SAMS, 2002, Vol. 42, pp.

16231637
A NEW APPROACH TO
FUZZY CONTROL OF
INTERCONNECTED SYSTEMS
MAGDI S. MAHMOUD
a,
*
, MANAR M. SABRY
b,y
and SALAH G. FODA
c,z
a
Faculty of Engineering, Arab Academy for Sciences and Technology,
P.O. Box 2033 Al-Horriya, Cairo, Egypt
b
Projects and Design Division, Saudi Arabian Texaco (J.O.),
P.O. Box 9720, Ahmadi 61008, Kuwait
c
Electrical Engineering Department, King Saud University,
P.O. Box 800, Riyadh 11421, Saudi Arabia
(Received 29 May 2000)
This paper develops a new approach to the control of interconnected system using fuzzy
system theory. The approach is based on incorporating a group of local estimators on
the system level to generate the inputoutput database. An array of feedback fuzzy con-
trollers is then designed to ensure the asymptotic stability of the closed loop system. The
developed technique is applied to an unstable large-scale system, and extensive simula-
tion studies are carried out to illustrate the potential of the new approach.
Keywords:
1. INTRODUCTION
In control engineering research, problems of decentralized control
and stabilization of interconnected systems are receiving considerable
interest in recent years [1,2] where most of the effort is focused on
*Corresponding author. E-mail: magdi@eng.kuniv.edu.kw
f
E-mail: mmanar@ncc.moc.kw

E-mail: sfoda@ksu.edu.sa
1623
dealing with the interaction patterns. It is concluded that a systematic
approach to deal with the problems of interconnected systems is two-
fold: First is to base the analysis and design effort on the subsystem
level using conventional control methods and second is to deal with
interactions effectively. These methods are facilitated, in general, by
virtue of several mathematical tools including linearization, delay
approximation, decomposition and model reduction. This constitutes
the so-called model-based control system approach for which we
have seen numerous techniques [3]. Most of the available results
have so far overlooked the operational knowledge of the intercon-
nected system under consideration. In [4], a knowledge-based
control system approach has been suggested to deal with the analysis
and design problems of interconnected systems by incorporating
both the simplest available model as well as the best available knowl-
edge about the system. For single physical systems, one of the earlier
efforts along this direction has been based on the development of
an expert learning system [56]. An alternative approach has been
to integrate elements of discrete event systems with differential equa-
tions [7]. A practically supported third approach has been the use of
fuzzy logic control by successfully applying fuzzy sets and systems
theory [9].
For interconnected systems, the foregoing approach motivates the
research into intelligent control by combining techniques of control
and systems theory with those from artificial intelligence. The main
focus should be on integrating a knowledge base, an approximate
(humanlike) reasoning and/or a learning process within a hierarchical
structure.
Fuzzy logic controllers [10] are generally considered applicable to
plants that are mathematically poorly understood (there is no accept-
able mathematical model for the plant) and where experienced
human operators are available for satisfactorily controlling the plant
and providing qualitative rules of thumb (qualitative control rules
in terms of vague and fuzzy sentences).
1. Hierarchical ordering of fuzzy rules is used to reduce the size of the
inference engine.
2. Real-time implementation, or on-line simulation of fuzzy control-
lers can help reduce the burden of large-sized rule sets by fusing
1624 M.S. MAHMOUD et al.
sensory data before imputing the systems output to the inference
engine.
A concerted effort has been made to formally reduce the size of the
fuzzy rule base to make fuzzy control attractive to interconnected
systems. Two of the difficulties with the design of any fuzzy control
system are:
. The shape of the membership functions.
. The choice of fuzzy rules.
The properties that a fuzzy membership function is used to charac-
terize are usually fuzzy. Therefore, we may use different membership
functions to characterize the same description. Conceptully, there are
two approaches to determine a membership function. The first
approach is to use the knowledge of human experts. Usually this
approach can only give a rough formula of the membership function;
fine-tuning is required. In the second approach, data are collected
from various sensors to determine the membership functions.
Specifically, the structures of the membership functions are specified
first. Then fine-tuning of the membership function parameters should
be implemented based on the collected data [8].
In this paper, we contribute to the further development of intelligent
control techniques of interconnected systems. It provides a new
approach to fuzzy control design for interconnected systems. The
approach consists of two stages: In the first stage, a group of local
state estimators is constructed to generate the data base of input-
output pairs. In the second stage, an array of feedback fuzzy controllers
is designed and implemented to ensure the asymptotic stability of the
interconnected system. Simulation studies on a large-scale system
with unstable eigenvalues are carried to illustrate the features and
capability of the new approach.
2. FUZZY SYSTEMS BACKGROUND
Fuzzy control is by far the most successful application of fuzzy sets
and systems theory to practical problems. Numerous applications of
fuzzy logic controllers to a variety of consumer products and industrial
systems have been recorded [4,9].
FUZZY CONTROL 1625
Fuzzy systems are linguistic knowledge based system. The heart of
a fuzzy system is what is so-called fuzzy IF-THEN rules. These rules
are statements in which some words are described by a continuous
membership function (Fig. 1). For example,
IF vessel temperature is high
THEN small opening of fuel value is required:
IF vessel temperature is low
THEN wide opening of fuel value is required: (1)
In general, the starting point of constructing a fuzzy system is to
obtain a collection of fuzzy IF-THEN rules from human experts,
experiments or based on domain knowledge.
The next step is to combine these rules into a single system. There are
three types of fuzzy systems that are commonly used:
1. Pure fuzzy systems,
2. TakagiSugenoKang (TSK) fuzzy systems, and
3. Fuzzy systems with fuzzifier and defuzzifier.
The three systems are described briefly hereinafter.
The configuration of a pure fuzzy system is illustrated in Fig. 2. The
fuzzy rule base represents the collection of fuzzy IF-THEN rules. The
fuzzy inference engine combines these fuzzy IF-THEN rules into a
mapping from fuzzy set in the input space UR
n
to fuzzy sets in the
output space VR based on fuzzy logic principles. If the dashed
FIGURE 1 a. Temperature membership functions; b. Valve opening membership
functions.
1626 M.S. MAHMOUD et al.
feed back line in Fig. 2 is exists, the system becomes a fuzzy dynamic
system (FDS).
The main disadvantage in the pure fuzzy system is that its input and
output is fuzzy set, whereas in design and engineering the input and
output are real-valued variables.
Takagi, Sugeno and Kang [11,12] introduced another fuzzy system
whose input and outputs are real-valued variables. This system uses
rules in the following:
IF the input x is high then the output y = cx: (2)
Where the word high has the same meaning as in (1), and c is a
constant. Comparing (1) and (2) we can see that the THEN part of
the rule changes from logistic to into a simple mathematical formula
which leads to combine the rule easier. In fact, the TSK fuzzy
system is a weighted average of the value in the THEN parts of
the rules. Figure 3 shows the basic configuration of TSK fuzzy system.
The main problem with TSK fuzzy system is its THEN part,
which may not reflect a good framework to represent human knowl-
edge. To solve this problem, the third type of fuzzy systems is used.
Figure 4 illustrates the main structure of the fuzzy system with fuzzifier
and defuzzifier.
Comparing this system with a pure fuzzy system, we can see that the
only difference between the two systems is that are the fuzzifier that
transfer the real-valued variable into a fuzzy set, and the defuzzifier
that transfer the fuzzy set into a real-valued variable.
FIGURE 2 Basic configuration of pure fuzzy systems.
FUZZY CONTROL 1627
3. STATE ESTIMATION OF INTERCONNECTED SYSTEMS
In the sequel, the terms large-scale and interconnected are used inter-
changeably. The term large-scale system (LSS) does not have a unique
established meaning, but it covers systems that possess several particu-
lar features, such as multiple subsystem, multiple control agents,
multiple objectives, decentralized and/or hierarchical information
structures [1,4]. Any LSS includes many variables but their control
is faced by a well-known fact [3] that the states are not always
available for measurement and state must be estimated.
Many authors have considered the state estimation of LSSs in input
decentralized fashion. Here we summarize one convenient algorithm
[2]. Let the state model of the ith subsystem be described by
x
i
(t) = A
i
x
i
(t) B
i
u
i
(t)
X
N
i,=j
G
ij
x
j
; (3)
y
i
(t) = C
i
x
i
(t), i, j = 1, 2, . . . , N: (4)
Where all vectors and matrices are appropriately defined and g
i
(.) is
the interaction function between the ith subsystem and the rest of
FIGURE 3 Basic configuration of TakagiSergenoKang (TSK) fuzzy systems.
FIGURE 4 Basic configuration of fuzzy systems with fuzzifier and defuzzifier.
1628 M.S. MAHMOUD et al.
the system. It is considered that (C
i
, A
i
) is completely observable for
i = 1, 2, . . . , N.
The following algorithm finds the optimal states of a LSS based on
decentralized estimation and control [4]:
Algorithm 1
Step 1: Read the matrices A
i
, B
i
and select Q
i
_0 and R
i
>0 as
weighted matrices.
Step 2: Solve the following 2N algebraic Raccati equations for H
i
, K
i
H
i
(A
T
i
I
i
) (A
i
I
i
)H
i
H
i
D
i
H
i
Q
i
= 0; (5)
K
i
(A
T
i
I
i
) (A
i
I
i
)K
i
K
i
S
i
K
i
Q
i
= 0: (6)
Where D
i
= C
T
i
C
i
, S
i
= B
i
R
1
j
B
T
:
Step 3: Integrate the following set of N simultaneous equations for
e
i
(t), i =1, 2, . . . , N, using the initial condition e
i
(0) =x
i
(0)
e
1
.
.
.
e
N
2
6
4
3
7
5
=
A
1
S
1
K
1
. . . G
IN
.
.
.
.
.
.
G
NI
A
N
H
N
D
N
2
6
4
3
7
5
e
1
.
.
.
e
N
2
6
4
3
7
5

B
1
v
1
.
.
.
B
N
v
N
2
6
4
3
7
5
(7)
Step 4: Integrate the following set of n simultaneous equations for
x
1
(t), i =1, 2, . . . , N
x
1
.
.
.
x
N
2
6
4
3
7
5
=
A
1
S
1
K
1
G
1N
S
1
K
1
0
.
.
.
.
.
.
.
.
.
G
N1
A
N
S
N
I
N
0 S
N
K
N
2
6
4
3
7
5

x
1
.
.
.
x
N
2
6
4
3
7
5

B
1
v
1
.
.
.
B
N
v
N
2
6
4
3
7
5
(8)
Step 5: Generate the inputoutput pairs {v
i
, ^ yy
i
= c
i
^ xx
i
].
FUZZY CONTROL 1629
4. INTERCONNECTED SYSTEM
Assume the following interconnected system of order 10:
A=
1:5 0:3 0:25 0:1 0:5 r1
11
r1
12
r1
13
r1
14
r1
15
0:1 0 0 0:2 0 r1
21
r1
22
r1
23
r1
24
r1
25
0 0:2 1 0 0:4 r1
31
r1
32
r1
33
r1
34
r1
35
0:6 0:1 0:25 2 0 r1
41
r1
42
r1
43
r1
44
r1
45
0:4 0:2 1 0:5 0:1 r1
51
r1
52
r1
53
r1
54
r1
55
r2
11
r2
12
r2
13
r2
14
r2
15
1:5 0:3 0:25 0:1 0:5
r2
21
r2
22
r2
23
r2
24
r2
25
0:1 0 0 0:2 0
r2
31
r2
32
r2
33
r2
34
r2
35
0 0:2 1 0 0:4
r2
41
r2
42
r2
43
r2
44
r2
45
0:6 0:1 0:25 2 0
r2
51
r2
52
r2
53
r2
54
r2
55
0:4 0:2 1 0:5 0:1
2
6
6
6
6
6
6
6
6
6
6
6
6
6
6
4
3
7
7
7
7
7
7
7
7
7
7
7
7
7
7
5
B =
0 0 0 0
1 0 0 0
0 0 0 0
0 1 0 0
0 1 0 0
0 0 0 0
0 0 1 0
0 0 0 0
0 0 0 1
0 0 0 1
2
6
6
6
6
6
6
6
6
6
6
6
6
6
6
4
3
7
7
7
7
7
7
7
7
7
7
7
7
7
7
5
C =
1 1 0 0 0 0 0 0 0 0
0 0 1 0 1 0 0 0 0 0
0 0 0 0 0 1 1 0 0 0
0 0 0 0 0 0 0 1 0 1
2
6
6
4
3
7
7
5
which is considered to be composed of two-coupled subsystems; each
of order 5. The coupling parameters are r1
jk
and r2
jk
where j and k
take values of 1, 2, 3, 4 and 5. In the sequel, we refer to the structure
of the interconnected system model as:
x =
A11 . . . G12( r1 )
.
.
.
.
.
.
G21( r2 ) A22
2
6
4
3
7
5
x
B1
. . .
B2
2
4
3
5
v (10)
1630 M.S. MAHMOUD et al.
Where G12(r1) and G21(r2) are the coupling matrices with r1 =
{r1
11
, . . . , r1
55
} and r2 ={r2
11
, . . . , r2
55
}.
For typical values of r1
15
=0.1, r1
24
=0.1, r1
42
=0.2, r2
22
=0.1,
r2
42
=0.15, r2
51
=0.11 and all other values of coupling parameters
being zeros, we examined the stability of the system by computing
the eigenvalues of matrix A. They are {1.0915, 1.0641,
0.477 j 0.0206, 0.477 j 0.00206, 0.022 j 0.0544, 0.022 j 0.0544,
1.8709 j 0.1713, 1.8709 j 0.1713, 1.9306 j 0.1413, 1.9306
j 0.1413}, and it is quite clear that there are four eigenvalues lying
in the open right half of the complex plane, and thus the interconnected
system is unstable. Further, it is easy to check that the interconnected
system is both controllable and observable.
4.1. Estimation of the System State Variables and Outputs
A Matlab program is written to implement the computational algo-
rithm (1) of section 3 on the interconnected system. Different positive
and negative step inputs are applied to estimate the outputs. The
results of two cases are illustrated in Figs. 5 and 6. It is observed
that the outputs tend to track conveniently the input signals.
FIGURE 5 Simulation results (case 1).
FUZZY CONTROL 1631
4.2. Design of an Array of Fuzzy Controller
We are going to treat the interconnected system at hand as being com-
posed of two identical and coupled subsystems. The control system to
be designed is such that each subsystem has its own fuzzy negative
feedback controller array with its input being the output of the
respective subsystem (Fig. 7). The array of fuzzy controllers is built
from a collection of individual controllers based on single-input
single-output (SISO) design. A schematic of the subsystem fuzzy con-
troller is shown in Fig. 7.
In order to build each fuzzy controller, the following steps are imple-
mented.
Step 1: The range of the inputs to each fuzzy controller [
i
,
i
] are
driven from the estimated value of the respective subsystem outputs,
where i =1, 2, 3, 4.
Step 2: 2N1 fuzzy set M
L
i
in [
i
,
i
] that are normal, consistent and
complete with triangular membership functions, as shown in Fig. 6 are
defined for each controller, where L=1, 2, . . . , 2N
i
1. That is we use
N
i
fuzzy set M
1
i
, . . . , M
Ni
i
to cover the negative internal (
i
, 0), the
other N
i
fuzzy sets M
Ni2
i
, . . . , M
2Ni1
i
to cover the positive internal
(0, ], and the center of fuzzy set M
Ni1
i
at zero.
FIGURE 6 Simulation results (case.2).
1632 M.S. MAHMOUD et al.
Step 3: The following 2N
i
1 rules are considered:
IF y
ai
is M
L
i
or y
bi
is M
L
i
THEN u is K
L
i
: (11)
Where L=1, 2, . . . , 2N
i
1, and a
i
, b
i
are the input to the fuzzy con-
troller i, and the center y
L
ai
and y
L
b
of the fuzzy set K
L
i
are chosen such
that
y
L
ai
and y
L
bi
_ 0 for 1 = 1, . . . , N
i
= 0 for 1 = N
i
1
_ 0 for 1 = N
i
2, . . . , 2N
i
1:
8
<
:
(12)
Step 4: Product inference engine, singleton fuzzyfier, and center aver-
age defuzzifier are selected to design the fuzzy controller.
4.3. Simulation Results
The behavior of the interconnected system outputs after implementing
the fuzzy controllers with unity step function input are shown in
Figs. 8 and 9. It is clearly evident that the system becomes asymptoti-
cally stable by using the negative fuzzy feedback controller array.
4.4. Performance of the Fuzzy Feedback Controller Array
Now, we examine the effect of coupling matrices on the performance
of fuzzy controlled interconnected system. Five additional cases with
FIGURE 7 Block diagram of the fuzzy feedback controller array.
FUZZY CONTROL 1633
deferent coupling ranks are implemented. Fine tuning of membership
functions was required to adjust their ranges. The following table
summarizes the test cases:
Case No. A11, A22 G12 G21 System Stability
Norm Sparsity Norm Sparsity Norm without
controller
with
controller
1 (Figs. 8,9) 2.2529 3/25 0.2 3/25 1.8028 Unstable Stable
2 (Figs. 10,11) 2.2529 12/25 0.4712 3/25 0.1803 Unstable Stable
3 (Figs. 12,13) 2.2529 3/25 .2 12/25 0.5341 Unstable Stable
4 (Figs. 14,15) 2.2529 1 3.0361 3/25 0.1803 Unstable Stable
5 (Figs. 16,17) 2.2529 3/25 .2 1 3.0364 Unstable Stable
6 (Figs. 18,19) 2.2529 1 3.0361 1 3.0417 Unstable Stable
*Sparsity equals the number of non zero elements divided by the total number of
elements.
The following figures illustrate the above test cases:
FIGURE 8 Output y
1
vs y
2
. FIGURE 9 Output y
3
vs y
4
.
FIGURE 10 Case 2 Output y
1
vs y
2
. FIGURE 11 Case 2 Output y
3
vs y
4
.
1634 M.S. MAHMOUD et al.
FIGURE 12 Case 3 Output y
1
vs y
2
. FIGURE 13 Case 3 Output y
3
vs y
4
.
FIGURE 14 Case 4 Output y
1
vs y
2
. FIGURE 15 Case 4 Output y
3
vs y
4
.
FIGURE 16 Case 5 Output y
1
vs y
2
. FIGURE 17 Case 5 Output y
3
vs y
4
.
FUZZY CONTROL 1635
5. CONCLUSIONS
This paper developed a new fuzzy control design approach to an inter-
connected system. It is shown that the approach consists of two stages:
In stage 1, a group of local state estimators was constructed to gener-
ate the inputoutput database. Then an array of feedback controllers
was designed and implemented to guarantee the overall asymptoticaly
system stability. Extensive simulation studies were performed to sup-
port the developed design approach.
References
[1] M.S. Mahmoud, M.F. Hassan and M.G. Darwish (1985). Large Scale Control
Systems: Theories and Techniques, Marcel Dekker Inc., New York.
[2] D.D. Silijak (1991). Decentralized Control of Complex Systems, Academic Press,
Boston.
[3] M.S. Mahmoud (1991). Computer-Operated Systems Control, Marcel Dekker Inc.,
New York.
[4] M. Jamshidi (1997). Large Scale Systems: Modeling, Control and Fuzzy Logic,
Prentice-Hall, New Jersey.
[5] M.S. Mahmoud, S.Z. Eid and A.A. Abou-Elseoud (September/October 1989).
A Real Time Expert System for Dynamical Processes, IEEE Transactions
Systems, Man and Cybernetics, Vol. SMC-19, No. 5, pp. 11011105.
[6] M.S. Mahmoud, S. Kotob and A.A. Abou-Elseoud (January 1992). A Learning
Rule-Based Control System. Journal of Information and Decision Technologies,
Vol. 18, No. 1, pp. 5566.
[7] M.S. Mahmoud, A.A. Abou-Elseoud and S. Kotob (April 1992). Development of
Expert Control Systems: A Pattern Classification and Recognition Approach.
Journal of Intelligent and Robotic Systems, Vol. 5, No. 2, pp. 129146.
[8] L.X. Wang (September 1994). A Supervisory Controller for Fuzzy Control Systems
that Guarantees. IEEE Trans. Automatic Control, Vol. 39, No. 9, pp. 18451847.
FIGURE 18 Case 6 Output y
1
vs y
2
. FIGURE 19 Case 6 Output y
3
vs y
4
.
1636 M.S. MAHMOUD et al.
[9] K.M. Passino and S. Yurkovick (1998). Fuzzy Control, Addison Wesley,
California.
[10] C.C. Lee (March/April 1990). Fuzzy Logic in Control Systems: Fuzzy Logic
Controller, Parts I and II. IEEE Systems, Man and Cybernetics, Vol. 20, No. 2,
pp. 404435.
[11] M. Sugeno and G.T. Kang (1998). Structure Identification of Fuzzy Model, Fuzzy
Sets and Systems, 28, 1533.
[12] Takagi and M. Sugeno (1985). Fuzzy Identification of Systems and its Applications
to Modelling and Control, IEEE Trans. On Systems, Man and Cybernetics, Vol. 15,
pp. 116132.
FUZZY CONTROL 1637
Copyright of Systems Analysis Modelling Simulation is the property of Taylor & Francis Ltd and its content
may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express
written permission. However, users may print, download, or email articles for individual use.