1398 IEEE TRANSACTIONS ON NEURAL NETWORKS, VOL. 21, NO. 9, SEPTEMBER 2010
The delay d(t) satisfies either (A1) or (A2) given as follows:
(A1) Constant time delay: 0 ≤ d(t) ≡ d;
(A2) Time-varying delay: 0 <d
1
≤ d(t) ≤ d
2
and
˙
d(t) ≤ τ
where d, d
1
, d
2
, and τ are known real constants.
For each possible value α(t)=i, i ∈ N, we will denote the
system matrices associated with mode i by
A(i)=A(α),A
d
(i)=A
d
(α),C(i)=C(α)
where A(i), A
d
(i), and C(i) are constant matrices.
Corresponding to the switching signal α,wehave
the switching sequence
{
(i
0
,t
0
), (i
1
,t
1
),... ,(i
k
,t
k
),... ,|
i
k
∈ N,k =0, 1,...
}
with t
0
= 0, which means that the i
k
th
subsystem is activated when t ∈
[
t
k
,t
k+1
)
.
Suppose that x
∗
[x
∗
1
,x
∗
2
,... ,x
∗
n
]
T
is an equilibrium of
system (3). As is usually done, we shift the equilibrium to the
origin by changing variables
ξ
i
(t) x
i
(t) −x
∗
i
f
i
(ξ
i
(t)) g
i
(ξ
i
(t)+x
∗
i
) − g
i
(x
∗
i
)
where i =1, 2,... ,n. Then the switched DNN (3) readily
transforms into
dξ(t)
dt
= −C(α)ξ(t)+A(α)f
(
ξ(t)
)
+ A
d
(α)f
(
ξ(t −d(t))
)
(5)
where
ξ
i
(t)
ξ
1
(t),ξ
2
(t), ... , ξ
n
(t)
T
f
i
(ξ
i
(t))
f
1
(ξ
1
(t)),f
2
(ξ
2
(t)), ... , f
n
(ξ
n
(t))
T
.
In addition, it is easily verified from (4) that f
i
(ξ
i
(t)) satisfies
f
i
(0) = 0 and ∀ξ
i
=0
0 ≤
f
i
(ξ
i
)
ξ
i
≤ k
i
, ∀i =1, 2,... ,n (6)
which can be rewritten as
f
i
(ξ
i
(·))
f
i
(ξ
i
(·)) − k
i
ξ
i
(·)
≤ 0, ∀i =1, 2,... ,n. (7)
Remark 1: For the existence of equilibrium point of system
(3), it is equivalent to prove that the states of system (5)
converge to the origin, just like the method applied in [17]
and [25]. On the contrary, from [3, Assum. 1, Lem. 1], it can
be obtained that there exists an equilibrium for each subsystem
of (5). Furthermore, if it is assumed that each subsystem has a
common equilibrium, such as the origin, then the equilibrium
point of the switched DNN (5) is the origin. So it is only
needed to prove that the system is asymptotically stable.
For the switching signal α(t), we revisit the average dwell
time property from the following definition.
Definition 1: For any T
2
>T
1
≥ 0, let N
α
(T
1
,T
2
) de-
note the number of switchings of α(t) over (T
1
,T
2
) [22]. If
N
α
(T
1
,T
2
) ≤ N
0
+(T
2
− T
1
)/T
a
holds for T
a
> 0, N
0
≥ 0,
then T
a
is called the average dwell time.
Remark 2: By the average dwell time switching, we mean
a class of switching signals such that the average time interval
between consecutive switchings is at least T
a
. Then, a basic
problem for such systems is how to specify the minimal T
a
and, thereby, get the admissible switching signals such that the
underlying system is stable and satisfies a prescribed perfor-
mance if the system dynamics meet some conditions. As com-
monly used in this paper, we choose N
0
= 0 in Definition 1.
The following definition and lemma are introduced, which
will play a key role in deriving our main results.
Definition 2: The equilibrium ξ
∗
= 0 of the switched DNN
in (5) is said to be exponentially stable under α(t)ifthe
solution ξ(t) satisfies
ξ(t)
≤ η
ξ
t
0
C
1
e
−λ(t−t
0
)
, ∀t ≥ t
0
for constants η ≥ 1 and λ>0, and
ξ
t
C
1
sup
−d
2
≤θ≤0
ξ(t + θ)
,
˙
ξ(t + θ)
.
Lemma 1: For any positive symmetric constant matrix M ∈
R
n×n
, scalars r
1
and r
2
satisfying r
1
<r
2
, and a vector function
ω :
[
r
1
,r
2
]
→ R
n
such that the integrations concerned are well
defined [10], then Jensen’s inequality holds
r
2
r
1
ω(s)ds
T
M
r
2
r
1
ω(s)ds
≤ (r
2
−r
1
)
r
2
r
1
ω
T
(s)Mω(s)ds.
Remark 3: Notice that the phrase α(t) appears in
Definitions 1 and 2, and in the problem formulation.
This serves to emphasize that all results to be obtained
subsequently in this paper are dependent on the switching
signal α(t), and α(t) is not an arbitrary switching signal but
is one with the average dwell time.
III. Main Results
A. Constant Time-Delay Case
In this section, we shall consider the constant time-delay
case, i.e., (A1): 0 ≤ d(t) ≡ d.
Theorem 1: Given an integer m ≥ 1 and a constant β>0,
suppose that there exist positive definite matrices P(i) ∈ R
n×n
,
Q(i) ∈ R
mn×mn
, R(i) ∈ R
n×n
, and = diag{γ
1
,γ
2
,... ,γ
n
}
such that for i ∈ N
(i) βW
T
P
P(i)W
P
+ W
T
R
(i)
¯
R(i)W
R
(i)+W
T
Q
¯
Q(i)W
Q
+ sym
W
T
P
P(i)W
R1
(i)+W
T
KW
P
− W
T
W
+ W
T
FD
KW
D
−W
T
FD
W
FD
< 0 (8)
where
¯
R(i)
R(i)0
0 −R(i)
,
¯
Q(i)
Q(i)0
0 −Q(i)
W
R
(i)
⎡
⎣
−
d
m
C(i)0
n×mn
d
m
A(i)
d
m
A
d
(i)
e
−
βd
m
m
d
I
n
−
e
−
βd
m
m
d
I
n
0
n×(m+1)n
⎤
⎦
W
R1
(i)
−C(i)0
n×mn
A(i) A
d
(i)
W
Q
I
mn
0
mn×n
0
mn×2n
0
mn×n
e
−
βd
m
I
mn
0
mn×2n
W
0
n×(m+1)n
I
n
0
n
,W
P
I
n
0
n×(m+2)n
W
D
0
n×mn
I
n
0
n×2n
,W
FD
0
n×(m+2)n
I
n
.