Int. J. Emerg. Sci.

, 3(3), 257-267, September 2013

ISSN: 2222-4254
IJES

PID-PSO Control of MIMO Nonlinear System

Adel TAEIB, Abdelkader ChAARI
Department of Electrical Engineering, High School of Sciences and Engineering of Tunis
(ESSTT), University of Tunis, Tunisia
taeibadel@live.fr, assil.chaari@esstt.rnu.tn

Abstract. In this paper, a novel design method for determining the optimal
proportional-integral-derivative (PID) controller parameters for multi-input
multi-output (MIMO) Takagi-Sugeno (T-S) fuzzy model using the particle
swarm optimization (PSO) algorithm is presented. In order to assist estimating
the performance of the proposed PSO-PID controller, a new time domain
performance criterion function has been used. The proposed approach yields
better solution in term of rise time, settling time, maximum overshoot and
steady state error condition of the system. The proposed method was indeed
more efficient and robust in improving the step response.
Keywords: Takagi-Sugeno fuzzy model, Particle Swarm Optimization, PID
controller.

1. INTRODUCTION
PID controller, which is usually known as a classical output feedback control for
SISO systems, has been widely used in the industrial world [1] and [2]. The tuning
methods of PID control are adjusting the proportional, the integral and the
derivative gains to make an output of a controlled system track a target value
properly. Several researchers focus on MIMO control systems. Because many
industrial processes are MIMO systems which need MIMO control techniques to
improve performance, though they are naturally more difficult than SISO systems.
As we know, MIMO PID controller design has developed over a number of years.
Luyben has been proposed a simple tuning method for decentralized PID controllers
in MIMO system from single loop relay tests [3]. Yusof and Omatu presented a
multivariable self-tuning PID controller based on estimation strategies [4]. We cite
also, An auto-tuning approach for centralized MIMO PID controllers and unknown
process models is presented. Classical multivariable PID controller tuning methods
are combined with iterative feedback tuning (IFT) to develop an auto-tuning method
with a bounded number of tuning parameters [5]. A PID controller calculates an
error value as the difference between a measured process variable and a desired set
point. The controller attempts to minimize the error by adjusting the process control
inputs. Kennedy and Eberhart proposed a swarm-intelligence-based parallel
optimization algorithm PSO [6]. This algorithm shows a realistic performance on
pattern classification, optimization and controller parameters design [7]. Nowadays;

257

Adel Ta, Abdelkader ChAARIas

there are many engineering fields where they are employed. Allouya et al has been
proposed an intelligent controller of DC Motor drive is designed using PSO method
for formative the optimal PID controller tuning parameters (15). Similarly, the
method is used to determine optimum PID controller parameters for a hybrid model
of the Buck-Boost DC-DC converter control (14). Moreover, the researchers
propose fuzzy PID controllers to replace the conventional ones but the values of
these parameters of the proposed fuzzy PID controllers are optimized using particle
swarm optimization (PSO) technique (16). Most of these works, in PID controller
design methods, the most common used criteria are integrated absolute error (IAE),
the integrated of time weight square error (ITSE) and integrated of squared error
(ISE). In this paper, besides demonstrating how to employ the PSO method to
obtain the optimal PID controller parameters of a nonlinear MIMO system, but, we
have proposed new criteria. The remainder of this paper is organized as follows.
Section 2 describes fuzzy T-S modeling and parameter identification. The proposed
PID tuning algorithm based on PSO is detailed in Section 3. Simulation results and
conclusion are given in Section 4 and Section5, respectively.
1.1 T-S Fuzzy model of MIMO process

Generally, modeling process consists to obtain a parametric model with the same
dynamic behavior of the real process. In this section, we are interested to the
problem of the MIMO process identification [8]. We consider a MIMO system with
ni inputs and n0 outputs. This system can be approximated by a set of discrete time
fuzzy MISO models. We consider also two wo polynomials A and B defined by:

A a0 a1q a2q 2 ... anAq nA

(1)

B b0 b1q b2q 2 ... bnB q nB

q is a backward shift operator where (qn y(k ) y(k n)) and two integers m
and n which define a delayed sample of a discrete time signal as:

ymn (k ) y(k m), y(k m 1),..., y(k n)

(2)

The MISO models are a input-output NARX (Nonlinear Auto Regressive with
exogenous input) defined by:

yl (k 1) fl ( xl (k )) l 1, 2,..., n0

where the regression vector represented by:

xl (k ) y1 (k )0yl1 , y2 (k )0yl 2 ,..., yn0 (k )

(4)

258

ny ln0

(3)

, u1 (k )ndlul1 , u2 (k )ndlul2 ,..., uni (k )

n

nulni

ndlni

International Journal of Emerging Sciences 3(3), 257-267, September 2013

Where k denotes the discrete time, n y and nu define the number of delayed
outputs and inputs respectively, nd define the number of pure delays and fl () are
unknown nonlinear functions. MISO models are estimated independently [9], so, to
simplify the notation, the output index l is omitted and we will be interested only in
the multi-input, mono-output case. The T-S MISO rules are estimated from the
system input- output data. The base rule contains r rules of the following form:

Ri : if y1 (k ) is Ai1

and if

unu (k n 1) Ain0 then

nu

j 1

j 1

l 1 j 1

yi (k 1) air yi (k j 1) bij ui (k j 1) bil ul (k j 1) ci

it i (k 1)

(4)

i=1, 2,,r.

with i ai1...ain bi1...bin ci and is a regression vector.

The computing of the values for r model parameters in Eq.3 is obtained by
using Weighted Recursive Least Squares method (WRLS) (14) to the N sample
data
( x(k ), y (k ), k 1,..., N ) as follows:

i (k ) i (k 1) Li (k ) yi (k ) (k )iT (k 1)
P(k 1) (k )
1/ ik (k ) P(k 1) T (k )
Pi (k ) Pi (k 1) Li (k ) (k ) Pi (k 1)

(5)

Li (k )

(6)
(7)

where P(k 1) is a covariance matrix and L(k ) referred to the estimator gain
vector. A common choice of initial value is to take i (0) 0 and Pi (0) I
where is a large number.
1.2 PID Control System MIMO Process

Consider a multivariable PID control structure as shown in Fig. 1, where:

Desired output vector: Yd yd 1 , yd 2 ,..., ydn .
T

Actual output vector: Y y1 , y2 ,..., yn .

T

Error vector: E Yd Y e1 , e2 ,..., en .

T

Control input vector: U u1 , u2 ,..., un .

T

n * n MIMO processes:

259

Adel Ta, Abdelkader ChAARIas

h11 ( z )...h1n ( z )
H ( z ) ... ... ...
hn1 ( z )... hnn ( z )

(8)

C11 ( z )...C1n ( z )
C ( z ) ... ... ...
Cn1 ( z )... Cnn ( z )

(9)

n * n MIMO PID controller:

The form of kij for i, j n and n 1, 2,..., n is given by:

z
( z 1)

)
Tiij ( z 1) Tdij z
Where kp is the proportional gain, Ti is the integral time constant,
C ( z) kpij (1

(10)
and

Td is

the derivative time constant. It can be also rewritten (9) as:

C ( z ) kpij

kiij z
( z 1)

kdij ( z 1)

(1.11)
z
where ki kp Ti is an integral gain and kd kp.Td is a derivative gain.

We can also rewrite as Controller design attempts to minimize the system error
produced by certain anticipated inputs [16]. The system error is defined as the
difference between the desired response of the system and its actual response.
Performance criteria are mainly based on measures of the system error ISE, IAE,
and ITSE. They are defined as follows:
N

k 1
N

k 1
N

k 1
N

k 1

k 1

k 1

ISE e12 (k ) e12 (k ) ... en 2 (k )

(1.12)

IAE e1 (k ) e2 (k ) ... en (k )

(1.13)

k 1

k 1

k 1

ITSE ke12 (k ) ke2 2 (k ) ... ken 2 (k )

(1.14)

A disadvantage of the IAE and ISE criteria is that its minimization can result in
a response with relatively small overshoot but a long settling time because IAE and
ISE performance criterion weights all errors equally independent of time. Although
the ITSE performance criterion weights errors with time, the derivation processes of
the analytical formula are complex and time consuming. In this paper, a new
performance criterion in the time domain is proposed for evaluating the PID
controller. A set of good control parameters kp , ki and kd can yield a good step
response that will result in performance criteria minimization in the time domain.

260

International Journal of Emerging Sciences 3(3), 257-267, September 2013

These performance criteria in the time domain include the overshoot Mp , Rise
Time ( Tr ), Settling Time ( Ts ), and Steady-State Error ( Ess ). Therefore, a new
performance criterion is defined as follows: time and settling time.
n

Fit (1 exp( ))( Mpi Essi ) exp(Tsi Tri ) i 1,..., n

i 1

(15)

: The weighting factor.

2. PARTICLE SWARM OPTIMIZATION
Particle Swarm Optimization is one of optimization algorithms. It was
developed through simulation of simplified social system, and has been found to be
robust in solving continuous nonlinear optimization problems [15]. The PSO
technique can generate a high quality solution within shorter calculation time and
stable convergence characteristic than other stochastic methods [13]. PSO is a
population based search process where individuals, referred to as particles, are
grouped into a swarm. Each particle in swarm represents a candidate solution to the
optimization problem. In PSO technique, each particle is flown through the
multidimensional search space, adjusting its position in search space according to its
own experience and that of neighboring particles. A particle therefore makes use of
best position encountered by itself and that of its neighbors to position itself toward
an optimal solution. The effect is that particles fly toward a minimum, while still
searching a wide area around the best solution. The performance of each particle is
measured using a predefined fitness function, which encapsulates the characteristics
th
of the optimization problem. As example, the i particle is represented as
th
xi ( xi1, xi 2 ,..., xid ) in the dimensional space. The best previous position of the j
particle is recorded and represented as pbest ( pbesti1 , pbesti 2 ,..., pbestid ) . The index of
best particle among all particles in the group is represented by the gbest . The rate of
the position change (velocity) for particle j is represented as vi (vi1, vi 2 ,..., vid )

the modified velocity and position of each particle can be calculated using the
current velocity and distance from pbestid to gbestid as shown in the following
formulas:

vid (k 1) wvid (k ) r1c1 ( pbestid (k ) xid (k )) r2c2 ( gbestgd (k ) xid (k ))

xid (k 1) xid (k ) vid (k 1)

(16)
(17)

w is the inertia weight and can be determined by

w wmax (wmax wmin )iter / itermax

261

(18)

Adel Ta, Abdelkader ChAARIas

2.1 Implementation of a PSO-PID controller

In this paper, a PID controller using the PSO algorithm was developed to improve
the step transient response of nonlinear system. It was also called the PSO-PID
controller. The PSO algorithm was mainly utilized to determine three optimal
controller parameters, and such that the controlled system could obtain a good step
response output. For our case of design, we had to tune all the three parameters of
PID such that it gives the best output results or in other words we have to optimize
all the parameters of the PID for best results. Here we define a three dimensional
search space in which all the three dimensions represent three different parameters
of the PID. Each particular point in the search space represent a particular
combination of kp ki kd for which a particular response is obtained The
performance of the point or the combination of PID parameters is determined by a
fitness function or the cost function. This fitness function consists of several
component functions which are the performance index of the design. The point in
the search space is the best point for which the fitness function attains an optimal
value. For the case of our design, we have taken four component functions to define
fitness function. The fitness function is a function of steady state error, peak
overshoot, rise time and settling time. However the contribution of these component
functions towards the original fitness function is determined by a scale factor that
depends upon the choice of the designer. For this design the best point is the point
where the fitness function has the minimal value.
The proposed PSO-PID controller for searching the optimal controller parameters,
kp , ki , and kd with the PSO algorithm. Each individual contains three
members kp , ki and kd . Its dimension is n 3 . The searching procedures of the
proposed PSO-PID controller were shown as below:

Step 1: Initialize randomly the individuals of the population including searching

points, velocities, pbest , and gbest . Specify the lower bound and the upper bound
Vmax , Vmin , max and min .
Step 2: Evaluate the objective criterion and calculate the values of the four
performance criteria Mp , Ess tr and ts .
Step 3: Compare the individual fitness of each particle to its previous gbest . If the
fitness is better, update the fitness as gbest .

Step 4: Modify the velocity v of each individual according to (8).

Step 5:
if Vmax Vid (k 1), then, Vid (k 1) Vmax

if Vid (k 1) Vmin , then, Vid (k 1) Vmin

Step 6: Modify the member position of each individual according to (9). Such
that min max
Step 7: If the number of iterations reaches the maximum, then go to Step 8.
Otherwise, go to Step 2.

262

International Journal of Emerging Sciences 3(3), 257-267, September 2013

Step 8: The individual that generates the latest is an optimal controller parameter.
3. SIMULATION RESULTS
This section presents a simulation example to shown an application of the proposed
control algorithm and its satisfactory performance.

Figure 1. Chemical reactor diagram.

The process is a Continuous Stirred Tank Reactor CSTR which is a nonlinear

system used for the conduct of the chemical reactions [17]. A diagram of the reactor
is given in Figure. 1. The physical equations describing the process are:

h(k 1) h(k ) Te (w (k ) w2 (k ) 0.2 h(k ))

1

k1Cb (k )
w1 (k )
w (k )
(Cb 2 Cb (k )) 2

Cb (k 1) Cb (k ) Te ((Cb1 Cb (k ))
h(k )
h(k ) (1 k2Cb (k ))2

(19)

where h( k ) is the height of the mixture in the vessel, w1 , w2 the volumetric flow
rates divided by cross-sectional are as with concentration Cb1 (resp, Cb 2 ), k1 and

k1 k2 1 . The flow rate of reactor-outlet divided

by cross-sectional area of the vessel is w0 and its concentration is Cb . The

k2 are

reaction rate constants

temperature in there actor is assumed constant which is used as a test for control
techniques introduced in this paper, to demonstrate the effectiveness of the proposed
algorithms. Here h1 and h2 are the outputs, w1 and w2 are the inputs which are
uniformly bounded in the region

2, 2 . We choose:
263

Adel Ta, Abdelkader ChAARIas

x1 (k ) h1 (k 1), h (k 2), w1 (k 1), w1 (k 2), w2 (k 1)

x2 (k ) Cb (k 1), Cb (k 2), w2 (k 1), w2 (k 2), w1 (k 1)

as inputs variables, and the number of fuzzy rules is four. The setup applied in this
work was the following: the population size was 20, the stopping criterion was 30
generations, wmax was 0.5, wmin was 0.9 and c1 c2 2 .
1.5
PSO-IAE
PSO-ITSE
PSO-ISE
PSO-Fit

0.5

0
0

0.5

1
Discrete time (sec)

1.5

Figure. 2. Step response of h output using different methods.

0.7
PSO-IAE
PSO-ITSE
PSO-ISE
PSO-Fit

0.6
0.5

Cb

0.4
0.3
0.2
0.1
0
0

0.5

1
Discrete time (sec)

264

1.5

International Journal of Emerging Sciences 3(3), 257-267, September 2013

Figure. 3. Step response of

Cb output using different methods.

Table 1. Parameters of different PID controllers of height output

Tuning Method
PSO-IAE
PSO-ITSE
PSO-ISE
PSO-fit

kp
32.3431
35.7236
10.3487
3.1225

ki

kd

3.2943
16.2192
45.7988
1.2092

6.4829
9.1360
46.4369
1.2675

Table 2. Parameters of different PID controllers of concentration output

Tuning Method
PSO-IAE
PSO-ITSE
PSO-ISE
PSO-Fit

kp
31.4310
35.2360
13.4870
1.6380

ki

kd

3.8430
15.1920
44.9880
1.0645

11.7290
35.2360
40.3690
2.9735

Table 3. Performance comparison of different methods ( h output)

Tuning Method
PSO-IAE
PSO-ITSE
PSO-ISE
PSO-Fit

Overshoot
44.0341
47.1406
40.8062
0.0000

Rise Time

Setting Time

0.1104
0.0368
0.0929
0.0419

0.7981
0.4864
0.7970
0.0514

Table 4. Performance comparison of different methods ( Cb output)

Tuning Method
PSO-IAE
PSO-ITSE
PSO-ISE
PSO-Fit

Overshoot
35.0070
44.8720
38.0510
0.0000

Rise Time

Setting Time

0.0828
0.0400
0.0614
0.0417

0.6939
0.5115
0.5366
0.0511

The results of the comparison are shown in Figure 2, Figure 3. Table 2 and
Table 3 in terms of overshoot, rise time and setting time. In these figures, the blue
line denotes the output control of the T-S fuzzy model based on PSO-IAE, The
green line denote the output control of the T-S fuzzy models based on PSO-ITSE,
265

Adel Ta, Abdelkader ChAARIas

the purple line shows the output control of the T-S fuzzy model based on PSO-ISE
algorithm and the red line shows the output control of the T-S fuzzy model based on
our algorithm (PSO-Fit). In these Tables, It is observed that the overshoot in MPCFit algorithm was zero; it is lesser than those of other algorithms. The rise time was
0.0419 for first output ( h ) and it was 0.0417 for second output ( Cb ), it is lesser than
those of other algorithms. The setting time in our algorithm is 0.0514 and 0.0511,
respectively, of h output and Cb output, it is lesser than those of other algorithms.
Therefore, the simulation results demonstrate the superiority of PSO-New criteria
(PSO-Fit) based on global search comparing to the other algorithms in terms of
overshoot, rise time and setting time. The overshoot, rise time and setting time are
the major factors for comparing the performance and robustness analysis of a
system.

4. CONCLUSION
This paper presents a novel design method for determining the PID controller
parameters using the PSO method for T-S fuzzy model. The proposed method
integrates the PSO algorithm with the new performance criterion into a PSO-PID
controller. Through the simulation, the results show that the proposed controller can
perform an efficient search for the optimal PID controller parameters.

5. REFERENCES
1. Astrom, K.J. Hgglund, T. PID Controllers: Theory, Design, and Tuning, ISA, ISBN
(1995), 978-1-55617-516-9, North Carolina.
2. Suda, N. PID Control. Asakura Publishing Co., Ltd., ISBN (1992), 978- 4-254-209662, Tokyo.
3. Luyben, W. L. (1986), A simple method for tuning SISO controllers in a multivariable
system, Industrial Engineering Chemistry Product Research and Development, 25: 654660.
4. Yusof, R., Omatu, S. (1993),A multivariable self-tuning PID controller, Internal Journal
of Control, 57: 1387-1403.
5. Steffen. S and Achim. K. Auto-Tuning of Multivariable PID Controllers Using Iterative
Feedback Tuning. at Automatisierungstechnik, (2012), 60: 20-27.
6. Kennedy. J and Eberhart. C ,Particle Swarm Optimization, Proceedings of the IEEE
International Conference on Neural Networks, (1995), Australia, 1942-1948.
7. Yongwei. Z, Fei. Q, Lu. J, Lei. W and Qidi. W, Performance Criteria Research on PSOPID Control System, College of Electronics and Information Engineering, Intelligent
Computing and Cognitive Informatics (ICICCI) 2010, 316-320.
8. Lagrat. I, Ouakka. H and Boumhidi. I, Fuzzy clustring for identification of TakagiSugeno fuzzy models of a class of nonlinear multivariable systems, Facult Des
Sciences, B. P 1796 Atlas, 30000 Fez MOROCCO 2007.

266

International Journal of Emerging Sciences 3(3), 257-267, September 2013

9. Takagi. T, Sugeno. M, Fuzzy Identification of Systems and its Application to Modeling

and Control, IEEE Transactions on Systems, Man and Cybernetics 1985, 15 (1), 116132.
10. Qin. S and Badgwell. T, A survey of industrial model predictive control technology,
Control Engineering Practice 2003, 11(7), 733- 764.
11. Wei-Der. C, A multi-crossover genetic approach to multivariable PID controllers
tuning, Department of Computer and Communication, ShuTe University, Kaohsiung
824, Taiwan, Expert Systems with Applications 2007, 33, 620-626.
12. Morkos. S, Kamal. H, Optimal Tuning of PID Controller using Adaptive Hybrid
Particle Swarm Optimization Algorithm, Int. J. of Computers, Communications and
Control, ISSN 2012, 6, 101-114,.
13. Soltani, M., Chaari, A. and Benhmida, F. A novel fuzzy c regression model algorithm
using new measure of error and based on particle swarm optimization, International
Journal of Applied Mathematics and Computer Science 2012, 22(3), 617-28.
14. Hoda. P, Assef. Z and Monfared. M, Hybrid Modeling and PID-PSO Control of BuckBoost Chopper, Erezglad Elektrotechniczny (Electrical Review) 2012, 8, 187-191.
15. Boumedine. A , Brahim. G and Brahim . M, Setting Up PID DC Motor Speed Control
Alteration Parameters Using Particle Swarm Optimization Strategy, Leonardo Electronic
Journal of Practices and Technologies ISSN 2009, 8, 19-32
16. Hassen .T Ahmed. M. , El-Garhy1. M and Shimy. E, Design of PSO-Based Optimal
Fuzzy PID Controllers for the Two-Coupled Distillation Column Process, Proceedings
of the 14th International Middle East Power Systems Conference (MEPCON10), Cairo
University 2010, 19-21.
17. Demuth H Beale M, Hagan M. Neural network toolbox 5. Users guide, the math works
2007.

267