4th International Conference on Power Engineering, Energy and Electrical Drives Istanbul, Turkey, 13-17 May 2013

Firefly Algorithm for optimal tuning of PID controller

parameters
Bendjeghaba O Ishak Boushaki S Zemmour N
University of Boumerdes University of bab ezouar University of Boumerdes
LREEI LRIA LFMEP
Boumerdes, Algeria Alger, Algeria Boumerdes, Algeria
[email protected] [email protected] [email protected]

Abstract— This paper presents a firefly algorithm (FA) for tuning the cooperating firefly’s agents to solve optimization problems
the proportional-integral-derivative (PID) controller parameters in was introduced by Xin-She Yang at Cambridge University and
order to achieve a desired transient response. The proposed presented in [7] then extended in his further works [8]. It has
approach has superior features including easy implementation, been recently adapted for solving various design problems
stable convergence characteristic and good computational [9,10,11].
efficiency. Comparing with Ziegler-Nichols method, the proposed
method is more efficient in improuving the step response The FA has been successfully applied to the Permutation
characteristics. Flow Shop Scheduling Problems in [9], to the Clustering
problem in [10], and to Mixed variable structural optimization
Keywords-firefly algortihm ; pid controller; optimization; in [11]. Ant algorithm shows very good results in each applied
metaheuristic. area. In this paper a new tuning method is proposed and used
for tuning the PID controller parameters of three different
I. INTRODUCTION processes. The obtained results are compared through
The conventional PID controller’s is the most widely used simulation with Ziegler-Nichols (ZN) existing method. The
control strategy in many industrial processes. This popularity is simulation experimentations show that the FA based approach
due to their functional simplicity, reliability and broad outperforms the Ziegler-Nichols method.
applicability. The PID control algorithm is used to control
The paper is organized as follows. Section II presents the
approximately all loops in process industries and it is also
PID parameters tuning problem formulation. In section III we
the cornerstone for many advance control algorithms and
describe the basics of the FA. The proposed approach
strategies [1]. However, industrial implementations of PID
implementation is explained in section IV. Numerical
controllers show that their parameter tuning still presents a
simulation and comparisons are provided in Section V. Finally,
challenge in many applications. Traditionally, the problem has
some conclusions are drawn in Section VI.
been handled by a trial and error approach. In the past decades
more systemic methods have been introduced. Ziegler-Nichols II. FORMULATION OF PID PARAMETERS TUNING PROBLEM
and Cohen-Coon’s are the best known tuning methods [2,3].
Despite the hug number of proposed approaches, they are not We consider the block diagram shown in Fig.1. In this
completely systemic and more of them occasionally yield poor figure GPID ( s ) is the controller and GP ( s ) is the process to be
performance in practice. controlled. In practice the output and the transfer function (in
parallel structure) of the PID controller are given respectively
Recently, metaheuristic approaches have received increased by the equations [1]:
attention from researchers dealing with engineering control
problems. In 1997 a modified GA approach was presented by ª 1 de ( t ) º
Bagis to find the optimal parameters of the PID controller so
that the desired system specifications are satisfied [4].
u ( t ) = k p ( t ) «e ( t ) +
¬« Ti
³ e ( t ) dt + T
d
dt ¼»
» (1)

Puangdownreon and Sujitjorn have presented an adaptive tabu

search procedure based on the additional use of the back ª 1 º
tracking and the adaptive search radius mechanism for the GPID ( s ) = k p «1 + + Td s » (2)
optimum PID controller parameters [5]. ¬ Ti s ¼

To achieve the desired closed loop system response, the

results of DEA based PID tuning study are discussed by Bagis Input Output
and Savascihabes [6]. In this paper an efficient tuning approach GPID ( s ) GP ( s )
is proposed to find the optimal PID parameters .the approach is
based on a firefly algorithm. In fact, the firefly algorithm (FA) Sensor
is one of the recent meta-heuristic optimization methods
inspired by the real fireflies’ behavior. The idea of employing Figure 1. block diagram of a basic PID control system

978-1-4673-6392-1/13/$31.00 ©2013 IEEE

POWERENG 2013 1293
4th International Conference on Power Engineering, Energy and Electrical Drives Istanbul, Turkey, 13-17 May 2013

For a given plant or process, the main goal of PID tuning Where, r is the distance between any two fireflies, β0 is the
problem, is to adjust optimally as fast as possible the PID initial attractiveness at r = 0 , and γ is the absorption
controller parameters for getting a desired performance with a
good closed loop time response of the considered process. For parameter which control the decrease of the light intensity.
this purpose, the proposed PID tuning based FA is
schematically shown in Fig.2 Procedure of the CFA Metaheuristic
Begin;
FA Initialize algorithm parameters;
Kp, Ki, Kd Define the objective function of f(x), where x=(x1,........,xd)T
Input Output
PID Plant Generate the initial population of fireflies or xi (i=1, 2 ,..., n)
Determine the light intensity of Ii at xi via f(xi)
Sensor While (t<MaxGen)
Figure 2. PID tuning scheme with FA For i = 1 to n (all n fireflies);
For j=1 to n (n fireflies)
The objective of the FA is to determine theses optimal
parameters values by minimization of predetermined fitness if (Ij > Ii), move firefly i towards j;
function. In time domain, the fitness function can be formed by end if
different performance specifications such as the integral of Attractiveness varies with distance r via Exp[-Ȗr2];
time multiplied by absolute-error value (ITAE), rise time,
settling time, overshoot and steady state error. In this paper, the Evaluate new solutions and update light intensity;
following performance function is used as fitness function End for j;
during the design of PID controller:
End for i;

(
F = (1 − e− ρ )(OS + Ess) + e− ρ(ts + tr ) ) (3) Rank the fireflies and find the current best;
End while;
Where, is a weighting factor, OS , Ess , ts and tr are
ρ
Post process results and visualisation;
respectively the integral of time multiplied by absolute-
End procedure;
error value, the maximum overshoot, the steady state
error, the settling time and the rising time of the
performance criteria in the time domain. Figure 3. Pseudo code of the CFA Metaheuristic.

III. BASICS OF FIREFLY ALGORITHM The distance r between any two fireflies i and j , at
Firefly Algorithm is a nature inspired algorithms, which is position xi and x j , respectively, can be defined as a Cartesian
based on the flashing light of fireflies. In fact the algorithm has
three particular idealized rules which are based in real on some or Euclidean as follows :
major flashing characteristics of real fireflies [7]. These are the d

¦ (x )
following: (1) all fireflies are unisex, and they will move 2
rij = i,k − x j,k (5)
towards more attractive and brighter ones regardless their sex.
k =1
(2) the degree of attractiveness of a firefly is proportional to its
brightness which decreases as the distances from the other Where, xi,k is the k th component of the spatial coordinate
fireflies increases. (3) If there is not brighter or more attractive
firefly then a particular one, then it will move randomly. xi of the i th firefly and d is the dimension number.
For an optimization problem, the flashing light is associated • Movement
with the fitness function in order to obtain efficient optimal
solutions. The main steps of standard firefly algorithm are The movement of a firefly i which is attracted by a
given by the pseudo code shown in Fig.3. brighter firefly j is given by the following equation:
In this algorithm, when searching for solutions the fireflies
uses two main procedures: attractiveness and movement, which (
xi = xi + β 0 * exp −γ rij2 * x j − xi + ) ( )
are defined as follows: (6)
α * §¨ rand − ·¸
1
• Attractiveness © 2¹
The form of the attractiveness function of a firefly is the Where the first term is the current position of a firefly, the
following monotonically decreasing function: second term is used for considering a firefly’s attractiveness to
light intensity seen by adjacent fireflies, and the third term is
(
β ( r ) = β0 exp −γ r m , with m ≥ 1 ) (4) used for the random movement of a firefly in case there are not
any brighter ones. The coefficient α is a randomization

POWERENG 2013 1294

4th International Conference on Power Engineering, Energy and Electrical Drives Istanbul, Turkey, 13-17 May 2013

parameter determined by the problem of interest, while rand V. SIMULATION RESULTS

is a random number generator uniformly distributed in the The proposed approach is implemented in MATLAB
space [0,1]. language on the Pentium-4 dual core 1.66 GHz PC.
Preliminary numerical tests were used to set the values of the
IV. IMPLEMENTATION OF FIREFLY ALGORITHM FOR TUNING
firefly algorithm parameters and the best obtained ones are
PID PARAMETERS presented in Table I.
Similar to other metaheuristics optimization methods,
firefly algorithm generates random initial population of feasible TABLE I. FA PARAMETERS VALUES
candidate solutions. All fireflies of the population are handled Parameter designation value
in the solution search space with the aim to guide the search to n number of fireflies 20
the best location in the search space using their flashing light α randomness 0.15
(Fitness) and randomness. Each firefly in the population moves β0 Initial Attractiveness 0.1
in the three-dimensional (number of decision variable) search γ Absorption coefficient 1.0
space with an attractiveness that is dynamically updated based ng generation number 100
on the knowledge of the firefly and its neighbours.
A. Test Processe
Applying this search mechanism over the iterations, the FA
find progressively the optimal set of PID parameters, while In order to illustrate the validity and the efficiency of the
minimizing the fitness function. The flow chart of the proposed proposed approach, we compared the closed loop to a step
firefly algorithm for PID controller is shown in Fig.4 change with Z-N method through simulation experimentations
on three different processes obtained from the literature [4].The
transfer functions of theses processes are given by equations
Begin
(7), (8) and (9). For time delay definition in process G1 the
Create initial firefly first padé approximation is used [1]:
population
e−0.5 s
Run the system closed loop G1 ( s ) = 2
(7)
control program (s + 1)
−1
Calculate ISE, ST and OS 4.228 ( s + 0.5 )
G2 ( s ) = (8)
For each firefly, evaluate
(S 2
+ 1.64S + 8.456 )
fitness function
27
G3 ( s ) = 3
(9)
Rank fireflies and find the
current best solutions and
(s + 1) ( s + 3)
fitness function
B. Results and discussion
The closed loops to a step change of each considered
processes are compared with Ziegler-Nichols method [3]. For
Maximum iteration this purpose the ZN PID controller parameter are obtained by
number reached
n using: K p = 0.6 Ku , Ti = 0.5Tu and Td = 0.125Tu . In here, Ku
y and Tu are the gain and the period of oscillation at the stability
Stop
limit under P-control, respectively
For each test process, Table II list the PID designed by the
Figure 4. Flow chart of the proposed firefly algorithm for PID controller proposed approach and the Ziegler-Nichols method, also we
present on the same tables the obtained fitness functions and
the processes performances in terms of OS% , Ess , t s and tr .

TABLE II. BEST SOLUTIONS USING FA AND Z-N FOR DIFFERENT PROCESSES WITH ρ = 1.5

Process Design PID parameters Performance Fitness

Method KP TI TD OS% ts tr Ess F
G1 ( s ) FA 2.765 2.1907 0.877 0 1.3412 0.7628 0 0.4695
ZN 2.808 1.64 0.41 27.030 5 3.0912 0.6781 0 21.8404
G2 ( s ) FA 2.7870 2.7876 0.7939 0.0037 4.9093 0.4718 0.0021 1.2052
ZN 2.19 1.03 0.258 16.5766 5.3980 0.7288 0 14.2456
G3 ( s ) FA 2.0291 1.8064 0.4778 0.0023 1.46753 0.9749 0 0.5468
ZN 3.072 1.352 0.338 32.5527 .7207 0.6649 0 26.2678

POWERENG 2013 1295

4th International Conference on Power Engineering, Energy and Electrical Drives Istanbul, Turkey, 13-17 May 2013

The time responses for the ZN and the FA based PID It is observed from these figures that the FA based PID
controllers are also plotted on Fig.5, Fig.6 and Fig.7 controller has better set point tracking compared to Ziegler-
Nichols based PID controller.
1.4
Step Response
VI. CONLUSION
1.2
The parameters tuning problem of PID controller has been
Z-N efficiently solved in this paper by using a new metaheuristic
1
based on firefly algorithm. Three test processes have been
0.8
FA examined and results have been attractive. When compared to
Z-N method, FA results in a superior system performance in
term of time domain specifications. Finally, we can say that the
A m p lit u d e

0.6

proposed algorithm can be used as an efficient alternative to

0.4
Ziegler-Nichols tuning method.
0.2
REFERENCES
0
[1] B.C. Kuo , Automatic control systems, 6ht ed, Prentice Hall, 1991.
-0.2
[2] G.H. Cohen, G.A. Coon, “theoretical consideration of related control”,
0 1 2 3 4 5 6 7 8 9 10 Transaction on ASME 75(1953),pp-827-7834
Time (sec)
[3] J.G.Ziegler, N.B. Nichols, “Optimum settings for Automatic
controllers”, Transaction on ASME 64(1942),pp-759-768
Figure 5. Step response of G1 system with FA and ZN based PID controllers
[4] A. Bagis, “Determination of the PID controllers parameters by using
binary and real coded genetic algorithms”, Journal of information
science and engineering 23(2007), pp 1469-1480
Step Response
1.4
[5] D. Puangdownreong , S. Sujitjorn, , “Obtaining an optimum PID
controller via adaptive tabu search”, in:Proceedings of the International
1.2
Conference on Adaptive and Natural Computing Algorithms
(ICANNGA2007), Part II, LNCS 4432, (2007), 747–755.
1 [6] A. Bagis,, A. Savascihabes, , “PID tuning by using differential evolution
Z-N
algorithm for desired closed loop system response”, in: Proceedings of
0.8 the International Symposium on Innovations in Intelligent Systems and
Applications (INISTA 2010), Kayseri & Cappadocia, Turkey, June
Amplitude

FA
2010, 170–174.
0.6

[7] X. S. Yang, “Nature-inspired Metaheuristic Algorithm”, Luniver Press

(2008)
0.4
[8] X. S. Yang, “Firefly Algorithms for Multimodal Optimization,” In:
Watanabe, O., Zeugmann, T. (eds.) SAGA 2009. LNCS, vol. 5792, pp.
0.2
169–178. Springer, Heidelberg (2009)
[9] M.K. Sayadia, R. Ramanziana, N. Ghaffari-Nasaba, “A discret firefly
0
0 1 2 3 4 5 6 7 8 9 10 meta-heuristic with local search for makespan minimisation in
Time (sec)
permutation flow shop scheduling problems, International Journal of
Industrial Engineering Computations,Vol.1, pp. 1-10, 2010.
Figure 6. Step response of G2 system with FA and ZN based PID controllers [10] J. Senthilnath, S.N. Omkar, V. Mani, “Clustering using firefly
algorithm: Performance study”, Swarm and Evolutionary Computation,
V.1(1), pp 164-171,2011
[11] H. Gandomi, X. S. Yang, A. H. Alavi, “Mixed variable structural
Step Response
1.4

optimization using firefly algorithm”, Computers and Structures, Vol.

1.2 89, No. 23-24, pp. 2325-2336 (2011).
Z-N

0.8 FA
A m p litu d e

0.6

0.4

0.2

0
0 1 2 3 4 5 6 7 8 9 10
Time (sec)

Figure 7. Step response of G3 system with FA and ZN based PID controllers

POWERENG 2013 1296