An LKF Method to H-Infinity State Estimator of Neural Networks with
Mixed Interval Delays
Guoquan Liu
1
, Chaomin Luo
2
, Shumin Zhou
1
, Xianxi Luo
1
, Hong Xia
1
1. Jiangxi Province Engineering Research Center of New Energy Technology and Equipment, East China University of Technology,
Nanchang, 330013, China
E-mail: gqlecit@hotmail.com
2. Department of Electrical and Computer Engineering, University of Detroit Mercy, Michigan, USA
E-mail: luoch@udmercy.edu
Abstract: This paper explains the H state estimation problem for neural networks with mixed interval delays. Firstly, a new
neural networks model is constructed, which contains an interval discrete time-varying delay and an interval natural-type
time-varying delay. Secondly, a new Lyapunov-Krasoskill Functional (LKF) is established, which contains several integral
terms. Lastly, by inequality techniques, and linear matrix inequality (LMI) method, a new criterion is presented so that the error
system is globally asymptotically stable with H performance. It is also shown that the estimator gain matrix can be solved by a
LMI.
Key Words: Neutral-type, Mixed interval delays, H state estimator
1 Introduction
The state estimation issue for neural networks has been
studied for many years [1-8]. Park and Kown [9] proposed
the state estimation problem for a class of neural networks
with discrete constant delay and neutral delay. Park and
Kown [10] extended the neural network model, a state
estimator for a class of neural networks with discrete
time-varying delay and neutral time-varying delay is
obtained. Next, they continued the same model and solved
the problem of state estimation for neural networks with
interval discrete time-varying delays and a natural-type
time-varying delay [11]. However, interval natural-type
delays are not taken into account in [9-11] yet.
Recently, H concept was proposed to reduce the effect
of the disturbance input on the regulated output to a
prescribed level. Hence the problem of H control for
neural networks with delay has received considerable
attention (see for example [12–14]). Based on the LKF and
LMI frameworks, the problem of H state estimation for
static neural networks with time-varying delays has been
solved in [15].Very recently, Mathiyalagan et.al [16]
studied the problem of robust exponential stability and H
control for switched neural networks with discrete
time-varying delay and natural-type time-varying delay. Liu
et.al [17] proposed H infinity state estimation for
neutral-type neural networks with continuously distributed
delays. However, the H state estimation problem for
neural networks with discrete time-varying delay and
natural-type time-varying delay hasn’t been fully
investigated. The main objective of this paper is to design a
H state estimator gain matrix such that the neural
networks model with mixed interval delays is globally
stable.
*
This work is supported by the Start-up Foundation for Doctors of East
China Institute of Technology (No.DHBK2012201), the Jiangxi foreign
science and technology cooperation plan (No.20132BDH80007), and the
National Natural Science Foundation (No. 11565002, 51409047,
61463003, 51567001, 61064009).
2 System description
Consider the following neural networks with mixed
interval delays described by the following state equation:
>@
12
12
() () ( ()) ( ( ()))
(()) (),
() () ( ()) ( ()) (),
() (), () (), ,0,
xt Axt Agxt Agxt t
Cx t h t Dw t J
yt Bxt Bxt t Bxt ht Ewt
zt Lxt xt t t
W
W
I
°
°
®
°
°
f
¯
(1)
where
>@
12
( ) ( ), ( ),..., ( )
T
n
n
xt x t x t x t
is the state vector of
the neural network with
n
neurons,
>
12
( ( )) ( ( )), ( ( ))gxt g xt g xt
@
,..., ( ( ))
T
n
n
gxt
is the neuron activation functions which
present the nonlinear parameter perturbations.
()wt
is the
noise input,
>@
12
, ,...,
T
n
JJJ J
is the external bias
vector,
()yt
is the measurement output of the
network ,
^`
12
diag , ,...,
n
Aaaa
is a diagonal matrix and has
positive and unknown entries
0, , 1,2, , , , ,
ii i
aAi CDBB!
1, 2, ,iEL
are the interconnection matrices representing
the weight coefficient of the neurons,
()t
W
and
()ht
are the
transmission time-varying delays satisfying
0 () , () 1,0 () , () 1,tt hhthhth
WW WW W
dd d dd dd d dd
(2)
where
,,,,hh
WWW
and
h
are known constants.
In this paper, the following Assumption for the neuron
activation function is given as follows.
Assumption 1. The neuron activation function
()
i
g
is
bounded and satisfies the following conditions:
() ()
,,
ii
ii
gx gy
llxy
xy
ddz
(3)
where
i
l
and
( 1,2,3,..., )
i
li n
are some constants and they
can be positive, negative and zero.
The following state estimator for estimation of
()zt
can
be constructed, which is based on the measurement
()yt
.
Proceedings of the 35th Chinese Control Conference
Jul
27-29, 2016, Chen
du, China
3694