Download as pdf or txt
Download as pdf or txt
You are on page 1of 6

PID Coefficient Design for the Automatic Voltage Regulator Using an Adaptive Inertia Particle Swarm Optimization

Hwan Il Kang, Min Woo Kwon, Hwan Gil Bae


Dept. of Information & Communications Eng. Myongji Univ. Yongin, Gyeonggi-do, South Korea hwan@mju.ac.kr, kasanokk@hotmail.com, bae5328@naver.com

AbstractIn this paper, we present a new particle swarm optimization. The original PSO has a weight term which is decreasing, increasing, or constant during iterations. In this paper, inertia terms are a vector instead of a scalar. Comparing a velocity and an updating term, the weight can be increased or decreased. That is, if the absolute value of velocity is larger or lesss than that of the update term, then the weight is increased or decreased, respectively. With the so-called adaptive inertia PSO algorithm we design PID coefficients for the automatic voltage regulators. We show that the proposed algorithm has a better performance than the original PSO under some conditions for the PID design of automatic voltage regulators. Keywords-a new particle swarm optimization; adaptive inertia; automatic voltage regulators; adaptive inertia PSO;

A. Modified Standard PSO algorithm PSO is initialized with a group of random particles (or solutions) and then an algorithm can be used to search for optima by updating generations of particle movements [5]. In every iteration, each particle is updated by following two best values. The first one is the best solution (fitness) it has so far achieved, called personal best (or pbest). The other best value that is tracked by the particle swarm optimizer is the best value obtained by any particle in the population which is global best (gbest). After finding the two best values, the particle updates its velocity and positions with following (1):

v(k + 1) = wv(k ) + c1r1 ( pbest (k ) x(k )) +


.

+ c2 r2 ( gbest (k ) x(k )) x(k + 1) = x(k ) + v(k + 1)

(1)

I.

INTRODUCTION

From the modified original particle swarm optimization, a weight term is increased, decreased or constant. But in this paper, we propose some conditions when a weight is increased or decreased. PID controllers have three control parameters: the derivative gain, the proportional gain and the integrating gain. Different design methods for automatic voltage regulators with PID controllers have been known such as various evolutionary methods [2,3,4]. This paper proposes a new PSO based on adaptive inertia. In Section II, definitions and the main idea of the proposed PSO and four theorems for the convergence regions used in the new PSO will be stated which are motivated from [1]. Section III gives the main results of the experiments. Conclusion will be followed in Section IV.

where

v(k ) is the current velocity vector for some particle,

x(k ) the current solution vector of the same particle, pbest (k ) the current optimal solution vector of the particle, gbest (k ) the current global optimal solution vector. The cognitive coefficient c1 is a fixed coefficient with 1 and the social coefficient c2 is a fixed coefficient with 3. r1 and r2 are uniform distribution between [0, 1]. The inertia w is
expressible as a coefficient with a fixed 1. With theses parameters, the convergence of the particle position is guaranteed using the methods in [1]. B. New PSO algorithm A new PSO has a property that the weight is updated according to the sign of difference of the absolute values of the velocity and the updated term.

II.

VARIANTS OF PARTICLE SWARM OPTIMIZATIONS

978-1-4244-9224-4/11/$26.00 2011 IEEE

If

| v(k ) |> | c1r1 pbest (k ) + c2 r2 ( gbest (k )) x(k )(


c1r1 + c2 r2 ) |
then

z z 1 X ( z ) z ( x(0) + v(0)) z w V ( z ) = B( z ) + zv(0)


If where

w(k ) is changed to w(k ) . | v( k ) | <| c1r1 pbest (k ) + c2 r2 ( gbest (k )) x(k )( c1r1 + c2 r2 ) |

X ( z ) = Z {x(k )},V ( z ) = Z {v(k )} B( z ) = Z {c1r1 pbest (k ) +c 2 r 2 gbest (k )}.

and

holds, then

w(k ) is changed to w(k ) + . We choose = 1, = 0.1.

Assuming that is expressible as

= c1r1 + c2 r2 , the characteristic equation

C. Theorems for a new PSO algorithm We present four theorems for the convergences of a new PSO algorithm. Theorem 1: Consider a discrete time polynomial P( z ) = z 2 + ez + f . For P ( z ) being Schur stable, we need the following conditions:

P( z ) = z 2 + ez + f where e = 1 w , f = w + . By theorem 1, e + f > 1 is equivalent to > 0, e + f > 1 is equivalent to 2( w + ) + 2 > , e + f > 1 is equivalent to 1 > w. Now, we complete the proof.
From theorem 2, we easily obtain Theorem 3 so we skip the proof. Theorem 3: Consider an adaptive inertia PSO described in equations (2). The systems are described by

e + f > 1 e + f > 1 , 1> f

Proof: We use the fact [6] that P ( z ) = z + ez + f is Schur stable if and only if Q( s ) = ( s 1) 2 P(( s + 1) /( s 1)) is Hurwitz stable. In addition, we use the fact [6] that

v(k + 1) = ( w )v(k ) + c1r1 ( pbest (k ) x(k )) +

+ c2 r2 ( gbest (k ) x(k ))
x(k + 1) = x(k ) + v(k + 1).

(3)

R( s ) = gs 2 + hs + j is Hurwitz stable if and only if g > 0, h > 0, j > 0. From P( z ) = z 2 + ez + f , we obtain Q( s ) = gs 2 + hs + j where g = 1 + e + f h = 22f j =1 e + f .
Now we complete the proof. Theorem 2: Consider an adaptive inertia PSO described in equations (2). The systems are described by

Assume that r1 and r2 are fixed real, > 0 and w > 0. The convergence regions are expressible as three inequalities: > 0, 0 < w < 1 + < 2 + 2( w ). Combing theorems 3 and 4, we obtain the following theorem 4 and we skip the proof. Theorem 4: Consider an adaptive inertia PSO described in equations (2). The systems are described by

v(k + 1) = ( w )v(k ) + c1r1 ( pbest (k ) x(k )) +

+ c2 r2 ( gbest (k ) x(k ))
x(k + 1) = x(k ) + v(k + 1).

(4)

v(k + 1) = ( w + )v(k ) + c1r1 ( pbest (k ) x(k )) + + c2 r2 ( gbest (k ) x(k )) x(k + 1) = x(k ) + v(k + 1).
Assume that r1 and r2 are fixed real > 0 and w > 0 . The convergence regions are expressible as three inequalities: > 0, 0 < w < 1 < 2 + 2( w + ). Proof: Using the z transform for systems in (2), we obtain (2)

Assume that r1 and r2 are fixed real, > 0 and w > 0. The convergence regions are expressible as three inequalities: > 0, 0 < w < 1 < 2 + 2( w ).

D. Convergence Region The convergence regions are depicted in Fig. 1 and = 0.08. The regions have the for = 0.1, 0.3 origin of the coordinates as a south-west corner of their boundaries. We choose a normal point as = 0.08, c1 + c2 = 4. The inertia w is expressible as a coefficient with decreasing from 1 to 0. In addition, the weight can be increased or decreased by = 0.08 if the first condition holds or second condition holds, respectively. The two conditions are described in Section 2.B.

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

A. AVR system parameters A practical AVR is used to verify the performance of the proposed PID controller. The system parameters are K A = 1, A = 1, K E = 1, E = 1, K R = 1, R = 1 and K G = 1. We study three different conditions for G = 0.7, 0.8, 1 . B. Design of PID using PSO for the AVR The lower bounds for

1.2

=0.1 =0.3 =0.08

k d , k p , k i are 0.1 and their upper

bounds are 2.0. The population size is 50 and the maximum iteration is set to 100 and we consider cases for k g = 1.0, g = 1.0. Executing the original and adaptive inertia PSOs, we obtain, in tables 1, 2 and 3, the optimal PID coefficients, minimum costs, overshoot and settling time for different and .
TABLE I. THE RESULTS OF COEFFICIENTS IN THE PID CONTROLLER WITH THE ADAPTIVE PSO ( = . 08 , = 0 . 1 ) Results for

0.8 w 0.6 0.4 0.2

= . 08 , = 0 . 1
Ki Minc 2.5739 2.8595 4.1910 4.1787 OS 0.0000 0.0001 0.0001 0.0001 ST 0.9090 1.0990 1.6970 1.6960

0.5

1.5

2.5

3.5

4.5

Adaptive

No. 1 2

Kd 0.1365 0.1457 0.1731 0.1690

Kp 0.4622 0.4879 0.5515 0.5413

0.3295 0.3493 0.3908 0.3840

Fig. 1. Convergence regions for

= 0 . 1, 0 . 3 , 0 . 08 .

Original

1 2

E. Cost Functions The cost function is


TABLE II. THE RESULTS OF COEFFICIENTS IN THE PID CONTROLLER WITH THE ADAPTIVE PSO ( = . 08 , = 0 . 5 ) Results for

Cost 2 = k =1 k | y ( k ) y r ( k ) | + st 2 + 10 8 OS 2
1000

= . 08 , = 0 . 5
Ki Minc 4.3065 4.3236 4.7196 4.1783 OS 0.0000 0.0001 0.0001 0.0001 ST 1.7790 1.7700 1.9260 1.6960

where k is the sample, y is the process output, reference. The reference voltage is 0.01.

y r is the

Adaptive

No. 1 2

Kd 0.1825 0.1898 0.2099 0.1690

Kp 0.5674 0.5850 0.6174 0.5413

0.3991 0.4110 0.4280 0.3840

Original

1 2

III.

MAIN RESULTS
TABLE III. THE RESULTS OF COEFFICIENTS IN THE PID CONTROLLER WITH THE ADAPTIVE PSO ( = . 08 , = 1 ) Results from Adaptive No. 1 2 Original 1 2 Kd 0.2879 0.1000 0.1000 0.1502 Kp 0.7015 0.3529 0.3611 0.5026

The block diagram of the linear automatic voltage regulator (AVR) with the PID controller is shown in Fig. 2.

= . 08 , = 1
Minc 9.4168 5.4613 4.5269 3.0209 OS 0.0001 0.0000 0.0001 0.0001 ST 2.9030 1.6180 1.3130 1.1210

Ki 0.4415 0.2515 0.2568 0.3602

C. Step responses and convergence results Figures 3, 5, 7, 9, 11 and 13 show the outputs for the closed loop systems with optimal coefficients in the PID controller for AVR system. Two convergences of minimum costs with the original and adaptive inertia PSOs are shown in Figures 4, 6, 8, 10, 12 and 14 for different = 0 . 1, 0 . 5 , 1 . Output responses in Figures 3, 5 , 7 and 13 show that the adaptive inertia PSO outperform the original PSO. Convergency results of minimum costs in Figures 4 and 6 show that the adaptive inertia PSO outperform the original PSO. On the other hand, output responses in Figures 9 and 11 show that original PSO outperform the adaptive inertia PSO. Convergency results of minimum costs in Figures 8, 10, 12 and 14 show that the original PSO outperform the adaptive inertia PSO. In summary, for = 0 . 1, = . 08 , the convergence results and the output responses show that the adaptive PSO outperform the original PSO.

Fig. 5. adaptive( No.2).

= . 08 , = 0 . 1 ) and the original PSO (see Table I

Output

voltages

for

an

AVR

system

with

the

Fig. ( = Fig. 3. adaptive( No.1).

. 08 , = 0 . 1 ) and the original PSO. (see Table I No.2).

6.

Convergence

tendencies

with

the

Adaptive

= . 08 , = 0 . 1 ) and the original PSO (see Table I

Output

voltages

for

an

AVR

system

with

the

Fig. 7. adaptive( Fig. ( =

= . 08 , = 0 . 5 ) and the original PSO (see Table II

Output

voltages

for

an

AVR

system

with

the

. 08 ,

4.

Convergence tendencies with the Adaptive = 0 . 1 ) and the original PSO. (see Table I No.1).

No.1).

Fig. ( =

. 08 , = 0 . 5 ) and the original PSO (see Table II No.1).

8.

Convergence

tendencies

with

the

Adaptive

Fig. 11. Output voltages for an AVR system with the adaptive( = . 08 , = 1 ) and the original PSO (see Table III No.1).

Fig. 9. Output voltages for an AVR system with the adaptive( = . 08 , = 0 . 5 ) and the original PSO (Table II No.2).

Fig. 12. Convergence tendencies with the Adaptive ( and the original PSO (see Table III No.1).

= . 08 , = 1 )

Fig. 10. Convergence tendencies with the Adaptive ( = . 08 , = 0 . 5 ) and the original PSO (see Table II No.2).

Fig. 13. Output voltages for an AVR system with the adaptive( = . 08 , = 1 ) and the original PSO (see Table III No.2).

cognitive and social indices for the tuning of coefficients in the PID controllers for AVR. ACKNOWLEDGMENT This research was supported by Basic Science Research Program through the National Research Foundation of Korea (NRF) funded by the Ministry of Education, Science and Technology under Grant 2009-0185. REFERENCES
[1] Fig. 14. Convergence tendencies with the Adaptive ( and the original PSO (see Table III No.2). A. P. Engelbrecht, Fundamentals of Computational swarm intelligence, Chichester, West Sussex, U.K., John Wiley & Sons, 2005, pp. 149-151. V. Mukherjee and S. P. Ghoshal, Intelligent particle swarm optimized fuzzy PID controller for AVR system, Electric Power Systems and Research, no. 77, pp. 1689-1698, 2007. Dong Hwa Kim, Hybrid GA-BF based intelligent PID controller tuning for AVR system, Applied Soft Computing Journal, 2008, doi:10.1016/j.asoc.2009.01.004. L. S. Coelho, Tuning of PID controller for an automatic regulator voltage system using chaotic optimization approach, Chaos, Solitions and Fractals, vol. 39, no. 4, pp.1504-1514, 2009. J. Kennedy and R. Eberhart, Particle Swarm Optimization, Proceeding of the IEEE International Conference on Neural Networks, pp. 1942-1948, 1995. S. P. Bhattacharyya, H. Chapellat and L. H. Keel, Robust Control : the Parametric Approach, Prentice Hall, Upper Saddle River, 1995, p. 53, p. 60.

= . 08 , = 1 )

[2]

[3]

IV.

CONCLUSIONS

[4]

In this paper, tuning of coefficients in the PID controller has been made by two different PSO methods. By experiments about the convergence results and the output responses, we show that the adaptive inertia PSO has outperformed the original PSO for = 0 . 1 . In future work, we will study a new adaptive inertia PSO with adpative

[5]

[6]

You might also like