impulsive Cohen–Grossberg neural networks with time-varying
delays and reaction–diffusion terms has been found by establish-
ing a delay differential inequality with impulsive initial condi-
tions and employing M-matrix theory in [35].
On the other hand, the fuzzy neural network is also a kind of
important neural networks [36,45]. Studies have shown that the
fuzzy neural network is a very useful paradigm for image processing
problems. Recently, some results on stability have been derived for
the impulsive fuzzy neural networks (IFNNs) with time delays and
reaction–diffusion terms (see [38,39]). To the best of our knowledge,
there are no results on the exponential stability of impulsive fuzzy
Cohen–Grossberg neural networks (IFCGNNs) with mixed time
delays and reaction–diffusion terms. From the view of mathematical
model, the IFCGNNs belong to new category of dynamical system,
which have not only fuzzy logic between its template input and/or
output besides the sum of product operation, but also impulsive
effects. Such a model displays a combination of characteristics of the
fuzzy logic and impulse, and has complex dynamical behaviors.
Therefore, it is necessary to further investigate the dynamical
behaviors of the IFCGNNs.
Motivated by the above discussions, in this paper, we will
investigate the global exponential stability of impulsive fuzzy
Cohen–Grossberg neural networks with mixed time delays and
reaction–diffusion terms. Based on the Lyapunov method, Poin-
care
´
Integral Inequality, and the linear matrix inequality (LMI)
approach, we obtain some sufficient conditions ensuring the
global exponential stability of equilibrium point for impulsive
fuzzy Cohen–Grossberg neural networks with mixed time delays
and reaction–diffusion terms. These results are less conservative
than some known results.
The rest of this paper is organized as follows. Model descrip-
tion and preliminaries are given in Section 2.InSection 3, we give
main results and their proof. Examples are given to illustrate our
theory in Section 4. Finally, in Section 5 we give the conclusions.
2. System descriptions and preliminaries
In this paper, we consider the following model:
@u
i
ðt, xÞ
@t
¼ w
i
D
u
i
ðt, xÞa
i
ðu
i
ðt, xÞÞ
b
i
ðu
i
ðt, xÞÞ
X
n
j ¼ 1
c
ij
f
j
ðu
j
ðt, xÞÞ
X
n
j ¼ 1
d
ij
v
j
2
4
^
n
j ¼ 1
a
ij
g
j
ðu
j
ðt
t
j
ðtÞ, xÞÞ
^
n
j ¼ 1
g
ij
Z
t
1
k
j
ðtsÞh
j
ðu
j
ðs, xÞÞ ds
^
n
j ¼ 1
T
ij
v
j
3
n
j ¼ 1
b
ij
g
j
ðu
j
ðt
t
j
ðtÞ, xÞÞ 3
n
j ¼ 1
y
ij
Z
t
1
k
j
ðtsÞh
j
ðu
j
ðs, xÞÞ ds
3
n
j ¼ 1
H
ij
v
j
J
i
#
, t Z t
0
, t a t
k
, xA
O
u
i
ðt
þ
k
, xÞ¼u
i
ðt
k
, xÞþI
ik
ðu
i
ðt
k
, xÞÞ, xA
O
, kA N,
@u
i
ðt, xÞ
@m
¼ 0, t Z t
0
, xA @
O
,
u
i
ðt, xÞ¼
f
i
ðt, xÞ, t A ð1, t
0
, xA
O
,
8
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
<
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
:
ð1Þ
for i ¼ 1; 2, ..., n.Where
O
is a bounded compact set in space R
l
with
smooth boundary @
O
and mes
O
Z 0inspaceR
l
;denoteby@=@m the
outward normal derivative,
O
¼
O
[ @
O
,and
D
u
i
ðt, xÞ¼
P
m
s ¼ 1
ð@
2
u
i
ðt, xÞ=@x
2
s
) represents the Laplace operator; n corresponds to the
number of units in the neural networks; uðt, xÞ¼ðu
1
ðt, xÞ, u
2
ðt, xÞ, ...,
u
n
ðt, xÞÞ
T
, u
i
ðt, xÞ corresponds to the state of the ith unit at time t and
position x; w
i
4 0 corresponds to the transmission diffusion coeffi-
cient along the ith neuron; a
i
ðu
i
ðt, xÞÞ represents an amplification
function; b
i
ðu
i
ðt, xÞÞ is an appropriate behavior function; f
i
, h
i
and g
i
are the activation functions;
a
ij
and
g
ij
are elements of the fuzzy
feedback MIN template;
b
ij
and
y
ij
are elements of the fuzzy feedback
MAX template; T
ij
and H
ij
are elements of fuzzy feed-forward MIN
template and fuzzy feed-forward MAX template, respectively; c
ij
and
d
ij
are elements of feedback and feed-forward template, respectively;
4 and 3 denote the fuzzy AND and fuzzy OR operations, respec-
tively; v
i
and J
i
denote input and bias of the ith neuron;
t
j
ðtÞ
corresponds to the transmission delay; k
j
ðsÞZ 0isthefeedback
kernel, t
k
is called impulsive moment and sati sfies 0r t
0
r t
1
r
t
2
r ,lim
k- þ1
t
k
¼þ1; u
i
ðt
k
, xÞ and u
i
ðt
þ
k
, xÞ denote the left
limitandrightlimitatt
k
and position x, respectively; I
k
ðuðt
k
, xÞÞ ¼
ðI
1k
ðu
1
ðt
k
ÞÞ, I
2k
ðu
2
ðt
k
ÞÞ, ..., I
nk
ðu
n
ðt
k
ÞÞÞ
T
, I
ik
ðuðt
ik
, xÞÞ shows impulsive
perturbation of the ith neuron at t
k
and position x.
Remark 1. The model (1) is new and more general than those
investigated in [7,8,11,12,14,17,18,23,24,29–39]. For instance, if
we do not consider continuously distributed delays and reaction–
diffusion terms, i.e, let f
i
ð
x
Þ¼g
i
ð
x
Þ, h
i
ð
x
Þ¼0, w
i
¼0, for
x
A R,
i ¼ 1; 2, ..., n, then (1) is the same as the model in [14]. Similarly,
we can illustrate the systems in [7,8,11,12,17,18,23,24,29–39] are
special cases of (1), since reaction–diffusion terms, impulsive
perturbations, fuzzy logic and mixed time delays are considered
in system (1).
To obtain our results, we give the following assumptions:
(H1) Each function a
i
ð
x
Þ is positive, continuous and bounded, i.e.
there exist constants a
i
, a
i
such that
0o
a
i
o a
i
ð
x
Þo a
i
o 1 for
x
A R, i ¼ 1; 2, ..., n:
(H2) b
i
ð
x
Þ is continuous, and there exists a positive diagonal
matrix B ¼ diagðb
1
, b
2
, ..., b
n
Þ such that
b
i
ð
x
1
Þb
i
ð
x
2
Þ
x
1
x
2
Z b
i
Z 0 for any
x
1
,
x
2
A R and
x
1
a
x
2
, i ¼ 1; 2, ...n:
(H3) The activation functions f
i
ð
x
Þ, g
i
ð
x
Þ and h
i
ð
x
Þ are continuous,
and there exist three positive diagonal matrices F ¼ diag
ðF
1
, ..., F
n
Þ, G ¼ diagðG
1
, ..., G
n
Þ and H ¼ diagðH
1
, ..., H
n
Þ
such that
F
j
¼ sup
x
1
a
x
2
f
j
ð
x
1
Þf
j
ð
x
2
Þ
x
1
x
2
, G
j
¼ sup
x
1
a
x
2
g
j
ð
x
1
Þg
j
ð
x
2
Þ
x
1
x
2
,
H
j
¼ sup
x
1
a
x
2
h
j
ð
x
1
Þh
j
ð
x
2
Þ
x
1
x
2
:
(H4) The transmission delay
t
j
ðtÞ is time-varying and satisfies
0r
t
j
ðtÞ r
t
and
_
t
j
ðtÞ r
m
, where
t
and
m
are some non-
negative constants.
(H5) Let r
k
ðuÞ¼uþI
k
ðuÞ be Lipschitz continuous in R
n
, that is,
there exists nonnegative diagnose matrix
G
k
¼ diagð
G
1k
,
G
2k
, ...,
G
nk
Þ satisfies
9r
k
ðuÞr
k
ðvÞ9r
G
k
9uv9, for all u, vA R
n
, kA N:
where r
k
ðuðt
k
, xÞÞ ¼ ðr
1k
ðu
1
ðt
k
, xÞÞ, r
2k
ðu
2
ðt
k
, xÞÞ, ..., r
nk
ðu
n
ðt
k
,
xÞÞÞ
T
, I
k
ðuðt
k
, xÞÞ ¼ ðI
1k
ðu
1
ðt
k
, xÞÞ, I
2k
ðu
2
ðt
k
, xÞÞ, ..., I
nk
ðu
n
ðt
k
, xÞÞÞ
T
.
(H6) k
j
ðsÞZ 0 is the feedback kernel, defined on the interval
½0, þ1Þ and satisfies the following conditions:
Z
1
0
k
j
ðsÞ ds ¼ 1 for j ¼ 1; 2, ..., n:
For convenience, we introduce several notations.
C. Zhou et al. / Neurocomputing 91 (2012) 67–7668