2015 12th International Multi-Conference on Systems, Signals & Devices

1  
2  
3   Indirect Fuzzy Sliding Adaptive Control
4  
5  
law For Nonlinear Systems
6  
7   Hafedh Abid , Ahmed Toumi
[email protected]
8  
9  
Abstract: In this paper we are interested to indirect adaptive fuzzy control of nonlinear SISO systems.
10  
The plant model is represented by a Takagi-Sugeno (T-S) type fuzzy system. The indirect adaptive
11  
fuzzy controller is basis of model reference. The adaptive algorithm exploits the lyapunov matrix and
12  
the sliding surface to adjust online the state matrices parameters of all local models. The Lyapunov ap-
13  
proach has been used to analysis and check the stability of the global closed loop system.. The plant
14  
state tracks asymptotically the state of the reference model for any bounded reference input signal. A
15   mechanical system has been used to check the performances of the proposed controller.
16  
17  
18   Keywords: Adaptive fuzzy controller, estimation state, T-S fuzzy systems, reference model, Lyapunov approach.
19  
20  
21  
22  
23   1. INTRODUCTION mechanichal system are given to show the
24   performances of suggested estimation approach and
25   Two mains approaches of adaptive control exist in the fuzzy adaptive control algorithm.
26   literature. , the first is called a direct adaptive control
27   where as the second is recognized as indrecte adaptive 2. TAKAGI -SUGENO FUZZY MODEL
28   control.
29   In the literature of adaptive fuzzy control, many works We consider a class of continuous SISO non linear
30  
are interested to affine systems. They combine classical dynamic systems which can writen as:
31  
theory with fuzzy systems to design an update
32  
parameters law such as: [9, 13, 14, 15]. Many researches  x&1 = x 2
33  
proposed an update laws basis of reference model [ 8]. 
34  
Li et al. [4] use Fuzzy systems to approximate the nonli-  x& 2 = x3
35  
near system functions. They combine the adaptive back- . (1)
36  
stepping technique with adaptive fuzzy control. Lin et al. .
37  
[5] develop an adaptive algorithm basis of type 2 fuzzy 
38  
systems and sliding mode control for nonlinear discrete-  x& n = f (x) + g (x) u + d
time systems. Li et al.[6] combines sliding mode control
39   with T-S fuzzy system to develop an adaptive controller. where x = [ x 1 x 2 ⋅ ⋅ ⋅ x n ] T is the state
40   The most important contribution of this work, which
41   vector, f(x) and g(x) are two nonlinear continuous
deals with the indirect fuzzy adaptive control for non-
42   functions, d is the disturbance.
linear continuous SISO systems is to suggest an
43   The continuous SISO non-linear system (1) can be
adjustable algorithm of plant parameters. This later
44   modeled by a Takagi-Sugeno fuzzy system. It is
includes in the same time Lyapunov matrix and slinding
45   expressed by a collection of linear local models. Each
surface. The plant model is represented by a series-
46   local model is presented by a fuzzy rule, which describes
parallel T-S fuzzy model. The controller includes three
47   a linear relations between input and output variables of
terms. The first is Parallel Distributed Compensation
48   the linear local system [1]. The ith rule of the fuzzy
"PDC" it is basis of state feedback method. The second
49   model can be presented by the following form:
use the Lyapunov matrix, the error and the row vector
50   which fixes the dynamic of the sliding surface and the ith Plant rule :
51   third term represents the feed forward term. The stability
52   analysis of the overall system has been made by the IF z 1 is M i 1 and .. and z n is M in
53   Lyapunov approach. (2)
54   THEN x& = Ai x + B i u, ; i = 1, 2...c
This work is organized as follows: Section II describes
55   the continuous type of T-S fuzzy model. The first part of
56   section III, presents the indirect fuzzy adaptive control where {M } are the fuzzy sets, u is the input vector,
ij
57   basis of reference model . In the second part of section
Ai ∈R is the state matrix, Bi ∈ Rnxm is the input matrix,
nxn
60   III, the estimation algorithm of the plant fuzzy model and
z1……zn are the premise variables.
61   the proof of the stability of the over all system are
The global fuzzy model of the nonlinear system can be
62   presented. In section IV, simulation results of a
expressed by the following form:
63  
64  
65   1

978-1-4799-1758-7/15/$31.00 ©2015 IEEE

1
 c zero in all circumstances. So, the plant evolves in accor-
∑ w i (z )[Ai x + B i u ] dance with its reference model.
i =1
x& = c (3)
∑ w i (z ) 3.1. Fuzzy control law
i =1
We propose the following local control law including
where c is the number of fuzzy rules, wi(z) is a weight three terms:
Attributed to each rule Ri, it depends on grade of
membership function of premise variables zj in fuzzy
ui = − K i x + l i r - GPε
sets Mij : (6)
n
With w i (z) = ∏ µ ij (z j ), w (z) >0 , for i = 1, … c .
j =1 i where the first term represents the fuzzy state
µij(zj) is the membership grade of zj to the fuzzy set Mij. feedback control. Its role is to place all closed-loop
The equation (3) can be written as: eigenvalues of each linear local model in the left open
c complex plane. It is used to stabilize the plant. It is
x& = ∑ hi (z )[A i x + B i u ] known as parallel distributed compensation PDC
(4)
i =1 techniques. In the PDC, each control rule is designed
wi ( z ) c from the corresponding rule of the T-S fuzzy model. The
With, h i ( z ) = c
, 0< hi(z) <1 ,and ∑ hi ( z (t )) = 1 , fuzzy controller and the fuzzy model share the same
∑ w j ( z) i =1
j =1 fuzzy sets in the premise part. However, the global
for , i = 1,…,c, controller which is obtained from an inference of all
local controllers is nonlinear [12].
3. INDIRECT FUZZY ADAPTIVE CONTROL
The ith fuzzy rule of the feedback state controller is:
The objective is to propose an adaptive controller so
that the performances of the controlled plant remains IF z1 is Mi1 and z2 is Mi2 and … and zn is Min
close to the performances of a desired model, despite Then u fi = − K i x (7)
variations in the plant parameters or uncertainties . The
classical controllers, which are designed for nominal The fuzzy rules of the feedback state controller have
parameters of system, are unable to assure required linear state feedback laws in the consequent part. Also,
performances when the parameters of plant are subject of for each rule Ri the same weight wi(z) is atributed to the
variations. Also, the stability of plant cannot be local gain controller and the local linear model.
guaranteed unless parameter uncertainty is progressively
reduced online by a suitable adaptation mechanism. To The state feedback gain parameters are computed for
surmount this difficulty, we suggest an indirect adaptive autonomous fuzzy closed loop model which is expressed
fuzzy controller basis of reference model. The plant by equation (8).
parameters are approximated on line and utilized to
compute the controller parameters. c c
The estimated fuzzy model of plant is given by a T-S x& = ∑ ∑ hi (z )h j (z )[Ai − B i K j ]x (8)
Fuzzy model where the structure of the matrix Ai and i =1 j =1
vectors Bi ( i=1,…,c) are known and the parameters of
plant are varying in time or imperfectly known . Our goal The vectors Ki are calculated such that each matrix of
is to choose the control law so that the plant state vector the closed-loop linear local model is equal to the
x tracks, in the closed-loop ,the state vector xm of a reference model matrix. So,
reference model. the latter can be described by: Ai - Bi Ki = Am
x& m = Am x m + B m r (9)
(5)
where Am is a stable matrix which has the same size of 1 i
the Ai matrix, Bm and the reference input signal r are Ki = [a - a m , ,... a ni - 1 - a nm- 1 ,a ni - a nm ]
bi 1 1 (10)
chosen such that x m = [ x m1 x m 2 ⋅ ⋅ ⋅ x m n ]T
represents the desired state vector of the plant. The second term which represents the feed forward
gain is adopted for considering the reference input r to
The basic idea of the control method is basis of the produce any reference signal xm. It will be caculated so
exploitation of the error between the model reference and that each linear local model becomes equal to the
the plant to modify the actions in order to force the plant reference model:
to track the model reference. The error is used by adap- ( Ai − Bi K i ) x + Bi li r = Am x m + Bm r
tive block to elaborate a corrective action which is added (11)
to the plant and T-S fuzzy model’s common input. The
adaptive block algorithm must adjust on-line the parame- B ilir = B m r (12)
ters of the controller so that the error converges toward

2
 c
Then l i = inv ( B iT B i ) B iT Bm e& = Am e + ∑ hi (z)(A%i x + B%i u)
i =1
(20)

Since the parameters of Ai and Bi are variable in time , ~ ~

it is not possible to compute the feedback gains Ki, and li where Ai = Ai − Aˆ i and Bi = Bi − Bˆ i .
To rise above this difficulty, we replace the matrices Ai
ˆ ˆ
and Bi by their estimated values A i and B i , that are
provided by an suitable estimation algorithm. So, Ki and For series-parallel fuzzy model we need to compute
li can be calculated on line as follows: the matrix P so that:

1 i m A mT G T GPG T G + G T GPG T GA m = −Q (21)

Ki = [aˆ1 - a1 , ,...aˆn-
i m ˆi m
1 - an-1 ,an - an ]
bˆ i
(13)
In this case the matrix P can be easily computed,
because Am is a known stable matrix.
l i = inv ( Bˆ iT Bˆ i ) Bˆ iT B m (14) We suggest the following algorithm to update the state
matrices and vector of local linear model:
We assume that (14) is invertible. & &
The third term is based on the error. It is used to A%i = Aˆi = −ri hi (z )G T GPG T S x T (22)
compensate the fuzzy approximation error and
uncertainties. & &
B% i = Bˆ i = − ri h i ( z ) G T G PG T S u
vi = − GPε
(15) Assumption1: All signals have been assumed to be
Where, G and P represent respectively, the row vector uniformly bounded.
which define the dynamic of the sliding surface and the Theorem: Consider the continuous nonlinear system
symetric definite positive matrix which verify the described by the equation (1). It is described by a T-S
lyapunov stability condition of equation (21). fuzzy model (4). The series-parallel fuzzy model (17)
The global control action is given by: and the reference model (5) are utilized to estimate the
fuzzy state model of the closed-loop system. We assume
c that the assumptions 1 and 2 are true. If we apply the
u = ∑ hi (z) (− K i x + l i r - GP ε ) (16) control law (16) with the estimation algorithm (22) to the
i =1
system, we promise that:
3.2. Estimation algorithm
By adding and subtracting Am, to the equation (4) we i) Ki(t) , li(t), e(t) are bounded for all i =1,…,c
obtain the fuzzy plant parameterisation:
c ii) e(t) → 0 , ε → 0 when t → ∞
x& = A m x + ∑ h i ( z )[( A i − A m ) x + B i u ] &
i =1
(17) iii) a&ˆi → 0 , bˆi → 0 when t → ∞
ˆ& &ˆ
The series-parallel fuzzy model is presented by the iv) Ai (t) → 0 , Bi (t) → 0 when t → ∞
equation (18) [16] [17]:
c
x&ˆ = A m xˆ + ∑ h i ( z )[( Aˆ i − A m ) x + Bˆ i u ] (18)
Proof
We suggest the following Lyapunov candidate
i =1
function:
Where, xˆ = [xˆ1 ⋅ ⋅ ⋅ xˆn ] ∈ℜ is the estimated
T n
1 T
ˆ and Bˆ are respectively the estimated Ai
state vector, A
(
V ( e , A%i , B%i ) =
2
S GPG T S )
i i
1 c  A% A%  1 c  B% B%  (23)
T T
and Bi at time t to be generated by an appropriate
adaptive law. + ∑ tr  i i  + ∑ tr  i i 
2 i =1  ri  2 i =1  ri 
We define the estimated error vector e by the
following equation:
Where, ri is a positive constant, it is called the learning
e = x − xˆ rate of the fuzzy model.S is a sliding surface. P is a
(19) symmetric definite positive matrix. It will be chosen as
the solution of the Lyapunov equation (21).
For series-parallel fuzzy model, the estimation error
should satisfy:

3
 & e ,A% , B% ) = 1 e&T GT GPGT S + S T GPGT Ge&
V( i i
2
( ) (
V& = − e T Qe )
 
1 c A A + A%i A%i 
 & T & c
~
+  ∑ tr i i
%T % ( (
+  ∑ tr h i ( z ) A iT G T GPG T
))
Sx T 
(24)  i =1  (29)
2  i =1 ri 
   c ~ 
( (
+  ∑ tr h i ( z ) B iT G T GPG T Su  ))
1 c  B% B% + B%i B%i 
&T T &
 i =1 
+ ∑ tr  i i ~
2 i =1  ri  c  − A iT h i ( z ) ri G T GPG T Sx T 
  + ∑ tr  

i =1  ri 
~
We know that: c  B iT h i ( z ) ri G T GPG T Su 
+ ∑ tr  − 

c
~ ~
e& = Am e + ∑ hi ( z ) Ai x + Bi u ( ) (25)
i =1  ri 
i =1
So, we have:
1 c
~ ~ 
(
V& = eT AmT + ∑hi (z) xT AiT + BiT u GT GPGT S )
2 i=1  V& = - e T Q e ≤ 0 (30)
1  c
~ ~  Thus, V& is negative semi definite which means V is a
+ ST GPGT GAme + ∑hi (z) Ai x + Biu 
2
( ) (26) non increasing function of time, that is, the tracking error
 
￡
i=1

~ ~ ~ ~ e is bounded. Notice that e = x − xm and xm ∈ ∞ we

1 c  AT A  1 c  BT B 
& &
+ ∑tr2 i i  + ∑tr2 i i  have that x ∈ ￡ . From (16) and r ∈￡ , we also
∞ ∞
2 i=1  ri  2 i=1  ri 
    have that x ∈￡ therefore, all signals in the closed-
∞;

loop system are bounded.

1 Since the reference trajectory xm is assumed to be
[ (
V& = eT AmT GT GPGT G + GT GPGT GAm e
2
)] bounded, the system output x is bounded.
To show asymptotic stability of the output, we note
1 c ~ ~ 
2  i=1
(
+  2∑ hi ( z) xT AiT + BiT u GT GPGT S  ) (27)
that:
∞ ∞
 ~ ~
~ ~& ~ ~&
∫ e T edt ≤ − ∫ V&(e, Ai , Bi ) = V (0) − V (∞)
1  c  AiT Ai BiT Bi  
0 0

+ 2∑ tr + This result establishes that

2  i=1  ri ri     ∞ 
     
 
∫
e(s) ∈ L2  L2 = z(t ) : z 2 (t )dt p ∞

since V ( 0)
By using the following properties of trace to  0

manipulate (27): and V (∞ ) are bounded. Since e is bounded and

e ∈ L2 , by Barbalat’s Lemma we have asymptotic
i) tr(AB) = tr (BA)
nxn stability of e (i.e, limt→∞ e = 0) .
ii) tr(A + B) = tr (A)+ tr(B) for any A,B ∈ ℜ
T T nx1
iii) tr(yx ) = x y for any x,y ∈ ℜ 4. ILLUSTRATIONS

To illustrate the performance of the presented ap-

After some manipulations on (27) using the properties proach, we choose a classical nonlinear mechanical
of trace, we obtain: system which is extensively used in the literature of con-
trol of nonlinear system :
~ ~&
 ~ ~&
c
A iT Ai BT B 
(
V& = − e T Qe + ) ∑  tr ri
+ i i 
ri 
- Mass-spring-damper
i =1
 
The system is described by the following equation [10]:
 c ~T T  (28)
+ ∑ ( (
tr h i ( z ) A i G GPG T Sx T  ))
 i =1  M&x&(t) + h1 (x(t), x& (t)) + h2 (x(t) ) = φ(x& (t))u(t ) (41)
 c ~ 
+ ∑ ( (
tr h i ( z ) B iT G T GPG T Su  ))
 i =1  where M and u(t) are respectively the mass and the force.
~& ~& The plant can be described as by the following equation:
By replacing Ai and Bi by the proposed algorithm
(22), we obtain:

4
  x&1(t) = x2(t) The feedback gains K1 , K2, K3 and K4 are computed
 online so that the closed loop eigenvalues of all local
 x& 2(t) = f(x1 ,x2 ) + g(x1,x2 )u + d (t) (43) linear model are equal to (-2).
 y(t) = x (t )
 1 We choose, G= [3 1] , r1=1, r2=1, r3=1, r4=1.
We present in figures 1, 2, 3, and 4 the simulation results
where of behavior of the state variables, errors and the control
1 2
law of the mass spring damper with parameter
f(x 1 ,x 2 ) = − (c 1 x(t) + c 2 x& (t) + c 3 x(t) + c 4 x(t) ) uncertainties. Between the time t= 15s and t=30s, the
M
1 parameter uncertainties are applied to the system
g(x 1 ,x 2 ) = 1 + c 5 x& (t) 3 (∆M=0.6 , ∆c2=0.6, ∆c3=0.04 and ∆c4=0.009).
M
The nominal parameters of system are: M=1.0, Initial conditions are: x( 0 ) = [− π 60 0]T .
c1=0, c2=1, c3=0.01, c4=0.1 and c5= 0.13. The input signal r is chosen as periodical sine wave.
The domain of state variables is [-1.5; 1.5]. Its equation is given by the following expression:
The equation system becomes: 4π
r = ( ) sin ( 5 t ) + 2 . 4 sin ( 50 t ) .
5
&x&(t ) = −x&(t ) − 0.01x(t) − 0.1x(t )3 + (1 + 0.13x& 3 (t ))u(t) Figures 1 and 2 present respectively the behaviours of
(45)
the response of the system and the desired signal. Figures
The premise variables are x(t) = z(t) = [x1 (t ) x2 (t ) ]T
3, and 4 represent respectively the behaviour of the
The membership functions are defined as[10]: position error, velocity error and the control law.

x12(t) 0.35
Evolution of x1 x1 estimate and x1d

µ11(x1(t)) = µ 21(x1(t)) = 1 −
X1estimate
X1d
X1

2 .25
0.3

0.25

2
x (t) 0.2

µ 31(x1(t)) = µ 41(x1(t)) = 1 0.15

2 .25 0.1

0.05

x 3(t)
µ12 (x 2(t)) = µ 22 (x 2(t)) = 0 .5 + 2
0

6 .75
-0.05

-0.1
0 5 10 15 20 25 30 35 40

x 3(t)
time (s)

µ 32 (x 2(t)) = µ 42 (x 2(t)) = 0 .5 − 2
6 .75 Fig. 1. Evolution of x1, x̂1 estimate and xm1 reference
(46)
in presence of parameter uncertainties
The T-S Fuzzy model of system described by equation Evolution of the state variables X2, X2dand X2estimate

(45) is defined by the following four fuzzy rules: 0.6

X2d
X2estimate
X2

0.4

Plant rules 0.2

x& = A1 x + B1u 0

Rule 1: IF x1 is µ11 and x2 is µ12 THEN -0.2

x& = A2 x + B2 u
Rule 2: IF x1 is µ21 and x2 is µ22 THEN
-0.4

-0.6

x& = A3 x + B3u
Rule 3: IF x1 is µ31 and x2 is µ32 THEN -0.8
0 5 10 15 20
time (s)
25 30 35 40

x& = A4 x + B4 u Fig.2. Evolution of x2, x̂2 estimate and xm2 reference in

Rule 4: IF x1 is µ41 and x2 is µ42 THEN
where: presence of parameter uncertainties
 0 1  0 
A1 = A2 =   , B1 = B3 =  ,
 − 0.01 − 1 1.4387 
Evolution of Errors (X1d-X1) and (X2d-X2)
0.1
X1d-X1
0.08 X2d-X2

0.06

. 0.04

0.02

 0 1  0  0

A3 = A4 =   , B 2 = B4 =  
-0.02

 − 0.235 − 1 0.5612 
-0.04

-0.06

-0.08

-0.1
0 5 10 15 20 25 30 35 40

The rules of the designed fuzzy model controller are:

time (s)

Fig.3. Evolution of the errors position and velocity

in presence of parameter uncertainties
Rule 1: IF x1 is µ11 and x2 is µ12 THEN u f (t) = − K 1 x(t)
Rule2: x1 is µ21 and x2 is µ22 THEN u f (t) = − K 2 x(t )
Rule 3: IF x is µ and x is µ THEN u f (t) = − K 3 x(t)
1 31 2 32
Rule 4: IF x1 is µ41 and x2 is µ42 THEN u f (t) = − K 4 x( t)

5
 10
Evolution of the control law
trol for a class of uncertain discrete-time nonli-
near systems, International Journal of Innovative
8

4
Computing, Information and Control 2012 ISSN
1349-4198 Volume 8, Number 1(A), January 2012
2

-2
pp.(347-359).
[6] Hongyi Li, Jinyong Yu, Chris Hilton, and Honghai
-4

-6

-8
Liu: Adaptive Sliding-Mode Control for Nonli-
near Active Suspension Vehicle Systems Using T–S
-10
0 500 1000 1500 2000 2500 3000 3500 4000
time (s)

Fig.4. Evolution of the control law u(t) Fuzzy Approach, IEEE Transactions on industrial
electronics, Volume. 60, N°. 8, August 2013
[7] M. Hojati and Saeed Gazor, “Hybrid Adaptive
Fuzzy Identification and Control nonlinear Sys-
5. COMMENTARIES tems,” IEEE Transactions on Fuzzy Systems vol 10,
N°2. April 2002 pp 198-210.
It is essential to note that the system state variables [8] Hafedh Abid, Mohamed Chtourou and Ahmed
track the desired state variables in spite of the variation Toumi, “Direct fuzzy adaptive control based
of the plant’s parameters. When a sudden change in on sliding surface for a class of SISO non-
parameters of plant is applied it appears a slight linear systems”, Int. J. Modelling, Identifica-
overshoot then the error decreases continuously to tion and Control, Vol. 16, No. 1, 2012, pp(70-
converge asymptotically to zero. The parameters of
78)
estimated state matrices of T-S model are adjusted online
[9] J. Perng Su, Tung-Min Chen and Chun-Chieh Wang
by adaptive algorithm to the new values. It is important
“Adaptive fuzzy sliding mode control with GA-
to mention that the control law is computed online after
based reaching laws,” Fuzzy Sets and Systems 120
adapting the state matrices parameters of the T-S fuzzy
(2001) pp 145-158.
model.
[10] K. Tanaka, Takayuki Ikeda and Hua O.Wang. “Ro-
bust stabilization of a class of uncertain nonlinear
6. CONCLUSIONS
systems via fuzzy control: Quadratic stability, H∞
This work deals with indirect fuzzy adaptive controller Control Theory, and linear Matrix Inequalities,”
for a class of continuous SISO nonlinear systems. The IEEE Trans on Fuzzy System, Volume 4 N°1 Feb
control action includes three terms: the first is a parallel 1996.
distributed compensation term basis of feedback control, [11] W. Chang, J.Bae Park, Y.Hoan Joob and G.chen
the second is a supervisor term which is basis of the er- "Design of robust fuzzy model-based controller
rors and Lyapunov matrix and the third is the feed for- with sliding mode control for SISO non-linear sys-
ward gain. The system remains stable in spite of the tems,” Fuzzy Sets and Systems 125 (2002) pp 1-22.
application of sudden variations of the system parameters. [12] H. Wang, Kazuo Tanaka and Michael F.Griffin “An
Simulation results illustrate the performances of the sug- approach to fuzzy control of nonlinear systems:
gested control algorithm. stability and design Issues,” IEEE Trans on Fuzzy
System, Volume 4 N°1 Feb 1996 pp (14-23).
REFERENCES [13] H. Abid Mohamed chtourou and Ahmed toumi “In-
[1] T.Takagi and M.Sugeno “Fuzzy identification of direct Adaptive Fuzzy Control Scheme Based on
systems and its applications to modelling and con- Lyapunov Approach For a Class of SISO Nonlinear
trol,” IEEE trans syst. Man,Cybern. Vol15. 116- Systems,” Studies in Informatics and Control
132 Jan/Feb1985. March 2006 volume 15 N° 1, pp (79-92).
[2] M. Sugeno “On Stability of fuzzy Systems ex- [14] J. Wang, A.B.Rad,P.T.chan “Indirect adaptive fuzzy
pressed by rules with singleton consequents,” IEEE sliding mode control: Part I: Fuzzy switching,”
Trans on Fuzzy Syst,Vol 7 N°2 Feb 1999. Fuzzy Sets and Systems 122 (2001) pp21-30.
[3] Š. Blažič, Igor Škrjanc and Drago Matko “Global [15] P. T. Chan, A. B. Rad ,J. Wang, “Indirect adaptive
Stable Direct Fuzzy Model Reference Adaptive fuzzy sliding mode control: Part II: Parameter pro-
Control,” Fuzzy Sets and Systems 139 (2003) pp3- jection and supervisory control,” Fuzzy Sets and
33. Systems 122 (2001) pp (31-43).
[4] Tsung-Chih Lin, Shuo-Wen Chang and Chao-Hsing [16] I. D. Landau, Adaptive Control: The Model Refer-
Hsu: Robuste adaptive fuzzy sliding mode control ence Approach, Marcel Dekker, Inc., New York,
for a class of uncertain discrete-time nonlinear 1979.
systems , International Journal of Innovative [17] K.S. Narendra,. and A.M. Annaswamy, Stable
Computing, Information and Control 2012 ISSN Adaptive Systems, Prentice Hall, Englewood Cliffs,
1349-4198 Volume 8, Number 1(A), January 2012 New Jersey, 1989.
pp.(347-359). [18] S. G. Cao, N. W. Rees and G. Feng “Universal fuzzy
[5] Tsung-Chih Lin, Shuo-Wen Chang and Chao-Hsing controllers for a class of nonlinear systems,” Fuzzy
Hsu: Robust adaptive fuzzy sliding mode con- Sets and Systems 122 pp 117–123 (2001).