Electrical Power and Energy Systems 77 (2016) 418–429

Contents lists available at ScienceDirect

Electrical Power and Energy Systems

journal homepage: www.elsevier.com/locate/ijepes

PID controller for automatic voltage regulator using teaching–learning

based optimization technique
Shamik Chatterjee, V. Mukherjee ⇑
Department of Electrical Engineering, Indian School of Mines, Dhanbad, Jharkhand, India

a r t i c l e i n f o a b s t r a c t

Article history: The present work presents teaching–learning based optimization (TLBO) algorithm as an optimization
Received 17 August 2015 technique in the area of tuning of the classical controller installed in automatic voltage regulator
Received in revised form 24 September 2015 (AVR). The proposed TLBO algorithm is applied with an aim to find out the optimum value of proportional
Accepted 10 November 2015
integral derivative (PID) controller gains with first order low pass filter installed in the AVR. The voltage
response of the AVR system, as obtained by using the proposed TLBO based PID controller with first order
low pass filter, is compared to those offered by the other algorithms reported in the recent state-of-the-
Keywords:
art literatures. The advantage of using this control strategy may be noted by providing good dynamic
Automatic voltage regulator
Figure of demerit
responses over a wide range of system parametric variations. For on-line, off-nominal operating condi-
Sugeno fuzzy logic (SFL) tions, fast acting Sugeno fuzzy logic technique is applied to obtain the on-line dynamic responses of
Teaching–learning based algorithm the studied model. Furthermore, robustness analysis is also carried out to check the performance of
the designed TLBO based PID controller. An analysis, based on voltage response profile, has been inves-
tigated with the variations of the model parameters. The simulation results show that the proposed TLBO
based PID controller is a significant optimization tool in the subject area of the AVR system. The essence
of the present work signifies that the proposed TLBO technique maybe, successfully, applied for the AVR
of power system.
Ó 2015 Elsevier Ltd. All rights reserved.

Introduction in terminal voltage changes the reactive power flow with a big
margin. The AVR, which is used to maintain the terminal voltage
An automatic voltage regulator (AVR) is a device that is of a synchronous generator at a specified level, is implemented
designed to automatically control, adjust or maintain a constant in power system to overcome these control problems. It also con-
voltage level of a synchronous generator. The main function of trols the reactive power flow and ensures proper sharing of the
the AVR is to maintain the voltage of an alternator at a definite reactive power amongst all the generators connected in parallel.
level. Thus, the security of the power system would be seriously With the variation of the exciter voltage of the alternator, the
affected by the stability of the AVR system. AVR maintains the consistency of the terminal voltage [1]. Stable
and fast response of the regulator is difficult to achieve due to
General the high inductance of the alternator field windings and load vari-
ation. Hence, improvement of the AVR performance is very impor-
In power system, one of the main control problem is to provide tant. Insulation breakdown may occur in different parts of the
constancy and stability of the nominal voltage level in an electrical power system due to high voltage which may damage the equip-
power network having all connected equipments, designed for a ment. Thus, proper controlling mechanism is required for the
certain voltage level (called as rated or nameplate voltage). There AVR system to perform properly.
may be decrease in the performance of these equipments and drop
in their expectancy, if the nominal voltage level deviates from the Literature survey
rated one. Another vital reason for this control is that the real line
loss depends on the real and reactive power flow. In fact, variation In the AVR system, so far, a number of different control strate-
gies such as adaptive control, robust control and optimal control
⇑ Corresponding author. Tel.: +91 0326 2235644; fax: +91 0326 2296563. have been proposed by the researchers to analyze the system with
E-mail addresses: [email protected] (S. Chatterjee), vivek_agamani@ an aim to gain better dynamic response. Self-tuning adaptive con-
yahoo.com (V. Mukherjee). trol technique is simple to apply than the other modern control

http://dx.doi.org/10.1016/j.ijepes.2015.11.010
0142-0615/Ó 2015 Elsevier Ltd. All rights reserved.
 S. Chatterjee, V. Mukherjee / Electrical Power and Energy Systems 77 (2016) 418–429 419

Nomenclature

ESS steady state error, p.u. De incremental change in error in terminal voltage, p.u.
KA gain constant of the amplifier Dei incremental change in error in terminal voltage at time
KD derivative gain of the controller i, p.u.
KE gain constant of the exciter Dei
1 incremental change in error in terminal voltage at time
KG gain constant of the generator ði
1Þ, p.u.
KI integral gain of the controller De_ derivative of incremental change in error in terminal
KP proportional gain of the controller voltage, p.u.
KS gain constant of the sensor DV ref ðsÞ incremental change in reference voltage, p.u.
MP maximum overshoot, p.u. DV s ðsÞ feedback voltage, p.u.
N filter co-efficient of the PID controller’s filter DV t ðsÞ incremental change in terminal voltage, p.u.
t sample time, s sA time constant of amplifier, s
T time constant of the PID controller’s filter, s sE time constant of exciter, s
TR rise time, s sG time constant of generator, s
TS settling time, s sS time constant of sensor, s
b parameter constant of figure of demerit

techniques and, with change in process parameters, it also makes a simplified version of PSO, called many optimizing liaisons (MOL),
the process less sensitive on being under control. This type of con- to have optimum PID controller gains for the AVR system and fur-
trol mechanism is implemented in the AVR system. The presence of ther a comparative analysis of their obtained results with the ABC,
more mathematical computation may make the conventional self- PSO and DEA based results has been reported in [11]. Mohanty
tuning control techniques inadequate in some operating conditions et al. have used local unimodal sampling (LUS) optimization algo-
because of intricacy of the power system such as variable operating rithm to tune the gains of the PID controller for the AVR system
points and non-linear load characteristics. Since 2000, the [12]. ABC-based obtained results were also compared to PSO and
researchers prefer the usage of optimization techniques and artifi- DEA based results in [12].
cial intelligence based self-tuning control strategy.
Proportional integral derivative (PID) controller is the most Motivation for the present work
preferable controller amongst all the proposed ones. Robust perfor-
mance over a wide-range of operating conditions and simplicity of Literature survey reveals that most of the previous researchers
structure design makes the PID controller different from other have used evolutionary optimization algorithms to tune the
types of controllers [2]. PID controller is designed with three con- parameters of the PID controller for controlling the AVR system.
trol parameters i.e. proportional, integral and derivative gains. For The simulation results of the existing research works have some
improving the voltage response of the AVR system, different types deficiencies of their own which may be overcome for having the
of heuristic optimization algorithms have been evolved. These most appropriate and desirable response. The responses, presented
methods have become popular across the world and acceptable in the earlier works, require lesser value of rise time (T R ), settling
to the researchers’ pool. PID controller gains are being tuned by time (T S ), overshoot (M P ) and steady state error (ESS ). Thus, the val-
many soft computing techniques for the AVR system. In 2004, ues of T R , T S , MP and ESS may even become optimum than those
Gaing [3] proposed a particle swarm optimization (PSO) technique offered by the reported algorithms viz. PSO [10,11], MOL [11], GA
based self-tuning PID controller for the AVR system and presented [3], LUS [12], ABC [10] and DEA [10] for the PID controlled AVR sys-
a comparison between PSO based method and genetic algorithm tem. PID controller, tuned by modern optimization technique, may
(GA) based method. In order to improve the performance of the overcome this drawback of the AVR system by reducing the values
self-tuning PID controller for the AVR system, Kim and Cho [4] of T R , T S , M P and ESS which may lead to decrease in overall objective
have developed a hybrid method containing GA and bacterial for- function value and, in turn, optimizing the system performance.
aging optimization technique. Mukherjee and Ghoshal have pre- Thus, it may be expected that the transient response of the AVR
sented Sugeno fuzzy logic (SFL) based self-tuning algorithm for system may be closer to the optimal one.
PID controller employing crazy PSO (CRPSO) [5]. In 2008, Kashki Some studied optimization techniques, as per the literature sur-
et al., based on reinforcement learning automata (RLA), have pro- vey, have a number of limitations and problems of their own. The
posed continuous action RLA optimization method in order to opti- notified limitations in GA are (a) requirement of more overall exe-
mize the parameters of the PID controller for the AVR system of cution time, (b) exposition of premature convergence i.e. ambush
synchronous generator and further they have also compared their in local minima and (c) involvement of lots of crossover and muta-
results with PSO and GA based controller [6]. In the year of 2009, tion operations in each iteration cycle. PSO uses the concept of sim-
Zhu et al. proposed a chaotic ant swarm algorithm to optimize ulation of bird flocking in multi-dimensional search spaces. PSO
the gains of the PID controller for the AVR system [7]. In the same undergoes various pragmatic studies which show that the particle
year, Coelho proposed chaotic optimization approach, based on may still diverge i.e. may go to infinity (a phenomenon known as
lozi map, for the tuning of the PID controller gains of the AVR sys- ‘‘explosion” of the swarm) even on currently defining the maxi-
tem [8]. Later on, Chatterjee et al. [9] have carried out a comparison mum velocity and acceleration constants [13]. ABC has also some
between the optimization response of CRPSO based optimization inherent deficiencies such as (a) improvement of performance
technique and velocity relaxed PSO based optimization technique requires new fitness tests on the new algorithm parameters, (b)
for the AVR system. Gozde and Taplamacioglu [10] have suggested chance of losing pertinent information on the behavior of the
artificial bee colony (ABC) algorithm to obtain optimum control for function to be optimized, (c) large number of objective function
the AVR system and have carried out a comparison between their calculations, (d) on usage of sequential processing, it slows down,
obtained results with the PSO and differential evolution algorithm (e) increment of computational cost occurs due to slowdown
(DEA) based results for the AVR system. Panda et al. have proposed which may further lead to the requirement of more iterations
420 S. Chatterjee, V. Mukherjee / Electrical Power and Energy Systems 77 (2016) 418–429

and memory capacity, (f) accuracy of solution is obtained in a slow Section ‘‘TLBO: an overview”. In Section ‘‘SFL for on-line tuning of
manner and (g) solutions achieved by deterministic methods have PID controller”, SFL technique is focused for on-line tuning of the
higher accuracy, provided, it does not ambush in local minima. PID controller. In Section ‘‘Results and discussions”, simulation
Teaching–learning based optimization (TLBO) is a population results are presented and its significances are discussed. Finally,
based optimization algorithm which is a new metaheuristic tech- the concluding remarks are drawn in Section ‘‘Conclusion”.
nique [14]. It is based on the concept of classroom environment
where knowledge is passed on from one individual to the other. Modeling of an AVR system
As it is a population based method, like other population based
methods, it also uses a population of solutions to get the global One of the most important factors to improve power system
solution. In TLBO, group of learners establish the population. There stability and quality of electrical power, is the excitation control
may be number of different design variables in any optimization of the synchronous generator. To retain the magnitude of the ter-
technique. In TLBO, these design variables are expressed as differ- minal voltage of a synchronous generator within a specified limit,
ent subjects suggested to the learners and the learners’ result is the AVR is required. The AVR, comprises of four main components
corresponding to the ‘‘fitness function value”, like in other popula- (namely, amplifier, exciter, generator and sensor), plays an impor-
tion based optimization algorithms. In TLBO, the best solution is tant role in the power system [15]. Voltage sensor’s role is to sense
analogous to the teacher, as the teacher is treated as the most the terminal voltage of the alternator continuously. This signal,
learned person in the colony. TLBO algorithm is divided in two after being rectified or smoothened, is compared to a reference sig-
activities which are the two phases of the algorithm, the learner nal in the comparator. The error voltage obtained from the com-
first get the knowledge from the teacher (i.e. teacher phase) and parator is used after being amplified to control the field winding
then from classmates, which is through the interaction between of the alternator by means of the exciter. The transfer functions
learners in a class (i.e. learner phase). An optimum solution is of the described four components of the AVR system are assumed
searched by TLBO through each learner as he/she tries to get the to be linear for the mathematical modeling of the system in the
experience of the teacher, thereby, achieving the optimum results present work.
rather than having learners undergoing genetic operations like In Fig. 1, the transfer function block diagram of the AVR system
selection, crossover and mutation. It is a powerful approach which is shown. An amplifier’s transfer function is modeled by a gain and
has been tested on different types of benchmark test functions a time constant, as given in (1) [5]
such as separable, non-separable, unimodal and multimodal in
KA
providing quality optimum results. It requires less time to con- GAmplifier ðsÞ ¼ ð1Þ
verge fully. TLBO has become a very important optimization tech- 1 þ ssA
nique and has been implemented, successfully, in many real-world where K A and sA are the gain and the time constant of the amplifier
problems because of its simple concept and high efficiency. Thus, system. The value of K A is in the range of 10–40 while the amplifier
an attempt may be made to deploy TLBO as an optimizing tool time constant ðsA Þ ranges from 0.02 s to 0.1 s. Like amplifier, the
for obtaining the optimal values of the controller gains of the stud- transfer function of a modern exciter may be represented by a gain
ied AVR system of the present work. and a time constant, as presented in (2) [5]
A SFL based controller is used in various applications. Over a
KE
wide range of operating conditions, its parameters are adjusted GExciter ðsÞ ¼ ð2Þ
on-line considering the environment in which it works. For varying 1 þ ssE
on-line off nominal conditions, fast acting intelligent SFL may be where K E and sE are the gain and the time constant of the exciter
applied to yield the incremental change in terminal voltage system. The value of K E is in the range of 1–10 and sE ranges from
response during real-time. 0.4 s to 1.0 s. Similarly, Eq. (3) represents the transfer function relat-
ing the generator terminal voltage to its field voltage [5]
Contribution of the present work
KG
GGenerator ðsÞ ¼ ð3Þ
1 þ ssG
The major contributions of this work are to:
where K G and sG are the gain and the time constant of the generator
(a) optimize the TLBO tuned PID controller gains and linear first model. The value of K G is in the range of 0.7–1.0 and sG ranges from
order low-pass filter parameters for the AVR system, 1.0 s to 2.0 s. The transfer function of the sensor is modeled by a
(b) compare the responses offered by the TLBO with those gain and a time constant, as presented in (4) [5]
offered by some other optimization techniques reported in
KS
the recent state-of-the-art literatures, GSensor ðsÞ ¼ ð4Þ
1 þ ssS
(c) study the appropriateness of the SFL based tuning of the PID
controller gains while changing in operating conditions of where K S and sS are the gain and the time constant of the sensor
the system in real-time environment, and model. The value of K S is in the range of 0.9–1.1 and sS ranges from
(d) study the robustness of the TLBO based PID controlled AVR 0.001 s to 0.06 s.
system under parametric variation of the studied model. In this work, the chosen parameters for the AVR system are as
follows: K A = 10.0, sA = 0.1 s, K E = 1.0, sE = 0.4 s, K S = 1.0,
Layout of paper sS = 0.01 s [3,10,11,16,17]. The values of K G and sG are load
dependent.
The rest part of the paper is arranged as follows. The modeling Taking K G = 1.0 and sG = 1.0 s along with the above mentioned
of AVR system is illustrated in Section ‘‘Modeling of an AVR values of the model, the closed loop transfer function of the AVR
system”. The structure of the PID controller is described in Sectio system (GAVR ) becomes as in (5).
n ‘‘Structure of PID controller”. In Section ‘‘Analysis of AVR system
DV t ðsÞ 0:1s þ 10
with PID controller”, the analysis of the AVR system with PID GAVR ðsÞ ¼ ¼
DV ref ðsÞ 0:0004s4 þ 0:045s3 þ 0:555s2 þ 1:51s þ 11
controller is investigated. In Section ‘‘Mathematical problem
formulation”, the mathematical problem of the present work is 250ðs þ 100Þ
¼ ð5Þ
formulated. An overview of the TLBO algorithm is presented in ðs þ 98:82Þðs þ 12:63Þðs2 þ 1:057s þ 22:04Þ
 S. Chatterjee, V. Mukherjee / Electrical Power and Energy Systems 77 (2016) 418–429 421

ΔVref (s ) ΔVt (s )
ΔVe (s ) KA KE
∑ KG
1 + sτ A 1 + sτ E 1 + sτ G
+
ΔVs (s ) _
Amplifier Exciter Generator

KS
1 + sτ S

Sensor

Fig. 1. Transfer function based model of the AVR system.

error and a less responsive or less sensitive controller. The effect

of an integral controller is to eliminate ESS but the transient
response gets worse by it. The value of M P will be high, if integral
gain is made higher and on putting it low, it may result in a slug-
gish system response. A derivative controller may increase the sta-
bility of the system, reduce the value of M P and, hence, there will
be an improvement in transient response [18,19]. The process
may become unstable on making the derivative gain, sufficiently,
large. For industrial control process, the conventional fixed gain
PID controller is extensively used. The three parameters of this
controller viz. proportional gain ðK P Þ, integral gain ðK I Þ and deriva-
tive gain ðK D Þ are required to be designed appropriately. Based on
the experience of the designer and plant behavior, the tuning of
these parameters is done by trial and error method. In Fig. 3, the
Fig. 2. Voltage response profile of the AVR system without controller. block diagram of PID controller for a closed loop system is shown.
In Laplace mode, the transfer function of PID controller may be
expressed by (7).
From (5), it may be noted that there is one zero and two real poles
KI
at Z ¼
100 and at S1 ¼
98:82 and S2 ¼
12:63, respectively, and TF ¼ KP þ þ sK D ð7Þ
s
the two complex poles are at S3;4 ¼
0:53  j4:66. Approximation of
GAVR ðsÞ may be done by cancelling the zero at
100 with the pole at For this system, a first-order low pass filter has been installed in
e AVR ðsÞ, as represented in (6). the derivative path of the PID controller for obtaining smooth and

98.82 to obtain G
better response profile of the terminal voltage. The transfer func-
250 tion of the studied filter (GFilter ðsÞ) may be presented in (8)
e AVR ðsÞ ¼
G ð6Þ
ðs þ 12:63Þðs2 þ 1:057s þ 22:04Þ s
GFilter ðsÞ ¼  1  ð8Þ
The voltage response of the AVR system without any controller s N T þ1
is shown in Fig. 2. From this figure, it may be observed that it exhi- where T is the time constant and N is the filter co-efficient. The
bits high oscillations with M P = 0.0150 p.u., T R = 0.2616 s and value of T lies in the range of 0.01–0.1 s and the value of N ranges
T S = 7.0191 s. In the steady state condition, the system terminal from 0.1 to 2.0. Thus, the overall transfer function of the employed
voltage ðDV t Þ deviates from the nominal value of 0.01 p.u. In power PID controller of the present work may be stated by (9).
system (where the operating voltage is in the order of hundreds of  
kV), this type of response is completely unwanted and is not per- KI 1
GPID ðsÞ ¼ K P þ þ sK D ð9Þ
missible. For this purpose, a controller is needed to be installed s sð1=NÞT þ 1
in the AVR system. It is known that the PID controller is playing
a vital role in the industries due its simplicity and robustness. Thus,
Analysis of AVR system with PID controller
in the studied AVR system, the PID controller may be implemented
for controlling the deviation in terminal voltage (if any). In the next
A PID controller, due to its simple structure, is implemented in
section, the structure of the studied PID controller is discussed.
the AVR system to improve the output response of the system. PID
controller is employed in the forward path which is capable of pro-
Structure of PID controller cessing the voltage difference ðDV e ðsÞ ¼ DV ref ðsÞ
DV s ðsÞÞ and pro-
duces a manipulated actuating signal.
The PID controller is the most acceptable feedback controller The PID controller is a combination of three control actions
used in the industries. For over 50 years, it is being, successfully, related to the error signal in proportional, integral and differential
used by the industry houses. It is easily understandable and as well manners and its transfer function is given in (7). The block diagram
as robust controller that can offer excellent control performance of the AVR system with PID controller is shown in Fig. 4. The gen-
despite the varied dynamic characteristics of the process plant. eral transfer function of the AVR system, controlled by a PID con-
PID controller has time controller gains i.e. proportional, integral troller GAVR PID ðsÞ, is given by (10).
and derivative gains. A proportional controller can reduce the value GPID ðsÞGAmplifier ðsÞGExciter ðsÞGGenerator ðsÞ
of T R but cannot eliminate ESS . The system can become unstable on GAVR PID ðsÞ ¼
1 þ GPID ðsÞGAmplifier ðsÞGExciter ðsÞGGenerator ðsÞGSensor ðsÞ
making the proportional gain too high, whereas a small value of
this gain may result in a small output response to a large input ð10Þ
422 S. Chatterjee, V. Mukherjee / Electrical Power and Energy Systems 77 (2016) 418–429

Fig. 3. Block diagram of the studied PID controller.

ΔVref (s ) ΔVe (s ) ΔVt (s )

KI ⎛ 1 ⎞ KA KE KG
∑ KP + + sK D ⎜⎜ ⎟⎟
s ⎝ s (1 / N )T + 1 ⎠ 1 + sτ A 1 + sτ E 1 + sτ G
+
ΔVs (s ) _ PID controller Generator
Amplifier Exciter

KS
1 + sτ S

Sensor

Fig. 4. Transfer function block diagram of the AVR system with PID controller.

9
Mathematical problem formulation K min 6 K P < K max >
>
P P >
>
max >
KI 6 KI 6 KP >
min
>
=
Mathematical problem formulation is a very important task for
K min 6 K 6 K max ð12Þ
obtaining the solution of an optimization problem. The main two D D D >
>
>
aspects of an optimization task are the design of the objective Nmin 6 N 6 Nmax > >
>
>
;
function and proper definition of the constraints of the problem. T min 6 T 6 T max
These two aspects are described in the next two sub-sections, in
order. where the superscripts min and max speak for the minimum and
the maximum values of the respective variables. For this work,
Design of objective function the value of K P ranges from 0.3 to 0.6, K I ranges from 0.1 to 0.5,
K D ranges from 0.01 to 0.4, N ranges from 0.1 to 2 and T ranges from
The performance of the AVR system is analyzed by using a time 0.01 s to 0.1 s.
domain objective function. This time domain performance analysis
is related to the involvement of some performance parameters. The TLBO: an overview
better performance is achieved when values of ESS , MP , T S and T R of
the step response are minimum corresponding to the step input Teaching–learning is a cardinal process where every solitary
perturbations. A time domain objective function, termed as figure tries to learn something from other solitary to improve themselves.
of demerit (FOD), is designed as in (11) [20] which is required to TLBO has been proposed by Rao et al. in [14,21]. The traditional
be minimized. teaching–learning phenomenon of a classroom is being simulated
in this algorithm. The two fundamental phases of learning are
FOD ¼ ð1
e
b ÞðM P þ ESS Þ þ e
b ðT S
T R Þ ð11Þ
being simulated in this algorithm viz. (a) teacher phase (learning
In (11), the value of b is set to be larger than 0.7 to reduce M P from teacher) and (b) learner phase (learning through the
and ESS . For reducing the value of T R and T S , it is recommended interaction between learners). Basically, TLBO is a population
to use the value of b smaller than 0.7. In the present work, the based algorithm. A class of students (i.e. learner) is being
value of b is set as 1.0. The values of M P , Ess, T S and T R are obtained considered as the population and the different subjects offered to
from the transient response. Hence, the main objective of the pre- the learners are analogous with the different design variables of
sent optimization task would be to get the minimum value of FOD, the optimization problem. The outputs of the learner are cognate
as defined in (11). to the fitness function value of the optimization problem. It may
be considered that the teacher may be the best solution in entire
Constraints of the present work population.

The constraints of the used optimization technique in the pre- Teacher phase
sent research work are the gains of the controllers which must
be bounded within some pre-specified limits. These limits may Through the teacher, at this juncture, the algorithm simulates
be mounted by (12) the learning of the students (i.e. learners). A teacher imparts
 S. Chatterjee, V. Mukherjee / Electrical Power and Energy Systems 77 (2016) 418–429 423

knowledge among the learners and makes an effort to increase the X 00j;P;i ¼ X 0j;P;i þ r i ðX 0j;P;i
X 0j;Q ;i Þ; If X 0total
P;i > X 0total
Q;i ð16Þ
mean result of the class. Let ‘m’ be the number of subjects (i.e. X 00j;P;i ¼ X 0j;P;i þ r i ðX 0j;Q ;i
X 0j;P;i Þ; If X 0total
Q ;i > X 0total
P;i ð17Þ
design variables) offered to ‘n’ number of learners (i.e. population
size, k = 1, 2, . . . , n). M ij is the mean result of the learns in a particu- Eqs. (16) and (17) are for the maximization problems, the
lar subject ‘j’ (j = 1, 2, . . . , m) at any sequential teaching–learning reverse will be true for minimization problems. If X 00j;P;i gives better
cycle, i. A teacher is being considered as the best learner in the function value, accept it. In Fig. 5, the flowchart of the TLBO algo-
entire population, as a teacher is the most experienced and knowl- rithm is shown.
edgeable person of a subject. On considering all the subjects,
X total
Kbest;i is assumed as the result of the best learner who is iden-
SFL for on-line tuning of PID controller
tified as a teacher for that cycle. To increase the knowledge level of
the whole class, the teacher will have to put the maximum effort. Fuzzy logic controller (FLC) implementation requires two inputs
But depending upon the quality of the learners present in the class which are incremental change in error in terminal voltage ðDeÞ and
and the quality of the teaching delivered by a teacher, the learners _
derivative of incremental change in error in terminal voltage ðDeÞ
will gain knowledge. The difference between the result of the tea-
to provide a control signal ðuÞ [22–24]. In Fig. 6, the FLC based
cher and the mean result of the learner’s in each subject is
PID controller is shown.
expressed as in (13) [14,21]
The difference between the change in reference voltage and ter-
Difference Meanj;i ¼ r i ðX j;kbest;i
T F M j;i Þ ð13Þ minal voltage is called incremental change in error in terminal
voltage while the rate of change of incremental change in error
where X j;kbest;i is the output of the teacher (i.e. the best learner) in the in terminal voltage is called derivative of incremental change in
subject j, ri is the random number in the range [0, 1] and T F is the error in terminal voltage. The expression for the error and deriva-
teaching factor which decides the value of mean to be changed. tive of error is presented in (18) and (19), respectively.
Ranging between 1 and 2, the value of T F is decided randomly with
equal probability by using (14) [14,21] De ¼ ðDV ref
DV s Þ ð18Þ
Dei
Dei
1
T F ¼ round½1 þ randð0; 1Þ  ð2
1Þ ð14Þ De_ ¼ ð19Þ
t
where rand is the random number in the range [0, 1] and the oper-
where Dei is the incremental change in error in terminal voltage at
ator round½ offers the nearest integer value. The value of T F is ran-
time i, Dei
1 is the incremental change in error in terminal voltage at
domly decided by the algorithm using (14). It may be noted here
time ði
1Þ and t is the sample time (in s).
that its value is not given as an input to the algorithm and it is
The error and derivative of error are divided into seven fuzzy
not a parameter of the TLBO algorithm.
classes as large negative (LN), medium negative (MN), small nega-
The extant solution is refurbished in the teacher phase, relying
tive (SN), zero (Z), small positive (SP), medium positive (MP) and
upon the value of Difference Meanj;i , as per the expression pre-
large positive (LP).
sented in (15) [14,21] The FLC consists of three membership functions with two
inputs and one output as shown in Fig. 7. The individual member-
X 0j;k;i ¼ X j;k;i þ Difference Meanj;i ð15Þ
ship functions include five triangular memberships and two trape-
zoidal memberships. In Table 1, the control rules are depicted for
where X 0j;k;i is the updated value of X j;k;i . If X 0j;k;i gives a better function the case of two inputs and one output. With two input member-
value, then it will be accepted. All the accepted function values ship functions, every cell shows the output membership function
become the input to the learner phase. of a control rule (refer in Table 1). The control rules are executed
The performance of the TLBO algorithm is affected by the values from the statement: if input 1 and input 2 then output 1. In
of ri and T F . The values of ri and T F are randomly generated in the Fig. 8, the surface plot profile of the control signal with the varia-
algorithm and these parameters are not supplied as input to the tion of incremental change in error and derivative of incremental
algorithm (unlike cognitive and social parameters and inertia change in error signal is presented.
weight in PSO, colony size and limit in ABC, mutation probabilities
and crossover in GA, etc.). Hence, in the TLBO, tuning of r i and T F is Comparison of Sugeno FLC and Mamdani FLC systems
not required. The common control parameters like number of
generations and population size are only required to tune for its FLCs have been implemented to the real world power system
operation. For the operation of all population based optimization applications as an intelligent control technique and due to that
algorithms, there is a requirement of the common control param- the intelligence in power system control is being deliberately
eters. Hence, TLBO may be treated as algorithm-specific increasing. The Sugeno system allows itself to the utilization of
parameter-less algorithm. adaptive techniques for constructing fuzzy models because of its
more compactness and, computationally, more efficient features
Learner phase than Mamdani system. The membership function customization
can be done by employing these adaptive techniques which will
Simulation of the learning of the students (i.e. learners) through lead to fuzzy system modeling for the given data. Sugeno method
interaction among themselves is done in this juncture of the has the following advantages.
algorithm. Knowledge is gained by the students by dis-
cussing and interacting with other students. If the other learner (a) With linear techniques, it works fine.
has more knowledge than him or her, a learner will acquire new (b) It works well with adaptive and optimization technique.
information. The learning phenomenon, at this juncture, is (c) It is, computationally, efficient.
illustrated below. (d) For the mathematical formulation and calculation, it per-
Arbitrarily select two learners, (P and Q ), such that forms nicely.
X 0total
P;i – X 0total
Q ;i , where X 0total
P;i and X 0total
Q ;i are the updated val-
ues of X total
P;i and X total
Q ;i , respectively, at the end of the teacher Mamdani FLC method has also some advantages which are as
phase as presented in (16) and (17), respectively. follows.
424 S. Chatterjee, V. Mukherjee / Electrical Power and Energy Systems 77 (2016) 418–429

Initialize number of students (population), termination criterion

Calculate the mean of each design variables

Identify the best solution (teacher)

Modify solution based on best solution

X 'j , k ,i = X j , k ,i + Difference _ Mean j ,i

No Is new solution better

Reject Accept
than existing?

Select any two solutions randomly P and Q

Is P better than Q ?
Yes No

X 'j' , P,i = X 'j , P,i + ri ( X 'j , P,i − X 'j ,Q,i ) X 'j' , P,i = X 'j , P,i + ri ( X 'j ,Q,i − X 'j , P,i )

No Is new solution better than Yes

Reject Accept
existing?

Is termination criteria No
satisfied?

Yes
Final value of solution

Fig. 5. Flowchart of the TLBO algorithm.

Fig. 6. Fuzzy logic based PID controller.

(a) It is widely accepted. For static system (with slow changing dynamics), Mamdani FLC
(b) For user input, it works fast. method will work better and for dynamic system (with fast chang-
(c) It is spontaneous. ing dynamics) Takagi–Sugeno method is a better option.
 S. Chatterjee, V. Mukherjee / Electrical Power and Energy Systems 77 (2016) 418–429 425

_ and (c) Output ðuÞ.

Fig. 7. Membership functions of TLBO-SFL based PID controller (a) Input 1ðDeÞ, (b) Input 2 (De)

Results and discussions

Table 1
Control rules of TLBO-SFL based PID controller. The AVR system is controlled by the PID controller to achieve the
desired response for which the gains and linear low-pass filter
Input 1 LN MN SN Z SP MP LP
parameters of the PID controller are tuned in this paper by an evo-
Input 2
lutionary algorithm viz. TLBO. The observation of this work is car-
LN LP LP LP MP MP SP Z ried out by analyzing the voltage response ðDV t Þ for different
MN LP MP MP MP SP Z SN
cases. The simulation has been done using MATLAB/SIMULINK (ver-
SN LP MP SP SP Z SN MN
Z MP MP SP Z SN MN MN sion: 7.10) in 2.77 GHz, Intel CoreTM i7 computer. The maximum
SP MP SP Z SN SN MN LN population size is set as 50 and maximum number of iterations
MP SP Z SN MN MN MN LN allowed is set as 100. Results of interest are bold faced in the
LP Z SN MN MN LN LN LN respective tables. The major observations of the present work are
illustrated below:

Fig. 8. Surface plot of control signal with the variation of error and derivative of error signal.
426 S. Chatterjee, V. Mukherjee / Electrical Power and Energy Systems 77 (2016) 418–429

Table 2
TLBO optimized PID controller gains and transient response parameters under varying K G and sG .
KG sG KP KI KD N T (s) T S (s) T R (s) M P (p.u.) FOD
0.7 1.0 0.5272 0.4000 0.1770 1.9074 0.0108 1.3675 0.6181 0.0100 0.2821
1.2 0.5376 0.4001 0.1780 1.9030 0.0109 1.2334 0.6953 0.0102 0.2045
1.4 0.5311 0.4002 0.1773 1.9034 0.0106 1.2964 0.8014 0.0104 0.1888
1.6 0.5469 0.3988 0.1753 1.9050 0.0108 2.3043 0.8536 0.0106 0.5406
1.8 0.5438 0.4003 0.1778 1.9029 0.0109 2.4221 0.9584 0.0108 0.5456
2.0 0.5409 0.3999 0.1700 1.9023 0.0107 2.4798 1.0564 0.0110 0.5192
0.8 1.0 0.5110 0.4015 0.1738 1.9049 0.0110 1.0105 0.4835 0.0100 0.2002
1.2 0.5200 0.4001 0.1773 1.9060 0.0119 1.0983 0.6060 0.0103 0.1877
1.4 0.5319 0.4115 0.1700 1.9063 0.0109 2.0694 0.6661 0.0104 0.5228
1.6 0.5375 0.4003 0.1762 1.9100 0.0108 2.3596 0.7419 0.0106 0.6020
1.8 0.5262 0.4100 0.1774 1.9023 0.0109 2.5106 0.8522 0.0108 0.6172
2.0 0.5314 0.4002 0.1754 1.9035 0.0111 2.5872 0.9218 0.0110 0.6200
0.9 1.0 0.5482 0.4020 0.1767 1.9005 0.0108 0.6916 0.4027 0.0100 0.2952
1.2 0.5382 0.4003 0.1765 1.9020 0.0105 0.8239 0.4887 1.0102 0.1298
1.4 0.5420 0.4001 0.1772 1.9019 0.0115 1.9162 0.5595 0.0104 0.5057
1.6 0.5387 0.4002 0.1757 1.9037 0.0110 2.3313 0.6362 0.0106 0.6304
1.8 0.5345 0.3999 0.1752 1.9064 0.0109 2.5068 0.7159 0.0108 0.6652
2.0 0.5346 0.4000 0.1737 1.9102 0.0108 2.6023 0.7879 0.0110 0.6747
1.0 1.0 0.5302 0.4001 0.1787 1.9104 0.0109 0.5603 0.3537 0.0100 0.0246
1.2 0.5382 0.4015 0.1777 1.9047 0.0119 0.6878 0.4249 0.0102 0.1032
1.4 0.5155 0.4011 0.1787 1.9084 0.0110 2.0200 0.5191 0.0104 0.5588
1.6 0.5370 0.4001 0.1786 1.9045 0.0109 2.2653 0.5606 0.0106 0.6339
1.8 0.5432 0.4005 0.1772 1.9002 0.0111 2.4389 0.6169 0.0107 0.6772
2.0 0.5353 0.4000 0.1780 1.9046 0.0108 2.5662 0.6961 0.0109 0.6950

Fig. 9. Voltage response profiles of the AVR system using TLBO based PID controller Fig. 11. Comparative TLBO-, GA-, PSO-, LUS-, ABC- and DEA-based time-domain
for different values of K G and sG . simulation response profile of the incremental change in terminal voltage (p.u.) for
the AVR system with K G ¼ 1:0 and sG ¼ 1:0 s.

response characteristic of the AVR system for the different four sets
of described values of K G and sG is shown in Fig. 9. Thus, TLBO
based PID controller has the capability to control the AVR system
with the variation in the generator’s parameters.

(ii) Convergence profile.

The minimum FOD value for the PID controlled AVR system is
recorded against number of iterations to get the convergence pro-
Fig. 10. TLBO based convergence profile of FOD value for K G ¼ 1:0 and sG ¼ 1:0 s. file of the proposed TLBO algorithm. In Fig. 10, the TLBO based con-
vergence profile of FOD for the studied system is shown
corresponding to the values of K G and sG mentioned alongside.
(i) Transient response characteristic under varying K G and sG .
From this figure, it is observed that the TLBO algorithm converges
Table 2 has been figured out to emphasize on the performance at a FOD value of 0.0246 in an around 19 iterations. Promising con-
analysis of the TLBO algorithm used for tuning the gains and the vergence mobility of the proposed TLBO algorithm may be noted
filter parameters of the PID controller while varying the value of from this figure.
K G and sG . In steps of 0.1, the value of K G is varied from 0.7 to
1.0 while the value of sG is varied from 1.0 s to 2.0 s in steps of (iii) Comparison of voltage responses of AVR system with
0.2 s. Hence, in this table, there are twenty-four different sets of different optimized PID controller.
input conditions of the AVR system. From this table, it may be
noted that TLBO based PID controller offers optimum value of Fig. 11 portrays a comparison of the terminal voltage response
M P , T S and T R . Thus, TLBO offers optimum value of FOD. The voltage of the studied PID controlled AVR system and voltage responses
 S. Chatterjee, V. Mukherjee / Electrical Power and Energy Systems 77 (2016) 418–429 427

Table 3
Comparative controller parameters and transient response for different controllers corresponding to K G ¼ 1:0 and sG ¼ 1:0 s.
Controller/algorithm KP KI KD M P (p.u.) T R (s) T S (s) FOD
PID/TLBO [Proposed] 0.5302 0.4001 0.1787 0.0100 0.3537 0.5603 0.0246
PID/GA/f [3] 0.8861 0.7984 0.3158 0.0109 0.2042 0.5963 0.1512
PID/PSO/f [3] 0.6568 0.5393 0.2458 0.0101 0.2743 0.9941 0.2712
PID/LUS/OF4 [12] 0.6190 0.4222 0.2058 0.0101 0.3112 1.0427 0.2755
PID/PSO/ITAE [11] 1.3541 0.9266 0.4378 0.0119 0.1485 0.8149 0.2527
PID/ABC/ITSE [10] 1.6524 0.4083 0.3654 0.0125 0.1516 0.9492 0.3015
PID/PSO/ITSE [10] 1.7774 0.3827 0.3184 0.0130 0.1565 1.0185 1.3255
PID/DEA/ITSE [10] 1.9499 0.4430 0.3427 0.0133 0.1479 1.2100 0.3993

is may be found that TLBO based PID controller values offers the
best FOD value as well as the performance index values (reported
in bold text) of the AVR system. Therefore, the TLBO based PID con-
troller has the capability to achieve the most accurate (minimum
oscillations), most stable (minimum M P ) and quiet faster (mini-
mum value of T R ) unit step response than the other indicated opti-
mization techniques.

(iv) SFL based response.

SFL based PID model is applied to achieve the on-line optimal

controller parameters for on-line and off-nominal input sets of
parameters. Consequently, to study the comparative performance
Fig. 12. Time-domain simulation response of the incremental change in terminal of the AVR system, the TLBO-SFL based PID controller gains are
voltage (p.u.) of the AVR system using SFL based PID controller. used. The voltage response profile of TLBO-SFL based PID controller
is depicted in Fig. 12. From this figure, it is observed that, the
yielded by the different optimization techniques viz. PSO [10,11], proposed TLBO-SFL method validates its importance in terms of
GA [3], ABC [10], DEA [10] and LUS [12] with 1% step perturbation on-line tuning of the PID controller gains. Thus, the proposed
in reference voltage for the ostensible input conditions as placed in TLBO-SFL method exhibits good voltage response with the
the figure. From this figure, it may be inferred that the proposed variation of operating conditions.
TLBO based PID controller yields a superior unit step response as
compared to the other optimization technique based PID controller (v) Robustness analysis.
for the same AVR system. In Table 3, a comparison of the numerical
results of the time-domain simulation responses of the incremen- The operating point of power system varies with the change in
tal change in terminal voltage (p.u.) (which includes the controller load as the parameters of the power system are not constant.
parameters, the time domain performance characteristics (such as Hence, robustness analysis is necessary for the designed AVR sys-
MP , T R and T S ) and the value of FOD) is reported. From this table, it tem. Robust controller is very much required for a system for

Fig. 13. Robustness analysis of the AVR system under varying (a) sA , (b) sE , (c) sG and (d) sS .
428 S. Chatterjee, V. Mukherjee / Electrical Power and Energy Systems 77 (2016) 418–429

Table 4 input condition for the AVR system. The dynamic response profile
Results of robustness analysis for the AVR system performance parameters tuned by of the terminal voltage, as obtained by using the TLBO based PID
TLBO algorithm.
controller, shows that TLBO is a significant optimization tool for
Model parameters Rate of change (%) M P (p.u.) T S (s) T R (s) the AVR system as compared to other techniques used in the recent
sA
50% 0.0100 1.9045 0.4348 state-of-the-art literatures. The FOD values also show that the

25% 0.0101 1.0311 0.3930 designed controller gains are towards the optimal ones. Addition-
+25% 0.0104 0.8610 0.3535 ally, the robustness analysis of the TLBO based PID controller is
+50% 0.0105 0.9555 0.3610
also carried out. It may be noted that it has given satisfactory
sE
50% 0.0100 1.4927 0.2534 results with the variation of parameter uncertainties. Moreover,

25% 0.0101 1.3202 0.3192
+25% 0.0103 1.1484 0.4054
the tuned values of the controller gains obtained with the nominal
+50% 0.0107 1.7599 0.4395 parameters need not be reset for a wide change in the system
sG
50% 0.0103 1.7094 0.1949
parameters. For on-line, off-nominal system parameters, the SFL

25% 0.0100 1.3348 0.2757 in conjugate with TLBO is applied in the present work to get on-
+25% 0.0102 0.7102 0.4419 line dynamic voltage profile of the investigated AVR model. The
+50% 0.0105 2.1134 0.5220 work of SFL is to extrapolate the nominal optimal PID controller
sS
50% 0.0100 0.5859 0.3612 gains and the filter parameters intelligently and linearly in order

25% 0.0101 0.6021 0.3670 to determine the off-nominal optimal controller gains and filter
+25% 0.0100 0.5449 0.3485
parameters. The on-line computational burden of SFL is found to
+50% 0.0101 0.5814 0.3598
be noticeably low. Consequently, TLBO-SFL based on-line dynamic
response is obtained for the AVR system. The simulation results
show that the proposed TLBO-SFL technique is very much effective
Table 5
Range of total deviation and percentage of maximum deviation for the AVR system and guarantees for good robust performance against parametric
tuned by TLBO algorithm. uncertainties.
Model Performance Range of total Percentage of
parameters parameters deviation maximum deviation
sA M P (p.u.) 0.0005 5
References
T S (s) 0.2335 29.93
T R (s) 0.0813 22.93 [1] Kundur P. Power system stability and control. New York: Tata McGraw Hill;
1994.
sE M P (p.u.) 0.0007 7 [2] Kiam AH, Chong G, Yun L. PID control system analysis, design, and technology.
T S (s) 0.6115 214.09 IEEE Trans Control Syst Technol 2005;13(4):559–76.
T R (s) 0.1861 24.26 [3] Gaing ZL. A particle swarm optimization approach for optimum design of PID
controller in AVR system. IEEE Trans Energy Convers 2004;19(2):384–91.
sG M P (p.u.) 0.0005 5
[4] Kim DH, Cho JH. A biologically inspired intelligent PID controller tuning for
T S (s) 1.4032 277.19 AVR systems. Int J Control Autom Syst 2006;4(5):624–36.
T R (s) 0.3271 47.58 [5] Mukherjee V, Ghoshal SP. Intelligent particle swarm optimized fuzzy PID
sS M P (p.u.) 0.0001 1 controller for AVR system. Electr Power Syst Res 2007;77(12):1689–98.
T S (s) 0.5148 7.46 [6] Kashki M, Abdel-Magid Y, Abido M. A reinforcement learning automata
T R (s) 0.0377 8 optimization approach for optimum tuning of PID controller in AVR system. In:
Huang D-S et al., editors. Advanced intelligent computing theories and
applications. With aspects of artificial intelligence, vol. 5227. Berlin,
maintaining the system’s acceptable response in abnormal cases. Heidelberg: Springer; 2008. p. 684–92.
[7] Zhu H, Li L, Zhao Y, Guo Y, Yang Y. CAS algorithm-based optimum design of PID
The behavior of the system is observed by the robustness analysis controller in AVR system. Chaos Soliton Fract 2009;42(2):792–800.
when subjected to the system uncertainties. In this paper, to [8] Coelho LS. Tuning of PID controller for an automatic regulator voltage system
observe the robustness of the proposed TLBO algorithm, the system using chaotic optimization approach. Chaos Soliton Fract 2009;39:1504–14.
[9] Chatterjee A, Mukherjee V, Ghosal SP. Velocity relaxed and craziness-based
uncertainties are represented in the form of variation of time con-
swarm optimized intelligent PID and PSS controlled AVR system. Int J Electr
stants of the different blocks of the AVR system within a range of Power Energy Syst 2009;31:323–33.
50% of the nominal value with steps of 25%. The behavior of [10] Gozde H, Taplamacioglu MC. Comparative performance analysis of artificial
the system is shown in different subplots in Fig. 13, to such varia- bee colony algorithm for automatic voltage regulator (AVR) system. J Frank
Inst 2011;348(8):1927–46.
tions along with the nominal response. The robustness is analyzed [11] Panda S, Sahu BK, Mohanty PK. Design and performance analysis of PID
with the evaluation of MP , T s and T R . In Table 4, the values of M P , T s controller for an automatic voltage regulator system using simplified particle
and T R of the system are reported under varying operating condi- swarm optimization. J Frank Inst 2012;349(8):2609–25.
[12] Mohanty PK, Sahu BK, Panda S. Tuning and assessment of proportional-
tions. Further, the range of total deviation and maximum percent- integral-derivative controller for an automatic voltage regulator system
age deviation of the transient response parameters of the AVR employing local unimodal sampling algorithm. Electr Power Comp Syst
system are calculated and presented in Table 5. From this table, 2014;42(9):959–69.
[13] Clerc M, Kennedy J. The particle swarm-explosion, stability, and convergence
it may be noted that the percentage of maximum deviation varies in multidimensional complex space. IEEE Trans Evol Comput 2002;6(1):58–73.
with a great margin. The percentage of maximum deviation of M P [14] Rao RV, Savsani VJ, Vakharia DP. Teaching-learning based optimization: a
with the variation in the value of sA is 5% and the same with the novel method for constrained mechanical design optimization problems.
variation in the value of sE is 7%. With the variation in sS , the per-
Comput Aid Des 2011;43(3):303–15.
[15] Saadat H. Power system analysis. New Delhi: Tata McGraw Hill Ltd; 2002.
centages of maximum deviation of MP and T R are found to be 1% [16] Tang Y, Cui M, Hua C, Li L, Yang Y. Optimum design of fractional order PID
and 8%, respectively. Thus, it may be said that the proposed TLBO controller for AVR system using chaotic ant swarm. Exp Syst Appl 2012;39
(8):6887–96.
based PID controller exhibits robustness in its characteristics for
[17] Pan I, Das S. Frequency domain design of fractional order PID controller for
the AVR system and is able to give the desired performance under AVR system using chaotic multi-objective optimization. Int J Electr Power
sudden changing parameter conditions of the system. Energy Syst 2013;51:106–18.
[18] Cominos P, Munro N. PID controllers: recent tuning methods and design to
specification. Proc IEE Control Theory Appl 2002;149(1):46–53.
Conclusion [19] Astrom KJ, Hagglund T. PID controllers: theory, design and
tuning. USA: Instrument Society of America; 1995.
[20] Banerjee A, Mukherjee V, Ghoshal SP. Modeling and seeker optimization based
In this paper, TLBO is presented for tuning the PID controller simulation for intelligent reactive power control of an isolated hybrid power
gains and the low-pass filter parameters with off-line nominal system. Swarm Evol Comput 2013;13:85–100.
 S. Chatterjee, V. Mukherjee / Electrical Power and Energy Systems 77 (2016) 418–429 429

[21] Rao RV, Savsani VJ, Vakharia DP. Teaching-learning based optimization: a [23] Mohagheghi S, Harley RG, Venayagamoorthy GK. Modified Takagi–Sugeno
novel optimization method for continuous non-linear large scale problems. fuzzy logic based controllers for a static compensator in a multi machine
Inform Sci 2011;183(1):1–15. power system. In: Proc 39th IAS annual meeting conf, vol. 4; 2004. p. 2637–42.
[22] Mukherjee V. Application of evolutionary optimization techniques for some [24] Banerjee A, Mukherjee V, Ghoshal SP. Intelligent fuzzy-based reactive power
selected power system problems. Ph.D. dissertation. Durgapur, India: Dept compensation of an isolated hybrid power system. Int J Electr Power Energy
Electrical Eng, National Institute of Technology; 2009. Syst 2014;57:164–77.