978-1-5090-4093-3/16/$31.00 ©2016 IEEE 321
2016 12th International Conference on Natural Computation, Fuzzy Systems and Knowledge Discovery (ICNC-FSKD)
LMI-based Criteria for Cohen-Grossberg BAM
Neutral Neural Networks with Continuously
Distributed Delays
Guoquan Liu
1
,
Zhicheng Wang
1
,
1
Jiangxi Province Engineering Research Center of New
Energy Technology and Equipment,
East China University of Technology,
Nanchang, China
Chaomin Luo
2
, Xianxi Luo
1
2
Department of Electrical and Computer Engineering,
University of Detroit Mercy,
Michigan, USA
Abstract—Under two different assumptions, the issue of global
stability problem is considered for Cohen-Grossberg BAM
(CGBAM) neutral neural networks (NNS) with continuously
distributed delays. The amplification functions are handled by an
interval value. By establishing a new Lyapunov–Krasovskii
functional (LKF) and using inequality technique, the novel global
stability criteria of CGBAM neutral neural network are
described by a linear matrix inequality (LMI). The result
establishes a relationship between the CGBAM neural networks
and the neutral neural networks.
Keywords-Cohen-Grossberg; global stability; Lyapunov
functional method; continuously distributed delays; BAM neural
networks
I.
I
NTRODUCTION
In 1983, Cohen and Grossberg firstly studied a class of n-
dimensional competitive dynamical systems [1]. After that,
researchers followed them, continued to do research about this
model, and extend the model. They called this model as Cohen-
Grossberg neural networks (CGNNS). In hardware
implementation of NNS, time delays may lead to instability.
Over the past decade, the stability of CGNNS with time delays
becomes a hot research area [2, 3]. For example, Ye, Michel,
and Wang carried out the exponential stability analysis for
CGNNS with multiple delays [2]. Wang and Zou further
studied the globally asymptotic stability (GAS) of a unique
equilibrium for CGNNS[3]. Chen and Zheng [4] extended the
results of [2] and [3]. By introducing a novel Lyapunov
function, some new sufficient conditions are derived for
impulsive CGNNS with time-varying delays [5].
Recently years, researchers have proposed some new
conclusions on the CGNNS with neutral-type delay. Neutral-
type delay also frequently causes systems instability or other
poor performance. Two delay independent sufficient stability
criteria for a class of CGNNS with neutral-type delays are
obtained in [6]. In [7], Mandal and Majee solved the existence
of periodic solutions problem for CGNNS with neutral-type
time-varying delays by establishing a k-set contractive
operator.
Recently, it is noted that the dynamic analysis issue of
BAM neural networks (BAM-NNS) with constant(variant)
delays has addressed in [8-9]. For example, Xu et al. [10]
provided the exponential stability of inertial general BAM-
NNS with proportional delays. Liu and Yang [11] solved the
GAS for stochastic BAM-NNS of neutral-type with variant
delays. Zhang et al. performed the GAS analysis for
generalized CGBAM-NNS with neutral type delays [12]. It is
pointed that Jian and Wang solved the Lagrange stability for
CGBAM-NNS of neutral-type with finite distributed delays by
constructing an appropriate LKF [13]. But so far there is just a
few research conclusions published on the stability of CGBAM
neutral NNS with continuously distributed delay.
Motivated by the above analysis, the problem of global
stability is considered for CGBAM neutral NNS with
continuously distributed delays. By constructing a new LKF
and combing inequality techniques with an LMI method, some
global stability condition are obtained.
There are some notions depicted as follows.
"T"
stands for matrix transposition; For a real square
matrix
Γ
, 0Γ< means that
Γ
is real symmetric and negative
definite;
12
diag{ , ,..., }
n
ΓΓ Γ
denotes a block diagonal matrix; A
symbol
∗
represents the entries implied by symmetry.
II. P
ROBLEM DESCRIPTION
Consider the following CGBAM neutral NNS:
]
[
]
111 2
3
41 1
221 2
3
42 2
() ( ()) () ( ()) ( ( ()))
()(())
(()),
() ( ()) () ( ()) ( ( ()))
()(())
(()),
( ) ( ), 0, 1, 2,..., ,
t
t
ii
ut e ut dut Wf vt W f vt t
WKt fvd
Wut h t I
vt e vt dvt Rgut Rgut t
RZt gud
Rvt h t I
uip
τ
ρρρ
σ
ρρρ
ρφρρ
−∞
−∞
=− − − −
−−
−−+
=− − − −
−−
−− +
=≤=
( ) ( ), 0, 1, 2,..., ,
jj
vjq
ρϕρρ
=≤=
(1)