietdl.org

because of its simple structure and robustness to variations of applied to obtain the optimal PID parameters of an AVR
the system parameters. system.

Proper selection of the PID controller parameters is 2 Modelling of automatic voltage

necessary for the satisfactory operation of the AVR. regulator system
Traditionally, the PID controller parameters are evaluated
using Ziegler–Nichols (ZN) [1, 2] and Cohen Coon In a synchronous generator, the terminal voltage is maintained
methods [6, 7]. In both these methods, the parameters of the constant at various levels by using an AVR. The AVR system
controller are obtained for an operating point where the consists of four major components, namely amplifier, exciter,
model can be considered linear. This implies that there is generator and sensor. Fig. 1 illustrates the block diagram
sub-optimal tuning when a process operates outside the representation of the AVR system. The transfer functions of
validity zone of the model. Internal model control [8] the individual components are given in Table 1 along with
overcomes the above said problem but its design calculations the limits of the parameters.
could be complicated for higher order process.
An increase in the reactive power load of the generator is
accompanied by a drop in the terminal voltage magnitude.
Alternatively, numerical optimisation techniques like
The voltage magnitude is sensed by a sensor. This voltage
gradient descent technique can be applied to obtain the
is compared with a dc set point signal to generate the error
parameters of the PID controllers. They are computationally
signal. A PID controller is used to reduce the error and to
fast, but with a non-monotonic solution surface these
improve the dynamic response. The PID controller is a
methods are highly sensitive to starting points and frequently
combination of the proportional, integral and derivative
converge to local optimum solutions or diverge altogether. In
control mechanisms that when used together effectively
[9], an optimal PID controller for a general second-order
stabilise the manipulated variable at the set point. The PID
system has been developed using linear-quadratic regulator
controller transfer function is given by
(LQR) technique. This approach requires the proper selection
of weighting functions for satisfactory performance. Recently, Ki
evolutionary computation techniques such as genetic G(s) ¼ Kp þ þ Kd s (1)
s
algorithm (GA) [10–13] and particle swarm optimisation
[14] have been applied to obtain the optimal controller The transfer function of AVR system with PID control is
parameters. GA is a global search algorithm based on the given by (as shown in (2))
principle of ‘survival of the fittest’. Devaraj et al. [12]
proposed an enhanced genetic algorithm (GA) for PI The AVR quality influences the voltage level during
controller tuning in pH process. A hybrid GA and bactorial steady-state operation and also reduces the voltage
foraging approach was proposed in [13] to tune the PID oscillations during transient periods, affecting the overall
controller of an AVR. Gaing [14] has proposed a novel system stability.
design method for determining the PID controller
parameters of the AVR system using the particle swarm
optimisation (PSO) method. PSO is a population-based 3 Optimisation of controller
optimisation algorithm which is inspired by social behaviour parameters
patterns of organisms such as bird flocking and fish schooling.
Proper selection of PID controller parameters is necessary for
the satisfactory operation of the system. In this work, the
Both GA and PSO suffer from computational burden and problem of PID controller parameter selection is
memory requirement and so they are not suitable for on-line formulated as an optimisation problem, the objective
applications. To overcome the above difficulties, this paper function of which is given by
proposes Sugeno fuzzy model [15] for on-line tuning of
PID controller. The optimal PID parameters required to Min F (Kp , Ki , Kd ) ¼ (1  eb )(Osh þ Ess ) þ eb (ts  tr )
formulate the fuzzy rule table are generated by employing
(3)
the real-coded genetic algorithm (RGA). In the RGA, the
optimisation variables are represented as floating point The above objective function uses a combination of transient
numbers instead of the binary string which is followed in response including rise time, overshoot, settling time and
the conventional binary-coded GA. Further, crossover and steady-state error. By selecting the proper value of the
mutation operators which can deal directly with the weighting factor b, the performance criterion can be made
floating point numbers are used. The proposed approach is to satisfy the designer requirements. The above optimisation

DVt (s) (s2 Kd þ sKp þ Ki )(Ka Ke Kg )(1 þ sts )

¼ (2)
DVref (s) s(1 þ sta )(1 þ ste )(1 þ stg )(1 þ sts ) þ (Ka Ke Kg Ks )(s2 Kd þ sKp þ Ki )

642 IET Gener. Transm. Distrib., 2009, Vol. 3, Iss. 7, pp. 641– 649
& The Institution of Engineering and Technology 2009 doi: 10.1049/iet-gtd.2008.0287

Authorized licensed use limited to: Arulmigu Kalasalingam College of Engineering. Downloaded on August 20, 2009 at 05:01 from IEEE Xplore. Restrictions apply.
 www.ietdl.org

Figure 1 Block diagram of AVR system along with PID controller

problem is subjected to the following constraints 4.1 Reproduction

Reproduction is a method that stochastically selects the
Kpmin  Kp  Kpmax
individuals from the population according to their fitness;
Kimin  Ki  Kimax (4) higher the fitness, more chance an individual has to be
selected for the next generation. There are three main types
Kdmin  Kd  Kdmax of selection methods: fitness proportionate selection, ranking
method and tournament selection. Tournament selection [17]
A RGA is applied to the above optimisation problem to search is used in this work. In tournament selection, ‘n’ individuals
for the optimum value of the controller parameters. The are selected randomly from the population, and the best of
details of the proposed GA are given in the next section. the ‘n’ is inserted into the new population for further genetic
processing. This procedure is repeated until the mating pool
is filled. Tournaments are often held between pairs of
4 Proposed GA individuals, although larger tournaments can be used.
GAs [16] are search algorithms based on the mechanics of
natural selection and genetics. They combine solution 4.2 Crossover operation
evaluation with randomised, structured exchange of
information between solutions to obtain optimality. The crossover operator is mainly responsible for the global
Starting with an initial population, the GA exploits the search property of the GA. Crossover basically combines
information contained in the present population and substructures of two parent chromosomes to produce new
explores new individuals by generating offspring using the structures, with the selected probability typically in the
three genetic operators namely, reproduction, crossover and range of 0.6– 1.0. The Blend crossover operator (BLX-a)
mutation which can then replace the members of the old [17] is applied in this work.
generation. After several generations, the algorithm
converges to the best chromosome, which hopefully Fig. 2 illustrates the BLX-a crossover operation for the
represents the optimum or near optimal solution. In the one-dimensional case. In the BLX-a crossover, the off
traditional binary-coded GA, the decision variables of the spring y is sampled from the space [e1 , e2] as follows
problem are represented by a fixed-length string of binary 
bits (0, 1). In this representation, the resolution of the e1 þ r(e2  e1 ) if umin  y  umax
y¼ (5)
solution depends on the number of bits used to represent repeat sampling otherwise
the variables. Further, the coding of real-valued variables in
finite-length strings causes a number of difficulties. To where
e1 ¼ u1  a(u2  u1 ) (6)
overcome these difficulties, in this paper, the decision
variables are represented in their natural form. Also, where
crossover and mutation operators which can operate directly e2 ¼ u2 þ a(u2  u1 ) (7)
with floating point numbers are used. The details of the
genetic operators used in the proposed GA are given below. r is the uniform random number [ [0, 1].

Table 1 Transfer function of AVR components

Component Transfer function Parameter limits

amplifier TFamplifier ¼ Ka/1 þ tas 10 , Ka , 40; 0.02 s , ta , 1 s
exciter TFexciter ¼ Ke/1 þ tes 1 , Ke , 10; 0.4 s , te ,1 s
generator TFgenerator ¼ Kg/1 þ tgs Kg depends on load (0.7 – 1.0); 1 s , tg ,2 s
sensor TFsensor ¼ Ks/1 þ tss 0.001 s , ts , 0.06 s

IET Gener. Transm. Distrib., 2009, Vol. 3, Iss. 7, pp. 641 – 649 643
doi: 10.1049/iet-gtd.2008.0287 & The Institution of Engineering and Technology 2009

Authorized licensed use limited to: Arulmigu Kalasalingam College of Engineering. Downloaded on August 20, 2009 at 05:01 from IEEE Xplore. Restrictions apply.
www.ietdl.org

Figure 2 Schematic representation of BLX-a crossover

It is to be noted that e1 and e2 will lie between umin and 5.2 Fitness function
umax , the variable’s lower and upper bounds, respectively. In
a number of test problems, it was observed that a ¼ 0.5 The performance of each individual in the population is
provides good results. One interesting feature of this type evaluated according to its ‘fitness’, which is defined as the
of crossover operator is that the created point depends on non-negative figure of merit to be maximised. It is
the location of both parents. If both parents are close to associated directly with the objective function value.
each other, the new point will also be close to the parents. Evaluation of the individual is accomplished by calculating
On the other hand, if parents are far from each other, the the performance criteria given by (3) for the problem
search is more like a random search. using the parameter set. The result of the performance
criteria calculation is used to calculate the fitness value of
the individual. The fitness function is the reciprocal of the
4.3 Mutation operation performance criterion F(Kp ,Ki ,Kd) given in (3). Hence,
The mutation operator is used to inject new genetic material the minimisation of performance criteria given by (3) is
into the population. Mutation randomly alters a variable with transformed to a fitness function to be maximised as
a small probability. ‘Uniform mutation’ operator is used in
this work. In uniform mutation, the variable is set to a k
Fitness ¼ (8)
uniform random number between the variable’s lower and F (Kp , Ki , Kd )
upper limits.
where k is a constant. This is used to amplify (1/F ), the value
of which is usually small, so that the fitness value of the
5 GA implementation for PID chromosome will be in a wider range.
controller tuning
While applying GA to obtain the optimal PID controller 6 Review of Sugeno fuzzy model
parameters, two main issues need to be addressed:
Fuzzy logic was first developed by Zadeh in the mid-1960s to
† representation of the decision variables and provide a mathematical basis for human reasoning. Fuzzy logic
[15] uses fuzzy set theory, in which a variable is a member of
† formation of the fitness function. one or more sets, with a specified degree of membership.
The degree of membership in a set is expressed by a number
5.1 Variable representation between 0 and 1. 0 means entirely not in the set, 1 means
completely in the set, and a number in between means
Each individual in the genetic population represents a partially in the set. Mathematically, a fuzzy set A in the
candidate solution. For the PID controller tuning problem, universe of discourse X is defined to be a set of ordered pairs
the elements of the solution consist of proportional gain
(Kp), integral gain (Ki) and derivative gain (Kd). These A ¼ {(x, mA (x))jx [ X } (9)
variables are represented as floating point numbers in the
proposed GA population. With floating point where mA(x) is called the membership function of x in A. The
representation, an individual in the GA population for parameterisable membership functions most commonly used
computing the optimal PID gains will look like the following in practice are the triangular membership function and the
trapezoidal membership function. Fuzzy logic when applied
0:937
|fflffl{zfflffl} 0:242
|fflffl{zfflffl} 0:320
|fflffl{zfflffl} to computers allows them to emulate the human reasoning
Kp Ki Kd process, quantify imprecise information, make decisions
based on vague and incomplete data, yet by applying a
With the direct representation of the solution variables, the ‘defuzzification’ process, arrive at definite conclusions.
computer space required to store the population is reduced.
Moreover, the efficiency of the GA is increased as there is There are three main types of fuzzy logic systems, namely,
no need to convert the solution variables to the binary string. Mamdani, Sugeno and Tsukamoto fuzzy logic systems. In

644 IET Gener. Transm. Distrib., 2009, Vol. 3, Iss. 7, pp. 641– 649
& The Institution of Engineering and Technology 2009 doi: 10.1049/iet-gtd.2008.0287

Authorized licensed use limited to: Arulmigu Kalasalingam College of Engineering. Downloaded on August 20, 2009 at 05:01 from IEEE Xplore. Restrictions apply.
 www.ietdl.org

the Sugeno fuzzy model, which is followed in this work, the

fuzzy rule is expressed as

If x is A and y is B then z ¼ f (x, y) (10)

where x and y are input variables, A and B are fuzzy sets in the
antecedent and f (x, y) is a crisp function in the consequent.
The fuzzy sets of each variable are described by appropriate
membership functions. A set of such rules form the heart
of the fuzzy logic system.

For a specific input signal condition, the fuzzy system

determines the rules to be fired and then computes the
effective output. For this, first the minimum of the Figure 4 Step response of change in terminal voltage with
membership functions of the inputs (wi) is obtained for ZN PID controller
each of the rules. This value is the firing value for a
particular rule. Then the overall output is determined by
weighted average of individual rule outputs given by of PID controller is shown in Fig. 4. From the figure, it is
observed that the response of the AVR system with PID
PM controller tuned using ZN method posses more than one
wi zi
z ¼ Pi¼1
M
(11) oscillatory mode and has large settling time. This shows that
i¼1 wi the system has not been tuned to its optimum.

In the present work, the Sugeno fuzzy system is used to

estimate the parameters of the PID controller under various 7.1 Performance of RGA–PID controller
operating conditions.
Next, the proposed GA was applied to obtain the optimal
PID controller parameters. The software for the proposed
GA was written in MATLAB and executed on a PC with
7 Simulation results 2.4 MHZ and 256 MB RAM. Proportional gain (Kp),
The proposed methodology for PID controller tuning was Integral gain (Ki) and derivative gain (Kd) are taken as the
tested on an AVR system. The AVR system consists of optimisation variables. They are represented as floating
amplifier, exciter, generator and sensor. The parameters of the point numbers in the GA population. The initial
AVR system are chosen as Ka ¼ 10, Ke ¼ Kg ¼ Ks ¼ 1.0, population is generated randomly between the variable’s
ta ¼ 0.1, te ¼ 0.4, ts ¼ 0.01 and tg ¼ 1.0. Only Kg and tg lower and upper limits. The fitness function given by (8) is
are load dependent. The AVR system was simulated in used to evaluate the fitness value of each set of controller
MATLAB Simulink. The MATLAB-Simulink model of parameters. Simulation was conducted with different values
AVR system along with PID controller is shown in Fig. 3. of b. The performance of GA for various values of
The PID controller was tuned using the ZN method. A step crossover and mutation probabilities in the ranges 0.6– 1.0
reference voltage of 0.01 p.u. is applied and the step response and 0.001 – 0.1, respectively, was evaluated. The best results
of change in terminal voltage of AVR system in the presence are obtained with the following control parameters.

Figure 3 MATLAB-Simulink model of AVR system along with PID controller

IET Gener. Transm. Distrib., 2009, Vol. 3, Iss. 7, pp. 641 – 649 645
doi: 10.1049/iet-gtd.2008.0287 & The Institution of Engineering and Technology 2009

Authorized licensed use limited to: Arulmigu Kalasalingam College of Engineering. Downloaded on August 20, 2009 at 05:01 from IEEE Xplore. Restrictions apply.
www.ietdl.org

Number of generations: 50

population size: 30

crossover probability: 0.8

mutation probability: 0.05

The proposed GA took 72 s to reach the optimal solution.

Fig. 5 shows the convergence characteristics of GA
algorithm. It can be seen that the fitness value increases
rapidly in the first 15 generations of the GA. During this
stage, the GA concentrates mainly on finding feasible
solutions to the problem. Then the value increases slowly Figure 6 Step response of RGA – PID change in terminal
and settles down near the optimum value with most of the voltage
individuals in the population reaching that point.
Table 3. The terminal voltage responses with these
The optimal values of the controller parameters obtained methods are also given in Fig. 6. On comparing these
using the proposed GA for different values of b are given in results, it is found that the proposed GA approach has
Table 2. The time-domain performance indices of the resulted in minimum values of rise time, settling time and
system, namely rise time, settling time, steady-state error and overshoot are less when compared with the results obtained
overshoot are also given in Table 2. The system response with using LQR method and binary-coded GA. In addition, the
optimal values of PID parameters is given in Fig. 6. The steady-state error with RGA – PID is low compared to
transient response of the AVR system has improved other two methods. Also the proposed algorithm took less
significantly and the response is faster than the one shown in time for convergence compared to the binary-coded GA.
Fig. 4. These results show that the proposed approach is able
to search the optimal values of the PID controller. To analyse the performance of the AVR system under severe
disturbances, a three phase fault is applied at the generator
For comparison, LQR technique and a binary-coded GA terminal and the response of the system was observed. Fig. 7
were applied to obtain the parameters of the PID controller. shows the system response for the above contingency with
The results obtained by these methods are also given in PID controller tuned using the RGA. It can be observed
that the controller is able to suppress the oscillations in the
terminal voltage and provide good damping characteristics.

7.2 Development of Sugeno fuzzy model

for online tuning
A Sugeno fuzzy logic model was developed to obtain the optimal
PID parameters during real-time operation. Kg and tg are the
input to the fuzzy model and the values of Kp , Ki and Kd are
the output. Four fuzzy sets, namely, ‘low (L)’, ‘medium low
(ML)’, ‘medium high (MH)’ and ‘high (H)’ are defined for the
variable Kg . Similarly, the fuzzy sets defined for the variable tg
are ‘very low (VL)’, ‘low (L)’, ‘medium low (ML)’, ‘medium
high (MH)’, ‘high (H)’ and ‘very high (VH)’. They are
associated with overlapping triangular membership functions.
To formulate the fuzzy rule table the value of Kg is varied from
Figure 5 Convergence characteristics of the RGA 0.7 to 1.0 in steps of 0.1 and tg is varied from 1 to 2 in steps

Table 2 Optimal PID gains and transient response parameters

b Kp Ki Kd Ts(s) Tr(s) Osh Ess(1025)

0.5 1.000 0.1448 0.7019 2.5595 1.9511 0.0158 46.911
1.0 1.000 0.1448 0.7019 2.5595 1.9511 0.0158 46.911
1.5 0.682 0.266 0.179 1.2682 1.0668 0.0004 4.3386

646 IET Gener. Transm. Distrib., 2009, Vol. 3, Iss. 7, pp. 641– 649
& The Institution of Engineering and Technology 2009 doi: 10.1049/iet-gtd.2008.0287

Authorized licensed use limited to: Arulmigu Kalasalingam College of Engineering. Downloaded on August 20, 2009 at 05:01 from IEEE Xplore. Restrictions apply.
 www.ietdl.org

Table 3 Comparison of PID gains and transient response parameters

Method Kp Ki Kd Ts(s) Tr(s) Osh Ess(1025)

LQR 1.01 0.5 0.1 2.3354 0.5004 0.3605 15.007
GA 0.5692 0.2484 0.1258 1.7019 0.8093 0.0586 8.2941
RGA 0.682 0.266 0.179 1.2682 1.0668 0.0004 4.3386

During real-time operation, corresponding to the present

operating condition, the values of Kg and tg are found out.
For this value of Kg and tg , the optimal values of Kp , Ki
and Kd can be computed using the fuzzy rule table and the
Sugeno inference system explained in Section 5. The
optimal gains and transient response parameters obtained
using the Sugeno model for a new set of operating
condition (off-nominal) are given in Table 5. To validate
the results obtained by the fuzzy model, RGA was applied
for the same values of Kg and tg and the results are given
in Table 6. From the table, it is observed that the values of
controller parameters obtained are almost same in all cases.
Fig. 8 shows the response of the system with PID
Figure 7 Terminal voltage response under three phase fault controller values obtained using the fuzzy model and RGA
for Kg ¼ 0.77 and tg ¼ 1.5. From the figure, it is observed
of 0.2. For each combination of Kg and tg, , the proposed GA is that the terminal voltage response is similar in both cases.
applied to obtain the optimal values of Kp , Ki and Kd. The fuzzy This demonstrates the suitability of the proposed approach
rule table formulated for Kp , Ki and Kd using the above approach to obtain the optimal PID gains during real-time operation
is given in Table 4 as (a), (b) and (c), respectively. of the system.

Table 4 Sugeno fuzzy rule table

tg Very low Low Medium low Medium high High Very high
Kg 1 1.2 1.4 1.6 1.8 2
(a) For proportional gain Kp
low (0.7) 0.8574 0.8574 0.7246 0.5408 0.6102 0.6102
medium low (0.8) 0.4752 0.7246 0.6848 0.7246 0.5864 0.6030
medium high (0.9) 0.5321 0.7246 0.8080 0.7379 0.5864 0.5864
high (1.0) 0.5408 0.8574 0.7246 0.7246 0.7379 0.5864
(b) For integral gain Ki
low (0.7) 0.2862 0.3601 0.3601 0.3601 0.3107 0.3107
medium low (0.8) 0.1719 0.2862 0.3601 0.3601 0.3601 0.3107
medium high (0.9) 0.1719 0.2505 0.3601 0.3601 0.3601 0.3601
high (1.0) 0.1719 0.2862 0.2862 0.3601 0.3601 0.3601
(c) For derivative gain Kd
low (0.7) 0.2187 0.2078 0.1643 0.1643 0.3578 0.4207
medium low (0.8) 0.2078 0.1643 0.1643 0.1643 0.2049 0.3547
medium high (0.9) 0.1351 0.1643 0.1757 0.1643 0.1643 0.2049
high (1.0) 0.1351 0.2049 0.1643 0.1643 0.1643 0.1757

IET Gener. Transm. Distrib., 2009, Vol. 3, Iss. 7, pp. 641 – 649 647
doi: 10.1049/iet-gtd.2008.0287 & The Institution of Engineering and Technology 2009

Authorized licensed use limited to: Arulmigu Kalasalingam College of Engineering. Downloaded on August 20, 2009 at 05:01 from IEEE Xplore. Restrictions apply.
www.ietdl.org

Table 5 Optimal gains and transient parameters for off-nominal operating condition

Kg tg Kp Ki Kd Ts(s) Tr(s) Osh(1024) Ess(1025)

0.77 1.50 0.723 0.36 0.164 1.1100 0.9699 1.1670 8.7792
0.79 1.15 0.682 0.266 0.179 1.1404 0.9698 0.4262 8.4564
0.85 1.30 0.736 0.314 0.167 0.8628 0.8356 8.3670 6.2673
0.75 1.67 0.621 0.361 0.205 1.6344 1.3304 0.0109 1.5927
0.99 1.45 0.731 0.31 0.165 0.8353 0.8082 2.2541 5.3307
0.99 1.96 0.614 0.36 0.176 1.4373 0.9671 0.0051 9.7682

Table 6 Optimal PID gains and transient response using RGA and SFL techniques

Kg tg Type Kp Ki Kd Ts(s) Tr(s) Osh(1024) Ess(1025)

0.77 1.50 RGA 0.7246 0.3601 0.1643 1.1960 0.8083 2.157 6.1102
SFL 0.723 0.36 0.164 1.1100 0.9699 1.1670 8.7792
0.79 1.15 RGA 0.6598 0.2927 0.1743 1.1103 0.9404 0.2889 7.5400
SFL 0.682 0.266 0.179 1.1404 0.9698 0.4262 8.4564
0.85 1.30 RGA 0.7379 0.2862 0.1643 0.9031 0.8354 6.3345 8.0084
SFL 0.736 0.314 0.167 0.8628 0.8356 8.3670 6.2673
0.75 1.67 RGA 0.6321 0.3601 0.2643 1.8021 1.0740 0.0592 2.0006
SFL 0.621 0.361 0.205 1.6344 1.3304 0.0109 1.5927
0.99 1.45 RGA 0.7080 0.3601 0.1652 0.7646 0.7646 1.960 4.8988
SFL 0.731 0.31 0.165 0.8353 0.8082 2.2541 5.3307
0.99 1.96 RGA 0.6030 0.3601 0.1757 1.4093 0.9997 0.0416 1.3932
SFL 0.614 0.36 0.176 1.4373 0.9671 0.0051 9.7682

8 Conclusion
This paper has proposed a RGA and Sugeno fuzzy logic
approach for obtaining the optimal gains of PID controller
in AVR system. In this paper, the problem of discretisation
in the representation of the decision variables in the binary-
coded GA has been alleviated by employing floating point
numbers to represent the PID parameters. Blend crossover
and uniform mutation which can directly deal with the real
variables have been applied. The proposed GA approach
has resulted in better dynamic performance of the AVR.
Further the proposed GA occupies less computer space and
takes less time for convergence compared with the
conventional binary-coded GA. The optimal gains
produced by the GA-based approach were used to develop
the Sugeno fuzzy system. The performance of the
Figure 8 Step response of change in terminal voltage for algorithm in obtaining the optimal values of PID controller
off-nominal values parameters under various operating conditions has been

648 IET Gener. Transm. Distrib., 2009, Vol. 3, Iss. 7, pp. 641– 649
& The Institution of Engineering and Technology 2009 doi: 10.1049/iet-gtd.2008.0287

Authorized licensed use limited to: Arulmigu Kalasalingam College of Engineering. Downloaded on August 20, 2009 at 05:01 from IEEE Xplore. Restrictions apply.
 www.ietdl.org

analysed through computer simulation. The simulation result [9] YU G.R. , HWANG R.C.: ‘Optimal PID speed control of
shows that the Sugeno fuzzy system can produce the optimal brushless DC motors using LQR approach’. IEEE Int. Conf.
PID values accurately in a fraction of the second and this Syst., Man Cybern., 2004, pp. 473– 478
solves the problems associated with evolutionary
computation techniques in applying to on-line application. [10] VISIOLI A.: ‘Tuning of PID controllers with fuzzy logic’,
Proc. Inst. Electron. Eng. Control Theory Appl., 2001, 148,
(1), pp. 1 – 8
9 References
[11] KROHLING R.A., REY J.P.: ‘Design of optimal disturbance
[1] ASTROM K.J. , HANG C.C., PERSSON P. , HO W.K.: ‘Towards rejection PID controllers using genetic algorithm’, IEEE
intelligent PID control’, Automatica, 1992, 28, (1), pp. 1 – 9 Trans. Evol. Comput., 2001, 5, (1), pp. 78– 82

[2] ASTROM K.J., HAGGLUND T.: ‘The future of PID control’, [12] VALARMATHI K., DEVARAJ D., RADHAKRISHNAN T.K.: ‘Enhanced
Control Eng. Pract., 2001, 9, (11), pp. 1163– 1175 genetic algorithm based proportional integral controller
tuning for pH process’, Instrumen. Sci. Technol., 2007, 35,
[3] LI Y., ANG K.H., CHONG G.C.Y.: ‘PID control system analysis (6), pp. 619– 635
and design, problems, remedies and future directions’,
IEEE Control Syst. Mag., 2006, 26, (1), pp. 32– 41 [13] KIM D.H., ABRAHAM A., CHO J.H.: ‘A hybrid genetic algorithm
and bacterial foraging approach for global optimization’, Int.
[4] LI Y., ANG K.H., CHONG G.C.Y.: ‘Patents, software and hardware J. Inf. Sci., 2007, 177, (18), pp. 3918 – 3937
for PID control’, IEEE Control Syst. Mag., 2006, 26, (1),
pp. 42–54 [14] GAING Z.L.: ‘A particle swarm optimization approach for
optimum design of PID controller in AVR system’, IEEE
[5] O’DWYER A.: ‘Handbook of PI and PID controller tuning Trans. Energy Convers., 2004, 19, (2), pp. 384– 391
rules’ (Imperial College Press, London, 2003)
[15] ROSS T.: ‘Fuzzy logic with engineering application’ (Tata
[6] SCHEI T.S.: ‘Automatic tuning of PID controllers based on McGraw Hill Inc., 1995)
transfer function estimation’, Automatica, 1994, 30, (12),
pp. 1983 – 1989 [16] GOLDBERG D.: ‘Genetic algorithms in search, optimization
and machine learning’ (Addison-Wesley, 1989)
[7] SHINSKEY F.G. : ‘Process control system: application,
design and tuning’ (McGraw-Hill, 1996, 4th edn.) [17] DEVARAJ D., YEGNANARAYANA B.: ‘Genetic algorithm-
based optimal power flow for security enhancement’,
[8] MORARI M. , ZUFIRIOU E.: ‘Robust process control’ IEE Proc. Gener., Transm. Distrib., 2005, 152, (6),
(Prentice-Hall Inc., Englewood Cliffs, NJ, 1987) pp. 899– 905

IET Gener. Transm. Distrib., 2009, Vol. 3, Iss. 7, pp. 641 – 649 649
doi: 10.1049/iet-gtd.2008.0287 & The Institution of Engineering and Technology 2009

Authorized licensed
View publication stats use limited to: Arulmigu Kalasalingam College of Engineering. Downloaded on August 20, 2009 at 05:01 from IEEE Xplore. Restrictions apply.