Nonlinear Dyn

https://doi.org/10.1007/s11071-018-4336-5

ORIGINAL PAPER

Fuzzy model-based nonfragile control of switched

discrete-time systems
Bo Wang · Dian Zhang · Jun Cheng ·
Ju H. Park

Received: 18 December 2017 / Accepted: 1 May 2018

© Springer Science+Business Media B.V., part of Springer Nature 2018

Abstract This paper is concerned with the problem numerical simulation is provided to show the effective-
of nonfragile control for a class of fuzzy switched sys- ness of achieved result.
tems with mode-dependent time-varying delays. By
constructing novelty Lyapunov function and employ- Keywords Switched system · Mode-dependent ·
ing a novel extended reciprocally convex discrete-time Nonfragile control · Discrete-time · Fuzzy model-
inequality, new sets of delay-variation-dependent sta- based
bility criteria are derived. A nonfragile fuzzy controller
is developed, via the parallel distributed compensa-
tion technique and stochastic analysis method. At last, 1 Introduction

This work was supported in part by the National Natural Switched systems are dynamical systems that con-
Science Foundation of China (NSFC: 61703150), the Natural sisted by a number of continuous-state or discrete-
Science Foundation of Shandong Provinces of China
(ZR2018LC010), the Program for Innovative Research Team of
state subsystems and switching signals. It is noted that
the Higher Education Institutions of Hubei Province switched systems are widely used due to its applica-
(T201812), the Natural Science Foundation of Hubei Provinces tion in many practical systems subject to different cate-
of China (2016CFB211). The work of J.H. Park was supported gories, such as mechanical systems, economic systems
by Basic Science Research Programs through the National
Research Foundation of Korea (NRF) funded by the Ministry of
and automotive highway systems. Up to now, many
Education (Grant Number NRF-2017R1A2B2004671). remarkable results have been achieved in the control
community over the past decades [1–10]. It is men-
B. Wang tioned that with further study on switched systems,
School of Electrical Engineering and Electronic
the major reported results aforementioned are focused
Information, Xihua University, Chengdu 610036,
People’s Republic of China on switched linear systems or simple dynamic non-
D. Zhang · J. Cheng (B)
linear systems, and many traditional used approaches
School of Automation and Electrical Engineering, Qingdao are average dwell time and linear matrix inequal-
University of Science and Technology, Qingdao 266061, ity [11,12]. The practical systems are very complex,
People’s Republic of China nonlinearities would be encountered inevitably, and
e-mail: [email protected]
the obtained results for switched linear systems can-
J. H. Park (B) not be applied to switched nonlinear systems directly
Department of Electrical Engineering, Yeungnam
University, 280 Daehak-Ro, Kyongsan 38541,
[13–15]. Therefore, the analysis and synthesis for gen-
Republic of Korea eral switched nonlinear systems have not been fully
e-mail: [email protected] studied yet.

123
 B. Wang et al.

It is well recognized that as an efficient technique Plant Rule i:

to deal with nonlinear systems, Takagi–Sugeno (T– IF ϕ1 is Xi1 and · · · and ϕ f is Xi f
S) fuzzy model is an alternative approach [16–20]. THEN
The switched nonlinear systems can be represented as 
convex of switched linear systems with the help of x(k + 1) = Ai,rk x(k) + Bi,rk x(k − τrk (k)) + Ci,rk u(k)
T–S fuzzy model. Hence, many valuable results are x(k) = φ(k), k = {−τ M , −τ M + 1, . . . , 0}
achieved on the fuzzy model-based switched nonlin- (1)
ear systems [4,21–28]. It should be pointed out that
the time delays in the most of derived results are where x(k) ∈ Rn , u(k) ∈ Rm are the vectors of states
mode-independent. It is well known that in the fuzzy and control input, respectively. The initial function
switched systems, the mode-dependent time-varying ψ(k) is defined on [− τ M , 0] with τ M = max{τ Ms , s ∈
delay may achieve less conservative criterion, and it N = {1, 2, . . . , N }}. rk : [0, + ∞) → N is the dis-
requires the controller accesses precise switching infor- crete switching signal, rk = s; it represents the sth sub-
mation. Thus, the mode-dependent control is often system is activated. (Ai,s , Bi,s , Ci,s ) means the matri-
introduced for switched nonlinear systems. ces subject to proper dimensions.
Note that the topic of nonfragile control problem Similar to [22], variable τs (k) stands for the mode-
might be very significant as well as the filter design; this dependent time-varying delay that satisfies
is due to the fact that the existence of small uncertainties τms ≤ τs (k) ≤ τ Ms
in controllers may lead to the unstable of closed-loop
where τms > 0 and τ Ms > 0 are known constant.
systems. Note that the literature [29–31] may involve in
Note that i j (ϕ j (k)) stands for the grade of func-
uncertain phenomenon and the obtained results might
tion ϕ j (k) in i j ; ϑi (ϕ(k)) is a membership function
degrade the performance of systems. To guarantee the
satisfying the following properties
desired performance of system, it is of importance to
devise a nonfragile control law such that some admis- 
g

sible uncertainties in terms of feedback control will ϑi (ϕ(k)) = 1, and ϑi (ϕ(k))

be not sensitive. For more results on this subject, the i=1

readers can refer to literature [31–33] and the relevant 

f 
g 
f

references. To the best of our knowledge, little effort = i j (ϕ j (k))/ i j (ϕ j (k)),

j=1 i=1 j=1
has been devoted to the nonfragile control problems
of fuzzy switched system. One of our main aims is to where ϑi (ϕ(k)) is positive function.
solve this issue. By employing the singleton fuzzifier, product infer-
Taking the above analysis into consideration, the ence and the center average defuzzifier, fuzzy system
purpose of this paper is to examine the problem of non- (1) can be rewritten as
fragile control for a class of fuzzy switched systems 
g

with mode-dependent time-varying delays. A multiple x(k + 1) = ϑi (ϕ(k)) Ai,s x(k)
Lyapunov function subject to mode-dependent time- i=1

varying delay is developed, which contains a different + Bi,s x(k − τs (k)) + Ci,s u(k) . (2)
type of switching law and an extended reciprocally con-
vex discrete-time inequality. The nonfragile control is Due to uncertainties of the controller implementa-
designed by means of the LMIs and sojourn probabil- tion, the nonfragile fuzzy controller is devised via using
ity approach. Finally, numerical example is presented the parallel distributed compensation (PDC) technique.
to illustrate the correctness and effectiveness of the pro- Plant Rule i:
posed theory. IF ϕ1 is Xi1 and · · · and ϕ f is Xi f
THEN
u(k) = −(K j,s + K j,s )x(k) (3)
2 Preliminaries
where K j,s is the state feedback controller matrix of the
Consider the following discrete-time switching system jth subfuzzy. K j,s denotes the drift of the controller
subject to fuzzy rules: and K j,s = E j,s F j,s (k)H j , where given matrices

123
Fuzzy model-based nonfragile control

E j,s and H j have appropriate dimensions. The uncer- In addition, if

tain matrix F j,s (k) are unknown time-varying matrices

satisfying F j,s (k)F j,s (k) < I . State feedback fuzzy E  x(k) 2 = 0
controller (3) can be written as
for any initial conditions, then system (8) is said to be

g mean square asymptotically stable.
u(k) = − h j (ϕ(k))K j,s x(k) (4)
j=1 Lemma 2.1 [24] For a given symmetric positive defi-
nite matrix R, integers b ≥ a, any sequence of discrete-
where
time variable x, the following inequality holds:
K j,s = K j,s + K j,s . (5) 

b−1
ζ1
(b − q) η (m)Rη(m) ≥
Combining (2) with (4), one has the following ζ2
m=a
closed-loop fuzzy switched system (CLFSS):
ζ1
diag{R, ρ(a, b)3R}

g 
g
 ζ2
x(k + 1) = ϑi (ϕ(k))ϑ j (ϕ(k)) Ai j,s x(k)
where η(k) = x(k + 1) − x(k), ζ1 = x(b) − x(a), ζ2 =
i=1 j=1 
 x(b) + x(a) − 2 bm=a x(m)/(b − a + 1), ρ(a, b) =
+Bi,s x(k − τs (k)) (6)
(b − a + 1)/(b − a − 1) for b − a = 1, and ρ(a, b) = 1
where for b − a = 1.

Ai j,s = Ai,s − Ci,s K j,s . Lemma 2.2 [34] For a real scalar α ∈ (0, 1), a sym-
metric matrix R > 0 and any matrix S, the following
To reflect the SP characteristics of the systems better, matrix inequality holds:
∀s ∈ N , k ∈ Z + , the indicative variables βs (k) are
employed
1
 αR 0
1, rk = s, 0 1
1−α R
βs (k) = (7)
0, otherwise (2 − α)R − (1 − α)S R −1 S  S
≥ .
0 (1 + α)R − αS  R −1 S
N
where s=1 βs (k) = 1, E{βs (k)} = βs .
Note that βs denotes the sth SPs of systems, obvi- Lemma 2.3 [35] Let A, B, C and D be a real matrices
ously, βs ∈ (0, 1). Let ks represents the times in [1, k]; of appropriate dimensions with A satisfying A = A,
one has the following property: then

A + BDC + C  D B  < 0, D D < I

βs = lim ks /k.
k→∞
if and only if there exists a scalar  > 0 such that
From (6)–(7), we have
A +  −1 B  B + CC  < 0.

N 
g 
g
x(k + 1) = βs (k)ϑi (ϕ(k))ϑ j (ϕ(k))
s=1 i=1 j=1
3 Main results
[Ai j,s x(k) + Bi,s x(k − τs (k))]. (8)
In this section, the mode-dependent control scheme is
Definition 2.1 [23] System (8) under the switching designed such that CLFSS (8) is MSS.
law is called mean square stable (MSS) if there exists
a scalar c > 0 such that Theorem 3.1 Given the positive scalars τms and τ Ms ,
∞  CLFSS (8) is called MSS if there exist matrices P > 0,

E  x(k) 2 ≤ c sup E  ς (i) 2 . Q ls > 0 (l = 1, 2, 3), Rms > 0 (m = 1, 2) and matri-
k=0 −τ Ms ≤i≤0 ces Ss with proper dimensions such that the following
conditions hold for all s = 1, 2, . . . , N

123
 B. Wang et al.

N 
lii ii1 φ2s φ2s
< 0, (l = 1, 2) (9) − (1 − μ)τ Ms s ,
∗ ii2 φ3s φ3s
  s=1
li j i1j l ji  1ji (1)
 
+ < 0, (l = 1, 2, i < j) i,s = βs P Ai j,s e1 + βs P Bi,s e1+N +s ,
∗ i2j ∗  2ji  
(2)
i,l = βl (Ai j,l − I )e1 + βl Bi,l e1+N +l ,
(10)
e1 − e1+s
where φ1s = ,
N e1 + e1+s − 2e1+3N +s
ϒ1 τ Ms φ  S
1i j = s=1 (1 − μ) 2s s ,
 τms +1
∗ τ
−(1 − μ) Ms  ,τ >1
ρ(τms ) = τms −1 ms
N τ Ms φ  S
1, τms = 1
ϒ2 s=1 (1 − μ) 3s s
2i j = τ , e1+s − e1+N +s
∗ −(1 − μ) Ms  φ2s = ,
 e1+s + e1+n+s − 2e1+4N +s
1i
1 1 1
i1j = j 2i j 3i j , e1+N +s − e1+2N +s
0 0 0 φ3s = ,
⎡ ⎤ e1+N +s + e1+2N +s − 2e1+5N +s
1i
2 0 0
⎢
j
⎥ diag{R2s , 3R2s } Ss
i2j = ⎣ ∗ 2i 2 0 ⎦, s = .
j ∗ diag{R2s , 3R2s }
∗ ∗ 3i j
2
Proof We introduce the following Lyapunov–Krasovs-

N
 kii functional:
ϒ1 = ϒ + φ2s (1 − μ)τ Ms diag{R2s , 3R2s }φ2s ,
s=1 
4


N V (k, xk , rk ) = Vr (k, xk , rk ), (11)

ϒ2 = ϒ + φ3s (1 − μ)τ Ms diag{R2s , 3R2s }φ3s , r =1
s=1 where

N
(1) V1 (k, xk , rk ) = x  (k)P x(k), (12)
1i
1
j = (i,s ) ,
s=1

N 
k−1
 N  V2 (k, xk , rk ) =

N  (2) s=1 m=k−τms
2i
1
j = τms i,l R1s ,
s=1 l=1 (1 − μ)k−m−1 x  (m)Q 1s x(m)
 N  
N 
k−τ ms −1

N 
3i
(2) + (1 − μ)k−m−1 x  (m)Q 2s x(m)
j = τm2s i,l R2s ,
1

s=1 l=1 s=1 m=k−τ Ms

1i
2
= diag{−P, −P, . . . , −P }, 
N 
k−1
j    + (1 − μ)k−m−1 x  (m)Q 3s x(m),
N
s=1 m=k−τs (k)
2i
2
j = diag{−R , . . . , −R11 , −R12 , . . . , −R12 , . . . , −R1N }, (13)
 11     
N N

N −τ
 ms 
k−1
3i
2
j = diag{−R , . . . , −R21 , −R22 , . . . , −R22 , . . . , −R2N }, V3 (k, xk , rk ) =
 21      s=1 n=−τ Ms +1 m=k+n
N N
= diag{R21 , 3R21 , R22 , 3R22 , . . . , R2N , 3R2N }, (1 − μ)k−m−1 x  (m)Q 3s x(m), (14)

N 
N −1
 
k−1
 
ϒ = −(1 − μ)e1 Pe1 + e1 (Q 1s + Q 3s )e1 V4 (k, xk , rk ) = τms
s=1 s=1 n=−τms m=k+n


N (1 − μ)k−m−1 η (m)R1s η(m)

+ e1+s (1 − μ)τms (Q 2s − Q 1s )e1+s N −τ
ms −1 
 k−1
s=1 +(τ Ms − τms )

N

s=1 n=−τ Ms m=k+n
− e1+2N +s (1 − μ)τ Ms Q 2s e1+2N +s (1 − μ)k−m−1 η (m)R2s η(m), (15)
s=1

N where η(k) = x(k + 1) − x(k).

+ e1 (τ Ms − τms )Q 3s e1 Denote the forward difference of Vi (k, xk , rk ) (i =
s=1 1, 2, . . . , 4) as Vi (k, xk , rk ) = Vi (xk+1 , rk+1 ) −

N

Vi (k, xk , rk ) and define the following notations:
+ φ1s (1 − μ)τms diag{R1s , ρ(τms )3R1s }φ1s
s=1
ξ(k) = [x  (k), x  (k − τm1 ), x  (k − τm2 ), . . . ,

123
Fuzzy model-based nonfragile control

x  (k − τm N ), 
N ms
k−τ
+ (1 − μ)k−m x  (m)Q 3s x(m), (17)
x  (k − τm (k)), x  (k − τ M (k)), . . . , s=1 m=k−τ Ms +1

x  (k − τ N (k)),
E{V3 (k, xk , rk ) + μV3 (k, xk , rk )}
⎧
x  (k − τ M1 ), x  (k − τ M2 ), . . . , ⎨ N ms
k−τ

 k
 
x  (k − τ M N ), ϕ11 (k), ϕ12 (k), . . . , =E (1 − μ)k−m x  (m)Q 3s x(m)
⎩
 s=1 n=k−τ M +1 m=k+n+1
ϕ1N (k), 
  
k−1
ϕ21 (k), ϕ22 (k), . . . , − (1 − μ)k−m x  (m)Q 3s x(m)
   
ϕ2N (k), ϕ31 (k), ϕ32 (k), . . . , ϕ3N (k)] ,
m=k+n
 

N



k
1 = E ξ (k) e1 (τ Ms − τms )Q 3s e1 ξ(k)
ϕ1s (k) = x(m), s=1
τms + 1
m=k−τms 
N ms
k−τ
− x  (m)(1 − μ)k−m Q 3s x(m),
ms
k−τ
1 s=1 m=k−τ Ms +1
ϕ2s (k) = x(m),
τs (k) − τms + 1 E{V4 (k, xk , rk ) + μV4 (k, xk , rk )} (18)
m=k−τs (k)  N
'

k−τs (k) =E η (k)(τms
2
R1s + (τ Ms − τms )2 R2s )η(k)
1
ϕ3s (k) = x(m). s=1
τ Ms − τk (s) + 1
m=k−τ Ms 
k−1
− τms η (m)(1 − μ)k−m R1s η(m)
The derivative of the Lyapunov function can be m=k−τms
⎞⎫
described as follows: 
k−τ ms −1 ⎬
− (τ Ms − τms ) η (m)(1 − μ)k−m R2s η(m)⎠
⎭
m=k−τ Ms
 N '
E{V1 (k, xk , rk ) + μV1 (k, xk , rk )} 
≤E η (k)(τms
2
R1s + (τ Ms − τms )2 R2s )η(k)
= E{x  (k + 1)P x(k + 1) − (1 − μ)x  (k)P x(k)} s=1
 N
 
k−1
=E βs (Ai j,s x(k) + Bi,s x(t − τs (k)) + Di,s ω(k)) − τms (1 − μ)τms η (m)R1s η(m)
s=1 m=k−τms
⎞⎫
× P(Ai j,s x(k) + Bi,s x(t − τs (k)) 
k−τ ms −1 ⎬
+ Di,s ω(k)) − (1 − μ)x  (k)P x(k)} − (τ Ms − τms )(1 − μ) τ Ms
η (m)R2s η(m)⎠

 N  ⎭
m=k−τ Ms
    N 
 −1
= ξ (k)E i,s P i,s − (1 − μ)e1 Pe1 ξ(k), (16)  
N
(2)
s=1 =E ξ  (k) (i,l ) (τms
2
R1s
E{V2 (k, xk , rk ) + μV2 (k, xk , rk )} s=1 l=1
 N (2)
 + (τ Ms − τms )2 R2s )i,l ξ(k)
=E (x  (k)(Q 1s + Q 3s )x(k) + x  (k − τms )

k−1
s=1
τms
−τms (1 − μ)τms η (m)R1s η(m)
(1 − μ) (Q 2s − Q 1s )x(k − τms ) m=k−τms
− x  (k − τ Ms )(1 − μ)τ Ms Q 2s x(k − τ Ms ) ⎞⎫

k−τ ms −1 ⎬
− x  (k − τs (k)) − (τ Ms − τms )(1 − μ)τ Ms η (m)R2s η(m)⎠ .
⎭
τs (k)
(1 − μ) Q 3s x(k − τs (k)) m=k−τ Ms
⎞⎫ (19)
s (k)+1
k−τ ⎬
+ (1 − μ) x (m)Q 3s x(m)⎠
k−m 
⎭ Employing Lemma 2.1, it is deduced from (19) that
m=k−τs (k+1)+1
 N
   
N 
k−1
≤ ξ  (k)E (e1 (Q 1s + Q 3s )e1 + e1+s −τms (1 − μ)τms η (m)R1s η(m)
s=1
s=1 m=k−τms
(1 − μ)τms (Q 2s − Q 1s )e1+s

− e1+2N +s (1 − μ)τ Ms Q 2s e1+2N +s 
N

 "# ≤ −ξ  (k) φ1s (1 − μ)τms
− e1+N +s (1 − μ)τ Ms Q 3s e1+N +s ξ(k) s=1

123
 B. Wang et al.

diag{R1s , ρ(τms )3R1s }φ1s ξ(k). (20) 

N
 τ Ms − τs (k)
+ φ2s (1 − μ)τ Ms
(τ Ms − τms )
On the other hand, setting Ss be any matrix with s=1
' +
approximate dimension and using Lemma 2.2, one has diag{R2s , 3R2s } − Ss diag{R2s , 3R2s }−1 Ss φ2s

ms −1

N
τs (k) − τms

N 
k−τ 
−(τ Ms − τms )(1 − μ) τ Ms 
η (m)R2s η(m) + φ3s (1 − μ)τ Ms
(τ Ms − τms )
s=1
s=1 m=k−τ Ms
' +
= −(τ Ms − τms )(1 − μ)τ Ms diag{R2s , 3R2s } − Ss diag{R2s , 3R2s }−1 Ss φ3s .
⎛ ⎞
N 
k−τ ms −1 s (k)−1
k−τ
⎝ + ⎠ η (m)R2s η(m) By using the Schur complement, inequalities (9) and
s=1 m=k−τs (k) m=k−τ Ms (10) are equivalent to
(τ Ms − τms )  N
≤ −ξ  (k)(1 − μ)τ Ms φ *1 s=1
ϒ

(1 − μ)τ Ms φ2s Ss
τs (k) − τms 2s *
ϒ (τms ) = < 0, (23)
∗ τ
−(1 − μ) Ms 
diag{R2s , 3R2s }φ2s ξ(k)
(τ Ms − τms ) 
−ξ  (k)(1 − μ)τ Ms φ and
τ Ms − τs (k) 3s N
* τ Ms φ  S
diag{R2s , 3R2s }φ3s ξ(k) · * (τ Ms ) = ϒ2
ϒ s=1 (1 − μ) 3s s < 0, (24)
 ∗ τ
−(1 − μ) Ms 
φ2s *(τs (k)) φ2s
≤ −ξ  (k) (1 − μ)τ Ms  ξ(k)
φ3s φ3s

where
φ2s φ2s
= −ξ  (k) (1 − μ)τ Ms (τs (k)) ξ(k)
φ3s φ3s

N
 τ Ms − τs (k) *1 = ϒ +
ϒ

i,s P −1 i,s
+ξ  (k)φ2s (1 − μ)τ Ms
(τ Ms − τms )
' + s=1
diag{R2s , 3R2s } − Ss diag{R2s , 3R2s }−1 Ss φ2s ξ(k) 
N 
N
(2) (2)
+ (i,l ) (τms
2
R1s + (τ Ms − τms )2 R2s )i,l
 τs (k) − τms
+ ξ  (k)φ3s (1 − μ)τ Ms s=1 l=1
(τ Ms − τms )
' + 
N

diag{R2s , 3R2s } − Ss diag{R2s , 3R2s }−1 Ss φ3s ξ(k) + φ2s (1 − μ)τ Ms diag{R2s , 3R2s }φ2s
s=1
(21)
and
where


τ Ms −τs (k)
diag{R2s , 3R2s } + (τ Ms −τms ) Z 1s Ss
*(τs (k)) =
 τ Ms −τs (k) .
∗ diag{R2s , 3R2s } + (τ Ms −τms ) Z 1s

It follows from (11)–(21), we have

E{V (k, xk , rk ) + μV (k, xk , rk )}

, # 
N
≤ E ξ  (k)(τs (k))ξ(k) (22) *2 = ϒ +
ϒ

i,s P −1 i,s
s=1
where 
N 
N
(2) (2)
+ (i,l ) (τms
2
R1s + (τ Ms − τms )2 R2s )i,l

N

s=1 l=1
(τs (k)) = ϒ + i,s P −1 i,s 
N

s=1 + φ3s (1 − μ)τ Ms diag{R2s , 3R2s }φ3s .
s=1

N 
N
(2) (2)
+ (i,l ) (τms
2
R1s + (τ Ms − τms )2 R2s )i,l With the aid of convex combination approach [36],
s=1 l=1 inequalities (23) and (24) are equivalent to

123
Fuzzy model-based nonfragile control

⎡ N τ Ms −τs (k) τ Ms φ  S
N τs (k)−τms τ Ms φ  S
⎤
* (τs (k))
ϒ s=1 (τ Ms −τms ) (1 − μ) s=1 (τ Ms −τms ) (1 − μ)
2s s 3s s
⎢ τ Ms −τs (k) τ Ms  ⎥
ϒ(τs (k)) = ⎣ ∗ − (τ Ms −τms )
(1 − μ) 0 ⎦<0 (25)
τs (k)−τms
∗ ∗ − (τ Ms −τms ) (1 − μ)τ Ms 

where Remark 2 The nonfragile mode-dependent controller

is derived with the aid of PDC and uncertainties. Fur-
thermore, the drift of the controller is fuzzy-basis-

N
* (τs (k)) = ϒ +
ϒ

i,s P −1 i,s dependent and devised as K j,s = E j,s F j,s (k)H j ,
s=1 which gives more flexible of nonfragile controller

N 
N design. Compared with [22,31,33,37], the nonfragile
(2) (2)
+ (i,l ) (τms
2 R + (τ
1s
2
Ms − τms ) R2s )i,l control developed in the paper has great practical sig-
s=1 l=1 nificance.
N
 τ Ms − τs (k)
+ φ2s (1 − μ)τ Ms diag{R2s , 3R2s }φ2s Remark 3 It is noted that matrices Ss in inequality
(τ Ms − τms )
s=1 (21) give more freedom in choosing feasibility results.
N Compared with [22], inequality (21) makes full use of
 τs (k) − τms
+ φ3s (1 − μ)τ Ms diag{R2s , 3R2s }φ3s . Lemma 2.2 and less conservative criterion is achieved
(τ Ms − τms )
s=1 in Theorem 3.1.
By employing Schur complement to (25), one has
Remark 4 To make the considered issue more efficient
E{ V (k, xk , rk ) + μV (k, xk , rk )} < 0. (26) sense, scalar μ is addressed in Theorem 3.1 and it can
be adjusted for the sake of engineering requirements.
* ) < 0, it is obvious that
Since λmax (ϒ
Theorem 3.2 Let positive scalars τms , τ Ms and εl
E{V (k, xk , rk ) + μV (k, xk , rk )}
(l = 1, 2, . . . , 9) are given, CLFSS (8) is called MSS
* )  x(k) 2 .
≤ λmax (ϒ (27) if there exist matrices P > 0, Q ls > 0 (l = 1, 2, 3),
Summing both sides of the above inequality from 1 R1s > 0 (m = 1, 2), K j,s and matrices Ss with proper
to N , with respect to k, yields dimensions, such that the following conditions hold for
all s = 1, 2, . . . , N
E{V (x N , r N )} ≤ (1 − μ) N E{V (x0 , r0 )}
Xlii Yii

N < 0, (l = 1, 2), (31)
*)
+ λmax (ϒ  x(k) 2
(28) ∗ Z
⎡   ⎤
k=1 Xli j + Xl ji M ji Ni j
⎣ ∗ G1 0 ⎦ < 0, (l = 1, 2), (32)
which implies that
∗ ∗ G2

N
*)
−λmax (ϒ  x(k) 2 ≤ (1 − μ) N E{V (x0 , r0 )} where
k=1

≤ E{V (x0 , r0 )}. (29) 1
li j  i j
∞ Xli j = 2 , (l = 1, 2),
When N → ∞, it can be seen that the series  ∗ i j
k=1
     
x(k) 2 is convergent. Therefore Yi j = [Xi j , ε1 Wi j , Xi j , ε4 Ui j , Xi j , ε7 Vi j ] ,
Z = diag{−ε1 I, −ε1 I, −ε4 I, −ε4 I, −ε7 I, −ε7 I },
 x(k) 2 → 0, (30)      
Mi j = [Xi j , ε2 Wi j , Xi j , ε5 Ui j , Xi j , ε8 Vi j ] ,
which the proof is completed.      
Ni j = [Xi j , ε3 Wi j , Xi j , ε6 Ui j , Xi j , ε9 Vi j ] ,

Remark 1 Note that the nonlinear switched systems are G1 = diag{−ε2 I, −ε2 I, −ε5 I, −ε5 I, −ε8 I, −ε8 I },
approximated by T–S fuzzy approach; thus, the subsys- G2 = diag{−ε3 I, −ε3 I, −ε6 I, −ε6 I, −ε9 I, −ε9 I },

1 1 1
tem is called local linear switched system. 1  1i j  2i j  3i j
i j = ,
0 0 0

123
 B. Wang et al.

⎡ ⎤  
−E −1 0 0 /(1) =
 βs (Ai,s − Ci,s K j,s )e1 + βs Bi,s e1+N +s ,
= ⎣ ∗ −F 0 ⎦,
2 −1 i,s
i j  
∗ ∗ −G −1 /(2) =
 βl (Ai,l − Ci,l K j,l − I )e1 + βl Bi,l e1+N +l .
i,l
 N 

N
(1) 
N  (2)
 1i j = (i,s ) ,  2i j = τms i,l , The above inequalities can be rewritten as
s=1 s=1 l=1
 N  

N  (2)
1
lii  ii
 3i j = (τ Ms − τms ) i,l , 2
s=1 l=1 ∗  ii
(1)      
+Xii Fi (k)Wii + Wii Fi (k)Xii + Xii Fi (k)Uii
i,s = βs (Ai,s − Ci,s K j,s )e1 + βs Bi,s e1+N +s ,
    
(2)   +Uii Fi (k)Xii + Xii Fi (k)Vii + Vii Fi (k)Xii < 0, (l = 1, 2).
i,l = βl (Ai,l − Ci,l K j,l − I )e1 + βl Bi,l e1+N +l ,
−1
(34)
E = diag{P −1 , P −1 , . . . , P −1 },
−1 −1 −1 −1 −1 
F −1 = diag{R11 , . . . , R11 , R12 , . . . , R12 , . . . , R1N }, Owing to Fi (k)Fi (k) < I , the following inequali-
−1 −1 −1 −1 −1
G −1 = diag{R21 , . . . , R21 , R22 , . . . , R22 , . . . , R2N }, ties hold
Xi j = [H j 0 · · · 0],
-   . 
      1
Yi j = β1 E j,1 Ci,1 , β2 E j,2 Ci,2 , . . . , β N E j,N Ci,N , lii  ii
2
Wi j = [0, . . . , 0, Yi j , 0, . . . , 0], ∗  ii
  
Ui j = [0, . . . , 0, Z1i j , 0, . . . , 0], +ε1−1 Xii Xii + ε1 Wii Wii + ε4−1 Xii Xii
Vi j = [0, . . . , 0, Z2i j , 0, . . . , 0],   
  +ε4 Uii Uii + ε7−1 Xii Xii + ε7 Wii Wii < 0, (l = 1, 2).
Z1i j = τm1 Yi j , τm2 Yi j , . . . , τm N Yi j ,
 (35)
Z2i j = (τ M1 − τm1 )Yi j , (τ M2 − τm2 )Yi j , . . . ,

(τ M N − τm N )Yi j . By Schur complement, it follows from (35) that (31)
is guaranteed. In the similar way, it is clearly observed
Proof Define E −1 = diag{P −1 , P −1 , . . . , P −1 }, that we can obtain (32).
−1 −1 −1 −1 −1
F −1 = diag{R11 , . . . , R11 , R12 , . . . , R12 , . . . , R1N },
−1 −1 −1 −1 −1 Remark 5 Notice that the inequalities in (31)–(32) are
G −1 = diag{R21 , . . . , R21 , R22 , . . . , R22 , . . . , R2N },
not strict linear matrix inequality due to the fact that
pre- and post-multiplying (11) with −1 −1
the matrices P −1 , R1s and R2s . Fortunately, above
diag{I, . . . , I , E −1 , F −1 , G −1 } inequalities can be solved with the aid of cone comple-
   mentarity linearization approach [38].
2+10N

it yields that
 4 Illustrative example
/1
lii  ii
2 < 0, (l = 1, 2) (33)
∗  ii In this section, one illustrative example is given to illus-
trate the effectiveness of proposed theoretical results.
where Consider the single link robot arm model as in [39]:
/1 /1 /1
/i1j = 1i j 2i j 3i j ,
 q̈(t) = −
Mk gL
sin(q(t)) −
Dk 1
q̇(t) + u(t)
0 0 0 Jk Jk Jk

N
(1)
where q(t) denotes the angle of the arm. u(t) stands
/1i j =
 / ) ,
( for the control input. Its mass Jk , inertia Mk and damp-
i,s
s=1 ing Dk are changing with respect to the angle q(t),
 N 

N  which makes the robot arm as a hybrid system. In
/2i j =
 τms /(2)
 ,
i,l real situation, time delays are inevitable in the sys-
s=1 l=1 tem due to feedback gains, network transmission and
 N 

N  external disturbance. By discretization with sampling
/3i j =
 (τ Ms − τms ) /(2)
i,l time T = 0.1s we get delayed discrete-time randomly
s=1 l=1
switched fuzzy system model as follows:

123
Fuzzy model-based nonfragile control

0.4 1.4
x (k)
1
1.2 x1 (k)
0.3 x (k)
2
1 x (k)
2
0.2
0.8
State response

State response
0.1 0.6
0 0.4
0.2
−0.1
0
−0.2
−0.2
−0.3 −0.4
−0.4 −0.6
0 20 40 60 80 100 0 20 40 60 80 100
Time (k) Time (k)

Fig. 1 The state response of CLFSS without nonfragile control Fig. 2 The state response of CLFSS with nonfragile control

A random switching sequence with known sojourn probabilities

4

3 
2
q(k + 1) = βs (k)h si (k)[Ai,s q(k) 3.5

s=1 i=1 3

+ Bi,s q(k − τs (k)) + Ci,s u(k)]

Subsystem
2.5

2
and the parameter values of above system are borrowed 1.5
from [23] as follows: 1

0.5
1 0.1 1 0.1
A1,1 = , A2,1 = ,
−0.49 0.8 0.42 0.8 0
0 20 40 60 80 100
Time (k)
1 0.1
A1,2 = ,
−0.43 0.96 Fig. 3 The switching signal of CLFSS
1 0.1 1 0.1
A2,2 = , A1,3 = ,
−0.25 0.96 −0.25 0.96 5 Conclusions
1 0.1
A2,3 = ,
0.01 0.96 The problem of nonfragile control for a class of fuzzy
0.1 0 switched systems with mode-dependent time-varying
Bi,s = , delays has been fully addressed in this paper. With the
0 −0.2
aid of novelty Lyapunov function and extended recip-
(i = 1, 2, s = 1, 2, 3)C1,1 = C1,2 = 0.1I,
rocally convex discrete-time inequality, new delay-
C2,1 = C2,2 = 0.02I, C3,1 = C3,2 = 0.01I. variation-dependent stability criteria are established.
In addition, a nonfragile fuzzy controller is developed
Assume that the delay parameters are τm1 = τm2 = via the parallel distributed compensation technique and
τm3 = 1, τ M1 = 4, τ M2 = 5, τ M3 = 3, and setting stochastic analysis method. Finally, numerical simu-
β1 = 0.4, β2 = 0.15, β3 = 0.45. With the obtained lation is provided to show the effectiveness of the
control gain matrices, simulations were performed with proposed results. In the future, the proposed methods
the help of MATLAB. Figures 1 and 2 demonstrate the can be extended to the following more general situa-
state response without/with nonfragile control. It can tions: aperiodically intermittent control, nonhomoge-
be seen from Fig. 2 that CLFSS (10) is stochastically neous switched systems, semi-Markovian jump sys-
stable based on the nonfragile controller. Furthermore, tems.
the switching signal of CLFSS (10) is shown in Fig. 3.
It is shown from the simulation that with the aid of
developed theoretical strategies, the proposed nonfrag-
ile control design is usefulness and applicability.

123
 B. Wang et al.

Compliance with ethical standards 15. Wang, Y., Xia, Y., Shen, H., Zhou, P.: SMC design for robust
stabilization of nonlinear Markovian jump singular systems.
Conflict of interest The authors declare that there is no conflict IEEE Trans. Autom. Control 63(1), 219–224 (2018)
of interests regarding the publication of this paper. 16. Su, X., Shi, P., Wu, L., Song, Y.D.: Fault detection filter-
ing for nonlinear switched stochastic systems. IEEE Trans.
Autom. Control 61(5), 1310–1315 (2015)
17. Cheng, J., Park, J.H., Zhang, L., Zhu, Y.: An asynchronous
References operation approach to event-triggered control for fuzzy
Markovian jump systems with general switching policies.
1. Shen, H., Wang, T., Chen, M., Lu, J.: Nonfragile mixed IEEE Trans. Fuzzy Syst. 26(1), 6–18 (2018)
H∞ /L1 − L∞ state estimation for repeated scalar nonlin- 18. Park, M.J., Kwon, O.M.: Stability and stabilization of
ear systems with Markov jumping parameters and redundant discrete-time T–S fuzzy systems with time-varying delay
channels. Nonlinear Dyn. (2018). https://doi.org/10.1007/ via Cauchy–Schwartz-based summation inequality. IEEE
s11071-017-3899-x Trans. Fuzzy Syst. 25(1), 128–140 (2017)
2. He, S., Xu, H.: Non-fragile finite-time filter design for time- 19. Park, M.J., Kwon, O.M., Choi, S.G.: Stability analysis of
delayed Markovian jumping systems via T–S fuzzy model discrete-time switched systems with time-varying delays via
approach. Nonlinear Dyn. 80(3), 1159–1171 (2015) a new summation inequality. Nonlinear Anal. Hybrid Syst.
3. Zhang, D., Shi, P., Zhang, W., Yu, L.: Energy-efficient dis- 23, 76–90 (2017)
tributed filtering in sensor networks: a unified switched 20. Kwon, O.M., Park, M.J., Lee, S.H., Park, J.H., Lee, S.M.:
system approach. IEEE Trans. Cybern. 47(7), 1618–1629 Stability and stabilization of T–S fuzzy systems with time-
(2017) varying delays via augmented Lyapunov–Krasovskii func-
4. Tian, E., Yue, D., Yang, T.: Analysis and synthesis of ran- tionals. Inf. Sci. 372, 1–15 (2016)
domly switched systems with known sojourn probabilities. 21. Meng, X., Wang, L., Zhang, T.: Global dynamics analysis
Inf. Sci. 277, 481–491 (2014) of a nonlinear impulsive stochastic chemostat system in a
5. Tian, E., Wong, W.K., Yue, D.: Robust H∞ control for polluted environment. J. Appl. Anal. Comput. 6(3), 865–
switched systems with input delays: a sojourn-probability- 875 (2016)
dependent method. Inf. Sci. 283, 22–35 (2014) 22. Wang, B., Cheng, J., Zhan, J.: A sojourn probability
6. He, S., Song, J., Liu, F.: Robust finite-time bounded con- approach to fuzzy-model-based reliable control for switched
troller design of time-delay conic nonlinear systems using systems with mode-dependent time-varying delays. Nonlin-
sliding mode control strategy. IEEE Trans. Syst. Man ear Anal. Hybrid Syst. 26, 239–253 (2017)
Cybern. Syst. (2017). https://doi.org/10.1109/TSMC.2017. 23. Banu, L.J., Balasubramaniam, P.: Robust stability and stabi-
2695483 lization analysis for discrete-time randomly switched fuzzy
7. Meng, X., Zhao, S., Feng, T., Zhang, T.: Dynamics of a systems with known sojourn probabilities. Nonlinear Anal-
novel nonlinear stochastic sis epidemic model with double ysis Hybrid Syst. 17, 128–143 (2015)
epidemic hypothesis. J. Math. Anal. Appl. 433(1), 227–242 24. Seuret, A., Gouaisbaut, F., Fridman, E.: Stability of discrete-
(2016) time systems with time-varying delays via a novel summa-
8. Wu, Z., Cui, M., Shi, P., Karimi, H.R.: Stability of stochas- tion inequality. IEEE Trans. Autom. Control 60(10), 2740–
tic nonlinear systems with state-dependent switching. IEEE 2745 (2015)
Trans. Autom. Control 58(8), 1904–1918 (2013) 25. Wang, B., Zhong, S., Dong, X.: On the novel chaotic secure
9. Li, F., Meng, X., Cui, Y.: Nonlinear stochastic analysis for communication scheme design. Commun. Nonlinear Sci.
a stochastic SIS epidemic model. J. Nonlinear Sci. Appl. Numer. Simul. 39, 108–117 (2016)
10(9), 5116–5124 (2017) 26. Zhang, T., Ma, W., Meng, X., Zhang, T.: Periodic solution of
10. Meng, X., Chen, L., Wang, X.: Some new results for a logis- a prey–predator model with nonlinear state feedback control.
tic almost periodic system with infinite delay and discrete Appl. Math. Comput. 266, 95–107 (2015)
delay. Nonlinear Anal. Real World Appl. 10(3), 1255–1264 27. Wu, L., Yang, X., Lam, H.K.: Dissipativity analysis and syn-
(2009) thesis for discrete-time T–S fuzzy stochastic systems with
11. Wang, B., Yan, J., Cheng, J., Zhong, S.: New criteria of time-varying delay. IEEE Trans. Fuzzy Syst. 22(2), 380–394
stability analysis for generalized neural networks subject (2014)
to time-varying delayed signals. Appl. Math. Comput. 314, 28. Xie, X., Yue, D., Zhang, H., Peng, C.: Control syn-
322–333 (2017) thesis of discrete-time T-S fuzzy systems: reducing the
12. Zhang, L., Zhuang, S., Shi, P., Zhu, Y.: Uniform tube conservatism whilst alleviating the computational bur-
based stabilization of switched linear systems with mode- den. IEEE Trans. Cybern. (2016). https://doi.org/10.1109/
dependent persistent dwell-time. IEEE Trans. Autom. Con- TCYB.2016.2582747
trol 60(11), 2994–2999 (2015) 29. Luo, J., Zhao, J.: A generalized multiple Lyapunov func-
13. Zhang, S., Meng, X., Zhang, T.: Dynamics analysis and tions method based non-fragile guaranteed cost control for
numerical simulations of a stochastic non-autonomous uncertain switched fuzzy systems. Appl. Math. Comput.
predator-prey system with impulsive effects. Nonlinear 237, 494–500 (2014)
Anal. Hybrid Syst. 26, 19–37 (2017) 30. Park, M.J., Kwon, O.M., Park, J.H., Lee, S.M., Cha, E.J.:
14. Wang, Y., Xia, Y., Li, H., Zhou, P.: A new integral slid- Synchronization of discrete-time complex dynamical net-
ing model design method for nonlinear stochastic systems. works with interval time-varying delays via non-fragile con-
Automatica 90C, 304–309 (2018)

123
Fuzzy model-based nonfragile control

troller with randomly occurring perturbation. J. Frankl. Inst. 36. Park, P., Ko, J.W.: Stability and robust stability for systems
351, 4850–4871 (2014) with a time-varying delay. Automatica 43(10), 1855–1858
31. Wu, Z.G., Park, J.H., Su, H., Chu, J.: Non-fragile synchro- (2007)
nization control for complex networks with missing data. 37. Sakthivel, R., Saravanakumar, T., Ma, Y., Anthoni, S.M.:
Int. J. Control 86, 555–566 (2013) Finite-time resilient reliable sampled-data control for fuzzy
32. Wu, Z.G., Dong, S., Shi, P., Su, H., Huang, T., Lu, R.: Fuzzy- systems with randomly occurring uncertainties. Fuzzy Sets
model-based nonfragile guaranteed cost control of nonlin- Syst. 329, 1–18 (2017)
ear Markov jump systems. IEEE Trans. Syst. Man Cybern. 38. Ghaoui, L.E., Oustry, F., AitRami, M.: A cone complemen-
(2017). https://doi.org/10.1109/TSMC.2017.2675943 tarity linearization algorithm for static output-feedback and
33. Sakthivel, R., Aravindh, D., Selvaraj, P., Vimal, S., Kumar, related problems. IEEE Trans. Auto. Control 42, 1171–1176
S., Anthoni, M.: Vibration control of structural systems via (1997)
robust non-fragile sampled-data control scheme. J. Frankl. 39. Zhang, L., Ning, Z., Shi, P.: Input-output approach to control
Inst. 3, 1265–1284 (2017) for fuzzy Markov jump systems with time-varying delays
34. Zhang, C., He, Y., Jiang, L., Wang, Q., Wu, M.: Stability and uncertain packet dropout rate. IEEE Trans. Cybern.
analysis of discrete-time neural networks with time-varying 45(11), 2449–2460 (2015)
delay via an extended reciprocally convex matrix inequal-
ity. IEEE Trans. Cybern. (2017). https://doi.org/10.1109/
TCYB.2017.2665683
35. Khargonekar, P.P., Petersen, I.R., Zhou, K.: Robust stabi-
lization of uncertain linear systems: quadratic stabilizability
and H∞ control theory. IEEE Trans. Autom. Control 35(3),
356–361(1990)

123