Jaya, PSO and ABC Algorithm-Optimized PID Controller

for AVR System

Department of Electrical Engineering, Heritage Institute of Technology,

Kolkata 700107, West Bengal

Abstract—This report presents an innovative approach for determining the best

tuning parameters of the PID (proportional-integral-derivative) controller for the AVR
(automatic voltage regulator) system by utilizing the Jaya, PSO and ABC algorithm.
This report aims to make betterment in the step response properties of the AVR
including a modified performance criterion and to comprehend how these algorithms
operate. Comparative research and simulation findings demonstrate that the suggested
approaches work very effectively.

I. INTRODUCTION

T HE automatic voltage regulator (AVR) is a typical component of electrical power

system that can automatically control, adjust or keep the terminal voltage output of a
generator at the desired level. The reliability of a standard power system, to this consequence,
is greatly impacted by the stability of AVR system. The Automatic Voltage Regulator system
has several ongoing issues including steady-state errors, a maximum overshoot and
inefficient oscillated transient response. The PID (proportional-integral-derivative) controller
is often applied in the AVR system to improve its performance [1].

Due to its stability, simplicity and ease of execution the PID controller is often used
nowadays in industrial control systems. Researchers and plant operators find it extremely
challenging to specify PID parameters in the AVR system in an effective manner. Many
heuristic methods including the Jaya Algorithm [1], PSO (Particle Swarm Optimization)
Algorithm [2-4], ABC (Artificial Bee Colony) Algorithm [5] etc. have been suggested to
optimize the parameters of PID controllers.

Originally introduced as heuristic algorithms, Jaya, PSO, and ABC are straightforward yet
effective optimization algorithms that are meant to approach optimal solutions. The Jaya
method may be easily applied to optimization issues in the practical world like restricted
mechanical layout difficulties [6], optimal power flow [7] etc. since it simply requires generic
control parameters like the population size and number of iterations [1]. PSO is an
optimization approach that obtains inspiration from nature and is used to create
approximations for both optimization and search issues. The population-based optimization

Page | PAGE \* MERGEFORMAT 15

technique known as the ABC technique was motivated by honey bee foraging behaviour [5].
It directs the search process toward better solutions by using a fitness function to assess the
quality of the results.
This report's remaining sections are arranged as follows: AVR systems without PID
controllers are modelled in Sect. 2 and AVR system including PID controller is modelled in
Sect. 3. Sections 4,5,6 provide descriptions of the Jaya, PSO and ABC algorithms. Section 7
presents the Jaya, PSO and ABC optimized PID controller and their simulation results.
Section 8 offers a succinct overview and conclusion.

II. AVR SYSTEM WITHOUT PID CONTROLLER

A. AVR modelling

Maintaining a generator's terminal voltage magnitude at the intended level is the major job of
an AVR system. The exciter, amplifier,sensor andgenerator are the four primary parts of the
AVR system. Below is a block schematic of a basic AVR system [1] :

Fig.1 Block diagram of a basic AVR system

The transfer functions, parameter range and selected values of the aforementioned four
components are shown in the below table [8] :

TABLE I
Name Transfer Function Parameter Range Selected Values
10 ≤ K A ≤ 40 K A =10
KA
Amplifier 0.02 ≤ τ A ≤ 0.1 τ A =0.1
1+ τ A S

1 ≤ K E ≤ 10 K E=1
KE
Exciter 0.4 ≤ τ E ≤ 1.0 τ E=0.4
1+ τ E S

0.7 ≤ K G ≤ 1.0 K G =1
KG
Generator 1.0 ≤ τ G ≤ 2.0 τ G =1
1+ τ G S

Page | PAGE \* MERGEFORMAT 15

 KS 0.9 ≤ K S ≤ 1.1 K S =1
Sensor
1+ τ S S 0.001 ≤ τ S ≤ 0.06 τ S=0.01

The AVR System’s transfer function is defined as follows [8,9] :

0.1 S+10
G ( S )= 4 3 2
0.0004 S +0.0454 S +0.555 S +1.51 S +11

The voltage at the terminals of an AVR system is displayed in the below figure :

Fig.2 Steady state response of an AVR system without PID Controller

With a maximum overshoot M p of 50.66%, a rise time T r of 0.2607 s, a settling time T s of

6.9856 sand a steady-state error E ssat 0.091 p .u the system exhibits a high oscillatory
response. These types of responses are completely unacceptable. These advantageous features
make the PID controller a typical installation in the AVR system [1].

III. AVR SYSTEM WITH PID CONTROLLER

The objective of the PID controller is to make improvements in the system’s dynamic
response performance and to reduce or eliminate steady-state errors. The PID controller’s
transfer function can be described as follows :

Ki
Gc ( S )=K p + +K d S
S

where K p , K i and K d are coefficients to be tuned for the proportional, integral and derivative
terms, respectively.

Page | PAGE \* MERGEFORMAT 15

 The system performance is impacted by these three coefficients in various ways. K p
reduces rise time T r but does not remove steady-state error E ss. K i lowers steady-state error.
By minimizing the overshoot, K d effectively dampens the dynamic response and improves
the system's stability [1]. For a typical AVR system, a PID controller was installed. Below is
a block diagram of an AVR system that uses a PID controller :

Fig.3 AVR system with a PID controller

Performing auto tuning of PID Controller parameters this is the response of an AVR System
that was obtained in Simulink :

Fig.4 Steady state response of an AVR system with auto tuned PID controller

In order to obtain the suitable tuning parameters used in the PID Controller for an AVR
System the following algorithms are going to be utilized.

IV. JAYA ALGORITHM

A. Concept

Let us assume that m(h=1 ,2 , … , m)is the number of design variables, g( j=1 , 2, … , g)is the
highest number of iterations and n(i=1 ,2 , … , n) is the population size and suppose we have

Page | PAGE \* MERGEFORMAT 15

to optimize the objective function f (x). Among all potential contenders for a solution,
assume the best one is the best value of f (x) and the worst is the worst value of f (x). If the
value of the hth variable during the j th iterations is denoted by X h ,i , j for the i th candidate, the
following update is made to this value [6] :

X h ,i , j = X h ,i , j+ r 1 , h , j ( X h ,best , j−¿ | X h ,i , j|)−r 2 , h , j ( X h ,worst , j−¿ | X h ,i , j|)

,
(1)
Where the value of variableh for the best value of f (x) is X h ,best , jand the value of variable h
for the worst value of f (x) is X h ,worst , j. A new value of X h ,i , j is X ,h ,i , j.r 1 , h , j and r 2 , h , j are two
random valueswithin the range [0, 1] of the hth variable during the j th [Link] term r 1 , h , j (
X h ,best , j−¿ | X h ,i , j|) depicts the result as it gets closer to the best value and the termr 2 , h , j (
X h ,worst , j−¿ | X h ,i , j|) portrays the result that refrains the worst solution. X ,h ,i , j is accepted if the
better value of f (x) is given [1]. Upon completion of every iteration,all accepted values for
the variables are kept and taken into consideration as the input for the subsequent iteration.

B. Algorithm framework

1) Initialize m = the number of variables, n = the size of population and g = the number of
iterations

2) Selection of n candidate solutions randomly

3) Repeat

4) Assessment of their fitness values

5) Identification of Best and Worst Solutions

6) Calculation of X ,h ,i , j based on Equ.(1)

7) if X ,h ,i , j< LBv X ,h ,i , j> UB then

8) Selection of r from U(0,1) randomly

9) X ,h ,i , j = LB+r .(UB – LB)

10) endif

11) if f ( X ,h ,i , j ¿< f ( X ,h ,i , j ) then

12) X h ,i , j=X ,h , i , j

13) f min =f ( X ,h ,i , j )

Page | PAGE \* MERGEFORMAT 15

14) endif

15) until stop condition is reached

16) the minimum fitness value f min and the corresponding solution X ,h ,i , jis returned

Where x ϵ [ LB ,UB ].U ( a , b) is uniformly distributed between a and b [1].

C. Algorithm flowchart

V. PSO ALGORITHM

A. Concept

PSO is founded on the ideas of cooperation and information exchange within a group and is
modelled after the social behaviours of fish schools and flocks of birds. To identify the best
solution in PSO, a population known as "particles" traverses a multidimensional search space.
The fundamental tenet of PSO is that particles collaborate by disclosing the discovery of their

Page | PAGE \* MERGEFORMAT 15

best solutions [5]. This social aspect of PSO helps the swarm of particles converge towards
optimal solutions in the search space. PSO is frequently utilized for solving optimization
problems where the objective function does not have a known mathematical form or when
other optimization techniques are less effective [10]. It has applications in various fields,
including engineering, data analysis, robotics and more.

B. Algorithm framework

1) Initialization:
 A population of particles with random positions and velocities within the problem's
search space is initialized
 To evaluate the quality of each particle's solution the fitness function is defined

2) Define Parameters:
 Parameters like the number of particles, maximum iterations, inertia weight,
cognitive and social coefficients w , c 1 , c 2 and the problem-specific constraints are
defined

3) Update Particle Velocity and Position:

 Update the position and velocity of each particle depending on the following
equations:
 Position update: x i ( t +1 )=x i ( t ) + v i ( t +1 )
Where v i(t) is the velocity of particle iat time t , x i (t) is the position of particle i at
time t , pBest i is the best position found by particle i so far, gBest is the best position
found by any particle in the whole population, the inertia weight is w and c 1and c 2
are the cognitive and social coefficients, r 1 and r 2 are random values between 0
and 1.
 Velocity update: v i ( t+ 1 ) = w * v i(t) + c 1 *r 1 * ( pBest i −xi (t)) + c 2 * r 2 * (
gBest−x i (t))

4) Update Personal Best p Best and Global Best gBest :

 Update the pBest of each particle if its current position has a better fitness value
than its previous pBest value
 Update the gBest if any particle's gBest has a better fitness value than the current
gBest

5) Termination:
 Update particle positions and velocities until a certain number of iterations or a
termination condition is reached

6) Result:

Page | PAGE \* MERGEFORMAT 15

  The solution represented by the gBest particle or the best position found in the
population is returned as the final output

C. Algorithm flowchart

VI. ABC ALGORITHM

A. Concept

Page | PAGE \* MERGEFORMAT 15

The ABC algorithm is a population-based optimisation technique that took its cues from the
foraging behaviour of honeybees. This algorithm uses an artificial bee population to
iteratively find the best solution to a given problem. Bees share information about possible
fixes among themselves in order to take advantage of their collective wisdom.

The algorithm consists of three main phases that represent the behaviour of different
types of artificial bees [11] :

1) Employed Bee Phase :

 The employed bees symbolize a portion of the population
 They explore the search space by introducing small changes to their current
solutions
 Current solution is replaced by the modified solution if it is found to be superior
 The objective of this phase is to improve existing solutions through exploration

2) Onlooker Bee Phase :

 Based on employed bees' fitness values onlooker bees select solutions to explore
 Solutions having better fitness value have a higher chance of being chosen
 Onlooker bees investigate particular solutions and potentially enhance them
 This phase improves exploration of possible solutions

3) Scout Bee Phase :

 After identification of the unimproved solutions, scout bees trash them and come up
with fresh, random solutions.
 By doing this, variety is preserved and local optima are avoided.
 The scout bee stage encourages flexibility and discovery.

The ABC algorithm directs the search process towards better solutions by assessing the
quality of solutions using a fitness function. It is renowned for being easy to use and efficient
in locating solutions that are close to ideal and is utilized to solve a variety of optimisation
issues.

B. Algorithm framework

1) Initialization:
 Create an artificial bee colony and start it off. Each bee represents a possible way to
solve the optimisation problem
 Assign employed bees from the population
 Establish boundaries and endpoints (e.g the number of iterations or the required
level of solution quality)

2) Employed Bee Phase:

Page | PAGE \* MERGEFORMAT 15

  Examine the neighbourhood to alter the present solution for every bee that is
employed
 Apply the fitness function of the problem to the new solution to assess its fitness
 Update the employed bee's solution if the new one proves to be superior

3) Onlooker Bee Phase:

 Onlooker bees use the fitness values of employed bees to choose which solutions to
explore
 Select solutions whose likelihood is in line with their level of fitness.
 Similar to employed bees, explore and potentially enhance the selected solutions

4) Scout Bee Phase:

 Determine which solutions, after a certain number of iterations, have not improved
 Reject these solutions and create new, arbitrary solutions to swap out those
 Preserve diversity and encourage exploration

5) Update Best Solution:

 Record the best solution that any bee in the population has found

6) Termination Check:
 Verify whether the termination requirements (e.g the highest number of iterations or
the required quality of the solution) have been satisfied
 If the criterias are met, proceed to the next step; otherwise, go back to the employed
bee phase

7) Output:
 The final output is the best solution the algorithm could find and it represents the
optimal or nearly optimal solution to the optimisation problem

C. Algorithm flowchart

Page | PAGE \* MERGEFORMAT 15

 VII. JAYA, PSO & ABC ALGORITHM OPTIMIZED PID CONTROLLER IN AVR SYSTEM

A. Formulation and application of fitness function

In any optimisation problem, an essential role is played by the fitness function. To obtain
optimised system variables, the selected optimisation algorithm must essentially maximise or
minimise a mathematical expression. Four fitness functions (FF) — IAE (Integrated Absolute
Error), ITAE (Integrated Time Absolute Error), ISE (Integral of Squared Error) and ITSE
(Integrated of Timeweighted-squared Error) — are often used extensively in research to
select the best gain value of PID controller.
The optimisation of PID controller parameters for the proposed AVR system based on
Jaya, PSO & ABC algorithm is accomplished by minimising an integrated time error
function. Because of its improved results and ease of implementation, ITAE is the most
frequently used error integrating FF [13,14]. Because of the squared error, the ITSE and ISE
yield extremely implausible outcomes and are, hence, regarded as extremely violent
functions. Furthermore, the ITAE is thought to be a better option than the IAE since it
generates more accurate error indexing due to the ongoing addition of the time multiplying
error function. In this study, ITAE is framed as the fitness function to maximise the system's
stability and dynamic response The ITAE can be stated mathematically as :

Page | PAGE \* MERGEFORMAT 15

 ∞
ITAE = ∫ t |e ( t )|dt
0

where e(t) or the error signal, is defined as the arithmetic difference between the reference
voltage (set-point) and the measured voltage at the AVR output terminals, and t is the
simulation time [12]. The suggested parameter optimisation in the AVR system using Jaya,
PSO, and ABC is shown in the block diagram by :

|u| x 1 ITAE
s
Ab Produc Integrat
t or delF

Cloc

Fig.5 Jaya, PSO & ABC optimized tunning of PID in AVR system

Page | PAGE \* MERGEFORMAT 15

 Noteworthy is the fact that only at the lowest Fitness Function values can the optimal
PID gains ( K p , K i and K d ) of the examined AVR system be guaranteed to exhibit optimised
dynamic behaviour. MATLAB 2017 was used to carry out the Jaya, PSO, and ABC coding
by taking population size as 20 and iteration value as 50. MATLAB/SIMULINK was used to
model the AVR system under consideration. The SIMULINK model, in which the Jaya, PSO,
and ABC Algorithms are utilised to minimise the FF, is combined with the computed values
of ITAE for the PID controller before being sent into the MATLAB workspace. Ultimately,
the ideal transient response of AVR system is obtained by feeding the optimised PID
coefficient values into the SIMULINK model of the PID regulator after the highest number of
iterations have been completed. Throughout the whole simulation, the best transient response
for the generated AVR system is obtained by the suggested controller with the least amount
of overshoot and settling time[12].

B. Performance

In order to achieve the ideal pairing of PID regulator gains that provide the AVR system's
ideal dynamic response, the Jaya, PSO and ABC algorithms are used in the system. To
complete the aforementioned objective, the said algorithms have each been used to minimise
FF value. The following figure graphically illustrates the behaviour of convergence of the
suggested techniques:

Fig.6 Convergence behaviour of Jaya, PSO & ABC algorithm

The above plot makes it pretty clear that as the number of generations increases, the
FF's magnitude decreases. Since the FF is represented by ITAE, the process of its minimising
guarantees the advancement towards the ideal vicinity. It is noteworthy that, when an
optimization process’s performance is assessed, two crucial parameters that must be taken
into account are the ultimate minimised or maximised value of the FF and the convergence
rate. While the former quantity determines the solution’s quality that the algorithm is able to

Page | PAGE \* MERGEFORMAT 15

get, the latter quantity provides clear details regarding the rate of convergence [12,15]. The
figure illustrates that the suggested optimisation strategies produce a satisfactory solution
with an appropriate rate of convergence and quality of solution.
A comparison of the convergence quality between the proposed techniques for the
tuning of AVR is given in the below tables :

TABLE II

Optimization Algorithm Value of minimised FF

Jaya 0.05199

PSO 0.051968

ABC 0.052616

TABLE III

Optimization Algorithm Optimized Gain Values of PID

Jaya K p=0.8879 , K i =0.643 , K d=0.28438

PSO K p=0.91197, K i=0.63631 , K d =0.29016

ABC K p=0.43543 , K i =0.90188 , K d=0.56583

C. Transient response analysis

Full details regarding the transient response assessment of the suggested AVR system
depending on Jaya, PSO and ABC are provided in this section. The examined system is
stimulated with a unit step input signal in order to evaluate its transient response. Three
important indicators were utilized in this study for transient response analysis — settling
time, overshoot and undershoot. The time taken by the response to reach 2% of the steady-
state value is known as the settling time [12].

Page | PAGE \* MERGEFORMAT 15

Based on % overshoot, % undershoot and settling time, the Jaya, PSO and ABC optimisation
algorithms were compared to determine the best AVR design. The findings are shown in the
table below :

TABLE IV

Comparison of Dynamic Response

Methods
of PID
tuning %overs h oot
Ts %unders h oot

Jaya 1.3 16 5

PSO 1.95 18 10

ABC 8 2 20

Below is a graphic representation of the results :

Fig.7 Transient response comparison of Jaya, PSO & ABC based AVR system
VIII. CONCLUSION

VIII. CONCLUSION

Page | PAGE \* MERGEFORMAT 15

In this study, the optimal tuning parameters of PID controller were found by applying the
Jaya, PSO and ABC algorithms. The effectiveness of the PID controller in a typical AVR
system was also assessed by using a modified performance criterion, known as the ITAE. The
outcomes of the simulation unambiguously illustrates that the suggested algorithms-
optimized PID controller operates incredibly well in the AVR system; it is reliable and
resilient to uncertainties [1]. When comparing the PID controllers optimised by the Jaya, PSO
and ABC algorithms, the findings show that the Jaya optimised controller performs better
than the others across a number of performance metrics. The simulation results show that the
Jaya based controller tuning provides better transient behaviour as compared to PSO and
ABC in terms of% overs h oot ,T s,% unders h oot values [12]. Therefore, the comparative
analysis demonstrates that the suggested Jaya-based AVR controller offers a superior design
that is resilient to variations in system parameters because it is not much influenced by them.

Page | PAGE \* MERGEFORMAT 15