Article

Transactions of the Institute of

Measurement and Control
Design of a stabilizing controller for 2019, Vol. 41(1) 14–22
Ó The Author(s) 2019

discrete-time switched delay systems Article reuse guidelines:

sagepub.com/journals-permissions
DOI: 10.1177/0142331217748191
with affine parametric uncertainties journals.sagepub.com/home/tim

NA Baleghi and MH Shafiei

Abstract
This paper studies the stabilization problem of discrete-time switched systems in the presence of a time-varying delay and parametric uncertainties.
The main goal is to provide a state feedback controller to guarantee the stability of the closed-loop system with an evaluated average dwell time. In this
regard, an appropriate Lyapunov–Krasovskii functional is constructed and the sufficient conditions for stability of the closed-loop system are developed
in terms of feasibility testing of proposed linear matrix inequalities. These conditions only depend on the upper bounds of the time delay and uncertain
parameters. Additionally, a numerical example is provided to verify the theoretical results.

Keywords
Discrete-time switched systems, parametric uncertainties, stabilization, time delay.

Introduction Hou et al. (2012) designed an observer-based L2-LN control-

ler for a class of discrete-time switched systems with norm-
A switched system is composed of multiple dynamical subsys- bounded uncertainties, using the average dwell-time method.
tems and a switching law, which determines the active subsys- Furthermore, Xiang and Wang (2011) have shown that the
tem at any time instant. This signal may depend on time, average dwell-time method can be used to stabilize the
state or both of them. Stability analysis and controller design discrete-time switched system under asynchronous switching.
of switched systems have attained much attention in the liter- In the case of polytopic uncertainties, the problem of robust
ature due to their widespread applications (Liu et al., 2015; L2–LN filtering for linear discrete-time switched systems has
Zhao et al., 2017; Zong et al., 2016). The procedure of con- been investigated by Zhang et al. (2007). In addition, based
troller design for switched systems includes three basic forms on switched Lyapunov functions, state and output feedback
(Zhao et al., 2017): 1) controller design under arbitrary controllers have been designed to stabilize a time-delay dis-
switching signals; 2) controller design under switching signals crete-time switched system with L2 performance index
with some constraints; 3) design of stabilizing switching sig- (Mahmoud et al., 2009). Moreover, the problems of robust
nals. In the case of arbitrary switching signals, finding a com- delay-dependent L2 gain analysis and feedback control synth-
mon Lyapunov function for all subsystems increases the esis have been investigated for a class of linear discrete-time
conservatism of design method. In order to solve the case 1, switched systems with time-varying delays (Mahmoud and
several approaches such as multiple Lyapunov functions Xia, 2009).
approach and average dwell-time method have been proposed Another type of uncertainty is affine parametric type in
(Hespanha and Morse, 1999; Mahmoud, 2010). In particular, which the structure of the uncertainty is known and is some-
the average dwell-time method is an effective tool used to times called structured uncertainty (Baleghi and Shafiei, 2017;
design the controller under constrained switching signals Bagherzadeh et al., 2016; Mahmoud, 2002; Mohammadpour
(Xiang et al., 2012). and Binazadeh, 2017; Zhai et al., 2012). This type of uncer-
On the other hand, uncertainties and time delays are both tainty arises in dynamical systems, when the state-space equa-
sources of unfavourable performance and instability of sys- tions are written based on physical considerations. It is worth
tems in the real world (Binazadeh and Bahmani, 2017a, noting that parametric uncertainties can also be converted to
2017b; Yousefi and Binazadeh 2016). For time-delay switched
systems, two kinds of uncertainties have been considered
more in the literature: polytopic and norm-bounded uncer- Department of Electrical and Electronic Engineering, Shiraz University of
Technology, Shiraz, Iran
tainties. In Sun et al. (2006) and Zhang et al. (2008), for
norm-bounded uncertain switched systems and based on Corresponding author:
switched Lyapunov functions, some conditions have been MH Shafiei, Department of Electrical and Electronic Engineering, Shiraz
derived to design a stabilizing state feedback. However, their University of Technology, Modares Blvd., Shiraz, Iran.
results are only admissible for arbitrary switching signals. Email: [email protected]
Baleghi and Shafiei 15

polytopic type, though this increases the conservatism of the system is composed of m subsystems with parametric uncer-
analysis and/or synthesis. Moreover, in contrast to the poly- tainties, which are expressed as follows:
topic uncertainty, which leads to 2n linear matrix inequalities 
(LMIs) (because of 2n vertexes for n uncertain parameters), in xðk + 1Þ = Ai xðk Þ + Bi xðk  d ðk ÞÞ + Ci uðk Þi 2 M,
Si :   d + 1, . . . , 0
affine parametric type, there are only n LMIs, and this leads xk0 ðlÞ = xðk0 + lÞ, l =  d,
to a less computational burden. Although there are many ð2Þ
existing results on the stability analysis and controller design
for the switched systems with polytopic or norm-bounded where Ai , Bi and Ci , i = 1, . . . , m, are parametric uncertain
uncertainties, the problem of controller design for time-delay matrices, defined as
discrete-time switched systems with affine parametric uncer-
tainties has not been fully resolved. X
r

In this paper, a switched feedback controller is designed Ai = A0i + dpj Eji ð3Þ
j=1
for a class of discrete-time switched systems in the presence of
a time-varying delay and parametric uncertainties. The system X
r
may include stable or unstable subsystems and the main goal Bi = B0i + dpj Fji ð4Þ
is to make the closed-loop switched system robustly stable. In j=1
this paper affine parametric uncertainty is considered, which
X
r
provides more feasibility in computation and less conserva- Ci = Ci0 + dpj Gji ð5Þ
tism compared with other types of uncertainties, such as poly- j=1
topic or norm-bounded uncertainties (as is shown in the
simulation section). Moreover, appropriate Lyapunov– where A0i , B0i and Ci0 are constant nominal matrices,
 r is the
Krasovskii functional is constructed to establish a stabilizing number of uncertain parameters and dpj 2 ej , ej , j =
state feedback controller to guarantee the stability of the 1, . . . , r is the perturbation around the nominal value of the
closed-loop switched system for a calculated average dwell jth uncertain parameter in matrices A0i , B0i . Moreover, the
time. It is worth noting that the proposed conditions in this matrices Ci0 . Eji , Fji and Gji are the uncertainty structure
paper depend only on the upper bound of uncertain para- matrices with known real parameters, describing how Ai , Bi
meters and the admissible upper bound of time delay. Finally, and Ci depend on the uncertain parameter dpj . If an uncertain
the obtained results are compared with existing results in Sun parameter does not enter in the ith subsystem, then Eji , Fji or
et al. (2006) for a switched system with norm-bounded uncer- Gji is a zero matrix.
tainty via a numerical example. It can be seen that the calcu- The objective is to find a mode-dependent state feedback
lated admissible upper bound of the time-varying delay for controller, which guarantees the exponential stability of the
the proposed method is less conservative than the paper, uncertain time-delay switched system (1). However, before
which has considered norm-bounded uncertainties. continuing, the following definitions and lemmas should be
The organization of this paper is as follows. Some preli- introduced.
minaries and useful lemmas are described in the next section.
Then we deal with the main results of this paper, which Definition 1 (Song et al., 2008; Zhang and Yu, 2009). A
includes a theorem on stabilization of a switched linear sys- switching signal sðtÞ is said to have an average dwell time Ta ,
tem subject to parametric uncertainties. A numerical example if for any k  k0 , the inequality Ns  N0 + kk Ta holds N0  0
0

is presented to investigate the main results of the paper, and and Ta . 0, where Ns denotes the total number of switching
finally, the paper is concluded. of sðt Þ during the interval ½k0 , k  and N0 is the chatter bound.
In this paper, without loss of generality, N0 = 0 is chosen.
Definition 2 The switched system (1) is robustly stabilized by
Preliminaries and problem formulation the control law uðk Þ under the average dwell time Ta , if for
Consider the following uncertain discrete-time switched sys- any initial conditions there exist an average dwell time Ta ,
tem with a time-varying delay in the state variable: such that its solution satisfies

xðk + 1Þ = Asðk Þ xðk Þ + Bsðk Þ xðk  d ðk ÞÞ + Csðk Þ uðk Þ kxðk Þk  clðkk0 Þ kfkL , 8k  k0 ð6Þ
S s ðk Þ :  k0  d + 1, . . . , k0
xðlÞ = fðlÞ, l = k0  d,
where kfkL = sup kfðlÞk, c . 0 is the decay coefficient
ð1Þ k0 d  l  k0
and l . 1 is the decay rate.
where xðk Þ 2 Rnx is the state vector, uðk Þ 2 Rnu is the control
input, sðk Þ : Z ! M = f1, . . . , mg is the switching signal, m is
Lemma 1 (Meyer, 2000). For any positive definite matrix P,
the number of subsystems of the switched system and Z is the
the following statements are equivalent:
set of positive integers. fðlÞ is a vector-valued initial function,
k0 is the initial time step and d ðk Þ is a bounded time-varying  xT Px . 0 for every nonzero x 2 Rn .
 Also, s = i, means that the
delay, which satisfies 0  d ðk Þ  d.  All eigenvalues of matrix P are positive.
ith subsystem is activated. It is assumed that the switched  P=BTB for some non-singular B.
16 Transactions of the Institute of Measurement and Control 41(1)

 0 T  0 T
Lemma 2. For a scalar e . 0, any matrices Xi 2 Rn 3 m ,  i,  
where j1 =  Xi + dT j2 = m1 (Xi A + Z1iT C ),
i = 1, . . . s, and any positive definite matrix P 2 Rn 3 n , the i i
 0 T  0
 T  i T  i
 T
following inequalities hold: j3 =m2 (Xi A  +Z T C  ), j4 =m3 Xi E  +Z T G  ,
i 1i i 1 1i 1
 T  i T  0 T
., j3+j =m3 Xi E  i +Z T G  , j4+j =m1 (Xi B  +
j 1i j i
X1T PX2 + X2T PX1  e X1T PX1 + e1 X2T PX2 ð7Þ   T
   T   T

Z2iT C 0 ), j5+j =m3 Xi F  i +Z T G i , ., j4+2j =m3
i 1 2i 1
   T  T  
ðX1 + X2 ÞT PðX1 + X2 Þ  ð1 + eÞX1T PX1 + 1 + e1 X2T PX2 Xi F ji +Z2iT G i , m1 = ð1+eÞ, m2 = ð1+eÞ 1ld ,

j
ð8Þ
 0 =lA0 ,
m3 =2rð1+e1 Þ, A i i
 0i =ld +1 B0i ,
B  0 =ld +1 C 0 ,
C i i
0 i
ðX1 + . . . + Xs ÞT PðX1 + . . . + Xs Þ  =lC 0 , E
C i
 i =lej Ei , F
i j
 i =ld +1 ej F i , G
j j
 =lej Gi and
j j j
  ð9Þ
 s X1T PX1 + . . . + XsT PXs  i =ld +1 ej Gi .
G j j
Then, the mode-dependent state feedback controller
uðk Þ = K1sðkÞ xðk Þ + K2sðkÞ xðk  d ðk ÞÞ robustly stabilizes the
Proof. Define H :¼ e X1T PX1 + e1 X2T PX2 X1T PX2  X2T PX1 :
According to Lemma 1, consider P = BT B. Let v1 :¼ e0:5 BX1 v uncertain switched system (1) with average dwell time Ta satis-
and v2 :¼ e0:5 BX2 v for any vector v, which implies ln m
fying Ta . Ta = , where the controller gains are
2 ln l
K1i = Z1i Xi and K2i = Z2i Xi 1 .
1
vT Hv = vT1 v1 + vT2 v2  vT1 v2  vT2 v1 = ðv1  v2 ÞT ðv1  v2 Þ  0
Proof. Consider the state-feedback controller for ith subsys-
Therefore, we have H  0, which is equivalent to (7). The tem of system (2) with the following structure:
inequality (8) is a direct consequence of (7) and with e = 1,
the expression (9) is obtained. Therefore, the proof is uðk Þ = K1i xðk Þ + K2i xðk  d ðk ÞÞ, i 2 M ð12Þ
completed. j
where K1i and K2i are the gain matrices. The state-feedback
controller is assumed to be switched synchronously by the
switching signal sðk Þ in the switched system (1).
Main results Applying the controller (12), leads to the following closed-
In this section, our objective is to design a switched feedback loop system:
controller such that the closed-loop switched system (1) is xðk + 1Þ = ðAi + Ci K1i Þxðk Þ + ðBi + Ci K2i Þxðk  d ðk ÞÞ, i 2 M
robustly stable. It is considered that the switching signal acti-
ð13Þ
vates the subsystem i at the instant k.
In the following theorem, first a mode-dependent state Now, if the transformation xðk Þ = lðkk0 Þ jðk Þ is applied to
feedback controller is designed to achieve a given decay rate, (13), the following switched system will be obtained:
l, for the proposed Lyapunov functional. Then, the average  
jðk + 1Þ = ðA  i K1i Þjðk Þ + B
i + C ~ i K2i jðk  d ðk ÞÞ, i 2 M
~i + C
dwell-time method is used to derive the exponential stability
conditions. ð14Þ
Theorem 1. Consider the time-delay discrete-time  switched where A  i = lAi , B ~ i = ld ðk Þ + 1 Ci , C
~ i = ld ðk Þ + 1 Bi , C  i = lCi and
system (1) with uncertain parameters dpj 2 ej , ej . For a the initial condition is as
scalar e . 0, a given scalar l . 1 and any time delay

0  d ðk Þ  d, suppose that there exist matrices   d + 1, . . . , 0
jk0 ðlÞ = jðk0 + lÞ = ll xk0 ðlÞ, l =  d, ð15Þ
Xi . 0, Ti . 0, Zi , i = 1, . . . , m, with appropriate dimensions
such that the following matrix inequalities hold: Consider the following candidate Lyapunov functional for
2 3
system (14):
j1 0 j2 j3 j4    j3 + j 0  0 X
k1
6 0 Ti j4 + j 0 0  0 j5 + j  j4 + 2j 7 Wi ðk Þ = jT ðk ÞPi jðk Þ + jT ðsÞQi jðsÞ
6 7
6   Xi 0 0  0 0  0 7
6 7 s = kd ðk Þ
6    Xi 0  0 0  0 7 ð16Þ
6 7 X
1 X
k1
6     Xi  0 0  0 7
6 7 + T
j ðsÞQi jðsÞ
6 .. 7\0
6      . 0 0  0 7 u = d + 1 s = k + u
6 7
6       Xi 0  0 7
6 7
6        Xi  0 7 The forward difference of the Lyapunov functional,
6 7
6 .. 7
4         . 0 5 DWi ðk Þ = Wi ðk + 1Þ  Wi ðk Þ, along the trajectories of system
         Xi (14) is given by
ð10Þ
DWi ðk Þ  jT ðk + 1ÞPi jðk + 1Þ  jT ðk ÞPi jðk Þ
ð17Þ
Xb  mXa , Ta  mTb 8a, b 2 M ð11Þ  T ðk ÞQi jðk Þ  jT ðk  d ðk ÞÞQi jðk  d ðk ÞÞ
+ dj
Baleghi and Shafiei 17

Using (3)–(5) and (14), we have

! ! !T
X
r
0 X
r X
r X
r
DWi ðk Þ  0 + l
A dpj Eji  +l
+ (C dpj Gji )K1i jðk Þ + ~0
B +l d ðk Þ + 1
dpj Fji ~ 0 + ld ðk Þ + 1
+ (C dpj Gji )K2i j ðk  d ðk ÞÞ
i i i i
j=1 j=1 j=1 j=1
! ! !
X
r
0 X
r X
r X
r
Pi 0 + l
A  +l
dpj Eji + (C ~ 0 + ld ðk Þ + 1
dpj Gji )K1i jðk Þ + B dp F
j j
i
+ (C~ 0
+ l d ðk Þ + 1
dp G
j j
i
)K2i j ð k  d ð k Þ Þ
i i i i
j=1 j=1 j=1 j=1

T  ðk ÞQi jðk Þ  j ðk  d ðk ÞÞQi jðk  d ðk ÞÞ

 j ðk ÞPi jðk Þ + dj T T

ð18Þ
0
 0 = lA0 , B
where A ~ 0 = ld ðk Þ + 1 C 0 and C
~ 0 = l d ðk Þ + 1 B 0 , C  = lC 0 . Using Lemma 2, we have
i i i i i i i i

 0
  0  T
DWi ðk Þ  ð1 + eÞ A 0 + C
 K1i jðk Þ + B ~ 0 K2i jðk  d ðk ÞÞ
~i + C
i i i

 0
  0    
Pi A0 + C
 K1i jðk Þ + B ~ 0 K2i jðk  d ðk ÞÞ + 1 + e1
~i + C
i i i

! ! !T
X
r   X
r  
i i d ðk Þ + 1 i i
l dpj Ej + Gj K1i jðk Þ + l dpj Fj + Gj K2i jðk  d ðk ÞÞ ð19Þ
j=1 j=1

! ! !
X
r   X
r  
d ðk Þ + 1
Pi l dpj Eji + Gji K1i jðk Þ + l i i
dpj Fj + Gj K2i jðk  d ðk ÞÞ
j=1 j=1

 T ðk ÞQi jðk Þ  jT ðk  d ðk ÞÞQi jðk  d ðk ÞÞ

 jT ðk ÞPi jðk Þ + dj
 
Now, using (9) and considering the time-varying delay d ðk Þ satisfying d ðk Þ  d and the intervals dpj 2 ej , ej , we have

 0
  0  T  0
  0  
DWi ðk Þ  ð1 + eÞ A 0 + C
 K1i jðk Þ + B
~ +C~ 0 K2i jðk  d ðk ÞÞ Pi A 0 + C
 K1i jðk Þ + B ~ 0 K2i jðk  d ðk ÞÞ
~ +C
i i i i i i i i

( ! ! )
  X
r   X
r
e2j (Eji + Gji K1i ) Pi Eji + Gji K1i jðk Þ + jðk  d ðk ÞÞT 2rl2ðd + 1Þ
T  T
+ 1 + e1 jðk ÞT 2rl2 e2j (Fji + Gji K2i ) Pi (Fji + Gji K2i ) jðk  d ðk ÞÞ
j=1 j=1

 T ðk ÞQi jðk Þ  jT ðk  d ðk ÞÞQi jðk  d ðk ÞÞ

 jT ðk ÞPi jðk Þ + dj
ð20Þ

Then we have
0 2 T 3 1
0 2 3
0 + C
 K1i  i
B T6
A i i 7 h 0
  0  i C Pi + dQ 0
DWi ðk Þ  ð1 + eÞB
@ h ðk Þ 6
4
7Pi A
5
0 + C
i
 K1i
i
~ +C
B i
~ 0 K2i hðk ÞC + hðk ÞT 4
i A
5hðk Þ
 0  0 Qi
~ +C
B ~ 0 K2i T
i i

2 r 
P i
T  i
 3 ð21Þ
2r E  K1i Pi E
 ji + G  K1i
 ji + G 0
6 j j 7
  6 j=1 7
+ 1 + e1 hðk ÞT 6
6
7hðk Þ
4 r 
P T  7
5
0 2r F  i K2i
i + G Pi F  i K2i
i + G
j j j j
j=1

i  
ji = ld + 1 ej Fji , G
 ji = lej Eji , F
where E  = lej Gi , G i = ld + 1 ej Gi and hðk Þ = jT ðk Þ, jT ðk  d ðk ÞÞ T .
j j j j

~ 0 = ld ðk Þd B
0, C 
~ 0 = ld ðk Þd C 0 , d ðk Þ  d and l . 1 that
It follows from the relations B i i i i
18 Transactions of the Institute of Measurement and Control 41(1)
2 T 3
0

A 0
+ C K h   i
6 i i 1i 7 0
hðk ÞT 4 5Pi A 0 + C K1i ~0 + C
B ~ 0 K2i hðk Þ
 0  i i i i
~ +C
B ~ 0 K2i T
i i
 0
T  0
  0
T
= jT ðk Þ A 0 + C  K1i Pi A 0 + C K1i jðk Þ + 2jT ðk Þ A 0 + C K1i
i i i i i i
 0   0 
~ +C
Pi B ~ 0 K2i jðk  d ðk ÞÞ + jT ðk  d ðk ÞÞ B ~ +C ~ 0 K2i T
i i i i
 0   0 
~i + C
Pi B ~ 0 K2i jðk  d ðk ÞÞ + jT ðk  d ðk ÞÞ B ~i + C~ 0 K2i T
i i
 0
T  0
  0
T  

= ld ðk Þd jT ðk Þ A 0 + C
i
 K1i Pi A
i
0 + C
i
 K1i jðk Þ + 2jT ðk Þ A
i
0 + C
i
 K1i Pi B
i
 0 K2i jðk  d ðk ÞÞ
0 + C
i i
 0   0  ð22Þ
+ jT ðk  d ðk ÞÞ B  +C  0 K2i T Pi B +C  0 K2i jðk  d ðk ÞÞ
i i i i
   0
T  0


+ 1  ld ðk Þd jT ðk Þ A 0 + C
i
 K1i Pi A
i
0 + C
i
 K1i jðk Þ
i
   0   0 
+ l2ðd ðk Þd Þ  ld ðk Þd jT ðk  d ðk ÞÞ B
 
 +C  0 K2i T Pi B  +C  0 K2i jðk  d ðk ÞÞ
i i i i
2 0 T 3
0  h   i
T 4 Ai + C i K1i 5Pi A 0 + C
0
 K1i  0 K2i hðk Þ
 0i + C
 hðk Þ B
 0  i i i
B +C  0 K2i T
i i
   0 T  0 
d T 0 + C  K1i Pi A 0 + C  K1i jðk Þ
+ 1l j ðk Þ A i i i i

Now, from (21) and (22), we have

DWi ðk Þ  ð1 + eÞ
2 0 T 3
0  h   i
T 4 Ai + C i K1i 5Pi A 0 + C
0
 K1i  0i + C  0 K2i hðk Þ
ðhðk Þ B
 0  i i i
 +C
B  0 K2i T
2 i i
 T   3
0 0
d 0 + C K1i Pi A 0 + C
 K1i  Pi + dQ  i
T4 1  l A 0 5hðk ÞÞ
+ hðk Þ i i i i ð23Þ
0 Qi
2 r  T   3
P i i
2r E ji + G K1i Pi E  ji + G  K1i 0
  6 j=1 j j 7
6 7
+ 1 + e1 hðk ÞT 6     7hðk Þ
4 P
 K2i 5
r T
0 2r F +G
i  K2i Pi F
i  +G
i i
j j j j
j=1

If we could find K1i and K2i such that

02  0
T 3 1
0 + C
A  K1i h 0
  i
ð1 + eÞ@4 i i 5Pi A0 + C
 K1i 0
B  0 K2i
+C A
  i i i i
B0 + C  0 K2i T
i i
2  T   3
0 0
d 0 + C  K1i Pi A0 + C  i
 K1i  Pi + dQ
4 1  l Ai i i i 0 5
+
0 Qi ð24Þ
2  T   3
P
r i i
2r E  K1i Pi E
i + G  K1i
i + G 0
 6
6 j=1
j j j j 7
7
+ 1 + e1 6 r  T  7\0
4 P
0 2r F 
j + Gj K2i Pi F
i i j + Gj K2i 5
i  i
j=1

Then DWi ðk Þ\0. Therefore, the forward difference of the Lyapunov functional in inequality (23) is negative, which implies that
Wi ðk Þ  Wi ðk0 Þ for any k  k0 .
Now, consider the following Lyapunov functional for the ith subsystems of (2):

X
k1 X
1 X
k1
Vi ðk Þ = xT ðk ÞPi xðk Þ + l2ðsk Þ xT ðsÞQi xðsÞ + l2ðsk Þ xT ðsÞQi xðsÞ ð25Þ
s = kd ðk Þ u = d + 1 s = k + u
Baleghi and Shafiei 19

where l . 1 is a given constant and Pi and Qi are the positive X

kt 1

definite weighting matrices of appropriate dimensions. Vsðkt Þ ðkt Þ = xT ðkt ÞPsðkt Þ xðkt Þ + l2ðskt Þ xT ðsÞQsðkt Þ xðsÞ
s = kt d ðk Þk(
Applying the transformation xðk Þ = lðkk0 Þ jðk Þ leads to:
X
1 X
k1
+ l2ðskt Þ xT ðsÞQsðkt Þ xðsÞ
Vi ðk Þ = l2ðkk0 Þ jT ðk ÞPi jðk Þ u = d + 1 s = k + u

X
k1  xT ðkt ÞmPsðkt 1Þ xðkt Þ
+ l2ðsk Þ l2ðsk0 Þ jT ðsÞQi jðsÞ X
kt 1
s = kd ðk Þ + l2ðskt Þ xT ðsÞmQsðkt 1Þ xðsÞ
s = kt d ðk Þ
X
1 X
k1
X
1 X
k1
2ðsk Þ 2ðsk0 Þ T
+ l l j ðsÞQi jðsÞ + l2ðskt Þ xT ðsÞmQsðkt 1Þ xðsÞ
u = d + 1 s = k + u
u = d + 1 s = k + u

= l2ðkk0 Þ jT ðk ÞPi jðk Þ ð26Þ = mVsðkt 1Þ ðkt Þ ð30Þ

X
k1 Using (29) and (30) and the fact that sðkt  1Þ = sðkt1 Þ, we
+ l2ðkk0 Þ jT ðsÞQi jðsÞ have
s = kd ðk Þ

X
1 X
k 1 Vsðk Þ ðk Þ  l2ðkkt Þ Vsðkt Þ ðkt Þ
+ l2ðkk0 Þ jT ðsÞQi jðsÞ  l2ðkkt Þ mVsðkt 1Þ ðkt Þ ð31Þ
u = d + 1 s = k + u
2ðkkt Þ 2ðkt kt1 Þ
l l mVsðkt1 Þ ðkt1 Þ
= l2ðkk0 Þ Wi ðk Þ
Iterating kt from kt1 to k0 , it follows that
According to the fact that Wi ðk0 Þ = Vi ðk0 Þ, we have
l2ðkk0 Þ Vi ðk Þ = Wi ðk Þ  Wi ðk0 Þ = Vi ðk0 Þ or Vsðk Þ ðk Þ  l2ðkk0 Þ mNs Vsðk0 Þ ðk0 Þ ð32Þ

Vi ðk Þ  l2ðkk0 Þ Vi ðk0 Þ: ð27Þ Applying the definition of the average dwell time Ta leads to
lnm
Therefore, along any trajectory of system (2), the function Vsðk Þ ðk Þ  l2ðkk0 Þ lNs lnl Vsðk0 Þ ðk0 Þ
Vi ðk Þ in (25) guarantees the given decay rate l. 2ðkk0 Þ Ns lnm

For a switching signal sðt Þ and any given integer k . 1, = l2ðkk0 Þ l 2ðkk0 Þ lnl
Vsðk0 Þ ðk0 Þ
ð33Þ
the switching instants of sðt Þ are considered k1 \ . . . \kt , and 2ðkk0 Þð2Tlnm + 1Þ
l a lnl Vsðk0 Þ ðk0 Þ
the series fxðk0 Þ; ði0 , k0 Þ, ði1 , k1 Þ, . . . , ðit , kt Þ, ðit + 1 , k Þg are the
= (lr )2ðkk0 Þ Vsðk0 Þ ðk0 Þ
switching sequence, which means that sðt Þ = ij or the ijth sub-
system is activated when kj \t\kj + 1 . Now, consider the fol-
Therefore, from the above inequality, it is obtained that
lowing Lyapunov functional for system (1):
b1 kxðk Þk2  Vsðk Þ ðk Þ  (lr )2ðkk0 Þ
X
k1
Vsðk Þ ðk Þ = xT ðk ÞPsðk Þ xðk Þ + l2ðsk Þ xT ðsÞQsðk Þ xðsÞ Vsðk0 Þ ðk0 Þ  (lr )2ðkk0 Þ b2 kfk2L
s = kd ðk Þ
which yields
X
1 X
k1
2ðsk Þ T
+ l x ðsÞQsðk Þ xðsÞ sffiffiffiffiffi
u = d + 1 s = k + u b2 r ðkk0 Þ
kxðk Þk  (l ) kfkL ð34Þ
ð28Þ b1

where Pi . 0, Qi . 0, i = 1, . . . , m are the solutions of (11) and where b1 = min lmin ðPi Þ and b2 = max lmax ðPi Þ +
i2M i2M
Oi \0. According to (27), we have
d max lmax ðQi Þ. Therefore, it is concluded from Definition 2
i2M
2ðkkt Þ that the closed-loop switched system (1) is robustly stable and
Vi ðk Þ  l Vi ðkt Þ ð29Þ
ensures the decay rate of lr .
Moreover, using the Schur complement, (24) is equivalent
Moreover, it is found from (11), (28) and Pi = Xi1 that
to
20 Transactions of the Institute of Measurement and Control 41(1)
2  T  T  T  T 3
0 0 i i
 0  0  i   K1i Pi
i + G
6 Pi + dQi 0 m1 Ai + C i K1i Pi m2 Ai + C i K1i Pi m3 E1 + G1 K1i Pi   m3 E j j 0  0 7
6  0   i   T 7
6  +C 0 K2i T Pi  i K2i T Pi
 +G  i K2i Pi 7
i +G
6 0 Qi m1 B i i 0 0   0 m3 F1 1    m3 F j j 7
6 7
6   Pi 0 0   0 0  0 7
6 7
6    Pi 0   0 0  0 7
6 7
6     Pi   0 0  0 7
6 7\0 ð35Þ
6 .. 7
6      . 0 0  0 7
6 7
6       Pi 0  0 7
6 7
6        Pi  0 7
6 7
6 .. 7
4         . 0 5
         Pi

 

where m1 = ð1 + eÞ, m2 = ð1 + eÞ 1  ld and
m3 = 2rð1 + e1 Þ. Applying the congruent transformation
diagfXi , Xi , Xi , Xi , Xi , . . . , Xi , Xi , . . . , Xi g, Xi = P1
i , we obtain
2  T  T  T  T 3
0 0 i i
 0  0 + C K1i  K1i
i + G  K1i
i + G
6 Xi + dTi 0 m1 Xi Ai + C i K1i m2 Xi A i i m3 Xi E 1 1   m3 Xi E j j 0   0 7
6  0   i   T 7
6  +C  0 K2i T  +G  i K2i T i +G i K2i 7
6 0 Ti m1 Xi B i i 0 0   0 m3 Xi F1 1   m3 Xi F j j 7
6 7
6   Xi 0 0   0 0   0 7
6 7
6    Xi 0   0 0   0 7
6 7
6     Xi   0 0   0 7
6 7\0 ð36Þ
6 .. 7
6      . 0 0   0 7
6 7
6       Xi 0   0 7
6 7
6        Xi   0 7
6 7
6 .. 7
4         . 0 5
         Xi

where Ti = XiT Qi Xi . With the change of variable of Z1i = K1i Xi where Ai , Bi and Ci are constant nominal matrices. The
and Z2i = K2i Xi , the inequality (10) is obtained and the proof matrices DAi , DBi and DCi , corresponding to the parameter
is completed. uncertainties, satisfy ½DAi DBi DCi  = Di F ðk Þ½Ei1 Ei2 Ei3 , where
Di and Eij , j = 1, 2, 3 are given constant matrices with appro-
Remark 1. It should be noted that because of some unknown priate dimensions. The matrix F ðk Þ is assumed to satisfy the
 the condition (10) in Theorem 1 is not a strict
scalars e and d, condition F ðk ÞT F ðk Þ  I. Now consider Example 2 in Sun
LMI. An efficient way to resolve this problem is to fix the et al. (2006). The system data are
unknown scalars and then solve a set of LMIs for the chosen
values of these parameters. For example, the value of para- 1 0 0:2 0:5 1 0:3
meters e and d (the admissible upper bound of time delay) A1 = , B1 = , C1 = ,
0 1:01 0 0:1 1 0:6
may be changed until the LMIs would not be feasible. If the
admissible upper bound of time delay is known, we can 0:96 0:1 0:3 0:1 1:2 0:5
A2 = , B2 = , C2 = ,
search for the maximal bound of uncertain parameters. For 0 1:03 0 0:2 2 1:6
this purpose, if the uncertain parameters are considered Eij = 0:01I, Di = I, i = 1, 2, j = 1, 2, 3
dpj 2 ½e, e, the parameter e may be increased until the
ð38Þ
LMIs (10) and (11) would not be feasible.
Remark 2. The designed controller (12) depends on time The subsystems 1 and 2 are both unstable. In order to be able
delay. If this controller is difficult to realize in practical to compare the results of the proposed method with those of
applications, one can set K2i = 0 in (12) and obtain a sim- Sun et al. (2006), the time-delay discrete-time switched system
pler state feedback controller with more conservative on (1) with two subsystems is considered. If we consider the
time delay. In this case, the condition (10) has a lower matrices of subsystems are
dimension.
1 + dp1 0 0:2 + dp1 0:5
Numerical example A1 = , B1 = ,
0 1:01 + dp1 0 0:1 + dp1
In this section, the results of the proposed method are com- 1 + dp1 0:3
C1 = ,
pared with the method based on norm-bounded uncertainties 1 0:6 + dp1
(Sun et al., 2006) using a computer simulation. Consider the 0:96 + dp1 0:1 0:3 + dp1 0:1
A2 = , B2 = ,
following switched system with norm-bounded uncertainty: 0 1:03 + dp1 0 0:2 + dp1
  1:2 + dp1 0:5
xðk + 1Þ = Asðk Þ + DAsðk Þ ðk Þ xðk Þ C2 =
  2 1:6 + dp1
+ Bsðk Þ + DBsðk Þ ðk Þ xðk  d ðk ÞÞ ð37Þ ð39Þ
 
+ Csðk Þ + DCsðk Þ ðk Þ uðk Þ
Baleghi and Shafiei 21

where uncertain parameters defined as dp1 2 ½0:01, 0:01, bound of time delay in the proposed method is bigger than
then the switched system (1) with the matrices (39) is equiva- result of Sun et al. (2006).
lent to system (37) with the matrices (38). According to uncer-
tain parameters, the nominal matrices are
Conclusion
1 0 0:2 0:5 1 0:3
A01 = , B01 = , C10 = , This paper studied the controller design for time-delay dis-
0 1:01 0 0:1 1 0:6 crete-time switched systems with parametric uncertainties.
0:96 0:1 0:3 0:1 1:2 0:5 For this purpose, a mode-dependent state feedback controller
A02 = , B02 = , C20 =
0 1:03 0 0:2 2 1:6 was designed, and by choosing the appropriate Lyapunov–
ð40Þ Krasovskii functional, the stability conditions were derived in
terms of LMIs. The proposed controller leads to a less com-
Moreover, uncertainty structure matrices are E11 = E12 = putational burden, which results in the evaluating of admissi-
F11 = F12 = I, where I denotes the identity matrix. Now, ble upper bounds of time delay with fewer LMI equations.
Theorem 1 can be used to investigate a switched feedback Furthermore, computer simulations were performed and the
controller such that the closed-loop system is stable. For the efficiency of the proposed controller was shown.
nominal matrices (40), when l = 1:1, e = 0:5 and m = 1, the
feasible solutions of (10) and (11) may be obtained using the Declaration of Conflicting Interests
LMI toolbox of MATLAB for the admissible upper bound of
time delay d = 7. Moreover, The author(s) declared no potential conflicts of interest with
respect to the research, authorship, and/or publication of this
0:295 0:644 0:509 1:022 article.
K11 = , K21 = ,
0:174 1:892 0:917 1:822
0:322 0:322 0:420 0:369 Funding
K12 = , K22 =
0:307 0:701 0:489 0:595 The author(s) received no financial support for the research,
ð41Þ authorship, and/or publication of this article.

Therefore, the system (1) with the parameters (39) and allow- ORCID iD
able upper bound of time delay d = 7 can be robustly stabi-
lized for uncertain parameters belonging to the mentioned M.H. Shafiei http://orcid.org/0000-0002-3989-4676
intervals, whereas in Sun et al. (2006) the admissible upper
bound of time- delay is d = 3, which shows the conservatism References
of its method with respect to the proposed method.
Bagherzadeh MA, Ghaisari J and Askari J (2016) Robust exponential
In Figure 1, the simulation results of the proposed method
stability and stabilisation of parametric uncertain switched linear
are compared with the existing result in Sun et al. (2006) for
systems under arbitrary switching. IET Control Theory & Applica-
d = 1. Compared with the method in Sun et al. (2006), the tions 10: 381–390.
proposed method has led to a better transient response and Baleghi NA and Shafiei MH (2017) Stability analysis for discrete-time
more rapidly convergence. Furthermore, the admissible upper switched systems with uncertain time delay and affine parametric
uncertainties. Transactions of the Institute of Measurement and
Control DOI: 10.1177/0142331216678061.
Binazadeh T and Bahmani M (2017a) robust time-varying output
tracking control in the presence of actuator saturation. Transac-
tions of the Institute of Measurement and Control DOI: 10.1177/
0142331216650022.
Binazadeh T and Bahmani M (2017b) Design of robust controller for
a class of uncertain discrete-time systems subject to actuator
saturation. IEEE Transactions on Automatic Control 62(3):
1505–1510.
Hespanha JP and Morse AS (1999) Stability of switched systems with
average dwell-time. In: Proceedings of the 38th IEEE Conference
on Decision and Control, pp. 2655–60.
Hou L, Zong G and Wu Y (2012) Observer-based finite-time expo-
nential L2-LN control for discrete-time switched delay systems
with uncertainties. Transactions of the Institute of Measurement
and Control 35(3): 310–320.
Liu T, Wu B, Wang Y, et al. (2015) New stabilization results for
discrete-time positive switched systems with forward mode-
Figure 1. The trajectories of the close-loop system states for our dependent average dwell time. Transactions of the Institute of
result and existing result in Sun et al. (2006). Measurement and Control 39(2): 224–229.
22 Transactions of the Institute of Measurement and Control 41(1)

Mahmoud MS (2002) Discrete-time systems with linear parameter- switching. Transactions of the Institute of Measurement and Con-
varying: stability and HN-filtering. Journal of Mathematical Anal- trol 33(5): 591–609.
ysis and Applications 269(1): 369–381. Xiang Z, Liu S and Chen Q (2012) An improved exponential stability
Mahmoud MS and Xia Y (2009) Robust stability and stabilization of criterion for discrete-time switched non-linear systems with time-
a class of nonlinear switched discrete-time systems with time- varying delays. Transactions of the Institute of Measurement and
varying delays. Journal of Optimization Theory and Applications Control 35(3): 353–359.
143: 329–355. Yousefi M and Binazadeh T (2016) Delay-independent sliding mode
Mahmoud MS, Al-Sunni FM and Shi Y (2009) Switched discrete-time control of time-delay linear fractional order systems. Transactions
delay systems: delay-dependent analysis and synthesis. Circuits, of the Institute of Measurement and Control DOI: 10.1177/
Systems, and Signal Processing 28(5): 735–761. 0142331216678059.
Mahmoud MS (2010) Switched Time-Delay Systems: Stability and Zhai D, Zhang QL and Liu GY (2012) Robust stability analysis of lin-
Control. Berlin: Springer Science & Business Media. ear systems with parametric uncertainty. International Journal of
Mohammadpour S and Binazadeh T (2017) Observer-based synchro- Systems Science 43(9): 1683–1688.
nization of uncertain chaotic systems subject to input saturation. Zhang L, Shi P, Boukas EK, et al. (2007) Robust l2–lN filtering for
Transactions of the Institute of Measurement and Control DOI: switched linear discrete time-delay systems with polytopic uncer-
10.1177/0142331217705435. tainties. IET Control Theory and Applications 1(3): 722–730.
Meyer CD (2000) Matrix Analysis and Applied Linear Algebra. Vol. 2. Zhang L, Shi P and Basin M (2008) Robust stability and stabilisation
Philadelphia, PA: SIAM. of uncertain switched linear discrete time-delay systems. IET Con-
Song Y, Fan J, Fei M, et al. (2008) Robust HN control of discrete trol Theory and Applications 2(7): 606–614.
switched system with time delay. Applied Mathematics and Com- Zhang WA and Yu L (2009) Stability analysis for discrete-time
putation 205(1): 159–69. switched time-delay systems. Automatica 45(10): 2265–71.
Sun YG, Wang L and Xie G (2006) Delay-dependent robust stability Zhao X, Kao Y, Niu B, et al. (2017) Control Synthesis of Switched
and stabilization for discrete-time switched systems with mode- Systems. Berlin: Springer.
dependent time-varying delays. Applied Mathematics and Compu- Zong G, Wang Q and Yang Y (2016) Robustly resilient memory con-
tation 180(2): 428–35. trol for time-delay switched systems under asynchronous switch-
Xiang X and Wang R (2011) Robust stabilization of discrete-time ing. Transactions of the Institute of Measurement and Control
switched non-linear systems with time delay under asynchronous DOI: 10.1177/0142331216636187.