114 H. Shen et al. / Fuzzy Sets and Systems 356 (2019) 113–128
inevitable phenomenon of time delays, the information latching is also an important issue in the study of NNs. In this
regard, the parameters of NNs can stochastically switch from one to another in finite modes at different time. As a
powerful tool, Markov jump neural networks (MJNNs) are capable of providing a suitable frame
work for modeling
NNs subject to information latching. Therefore, MJNNs have naturally become the focus of research in the control
field, and many significative results have been studied in the literature, see, e.g., [33,42] and the references therein.
Another research direction in the study of delayed NNs is to handle the comple
x nonlinear states by utilizing the
Takagi–Sugeno (T–S) fuzzy model. Such a model, as we all know, has many advantages in inaccurate mathematical
model of the controlled plant, strong robustness and powerfully modeling complex nonlinear systems [2,9,13,20,21,
26,28,32,44]. Consequently
, various control issues of T–S fuzzy systems have been investigated, such as stability
analysis, fault detection, filtering/state estimation, performance monitoring, adaptive control problems and passivity
analysis [5,11,15,30,31,43,45]. In particular, by the aid of the T–S fuzzy model approach, the stability analysis for
some complex nonlinear NNs and observer design problems ha
ve been addressed [3,12,34,36,47]. To mention a few,
for non-autonomous stochastic T–S fuzzy cellular neural networks, a global exponential p-stability criterion was
established in [17]. Based on the neural network-based approach, the adaptive control problem for a class of fuzzy
stochastic systems with the non-affine pure-feedback form and unknown functions wa
s investigated in [6,35]. Based
on a fuzzy method, the adaptive synchronization problem for memristive neural networks was addressed in [41]. It is
worth noting that the states of neural network are normally incomplete available in the network outputs. To overcome
this difficulty, it could be reasonable to estimate the neuronal states on the basis of the utilized measurements. More
recently
, in [29], a mixed H
∞
and passive filter design method for discrete-time fuzzy MJNNs (DFMJNNs) with time
delays was presented.
It should be noted that the aforementioned literature related to the state estimation of fuzzy NNs were considered
ove
r an infinite-time interval in the conventional Lyapunov asymptotic stability framework. In practical applications,
on the contrary, the states of closed-loop system only could be required to maintain in specified bounds over a gi
ven
time interval [1]. Accordingly, it is imperative to pay a great deal of attention on finite time stability of the underlying
systems [16,22]. Besides, in modern control systems, the information on measurement output is usually collected by
the sensors networks. As a result, it is inevitable that the occurrence of the netw
orked-deduced phenomena may be
displayed in form of parameter uncertainties and asynchronous mode jumping. Such phenomena may be caused by
random failures, imperfect network communications, failures of the components and sudden external disturbances, etc.
[8,14,25,48]. Although there is no denying that we cannot recognize the significance of the study of fuzzy NNs subject
to such phenomena too much, there are fe
w literatures which investigate finite-time H
∞
asynchronous state estimation
problem for DFMJNNs with randomly occurring uncertain measurements. The problem is of great significance but
also practicability, which motivates our study.
Summarizing the a
bove discussions, we are interested in addressing the asynchronous H
∞
state estimation for
DFMJNNs with randomly occurring uncertain measurements on a finite-time interval in this paper. A Bernoulli dis-
tributed white sequence is employed to determine whether the uncertain measurements occur or not [8,14]. The main
contributions of the work lie in the following three aspects: 1) It is significant that this paper could be re
garded as
the first attempt to cope with the finite-time H
∞
asynchronous state estimation problem against randomly occurring
uncertain measurements for DFMJNNs. 2) We consider the system which is comprehensive to cover asynchronous
jump mode information, randomly occurring uncertain measurements, hence reflecting the reality more closely. 3) By
employing the stochastic analysis technique, the design of finite-time H
∞
asynchronous state estimator is provided
for DFMJNNs.
Most notations emplo
yed in the paper are fairly standard as follows:
Z
+
/R
a
the set of non-negative integers/the a-dimensional vectors
R
a×b
the set of all a × b real matrices
P ≥ 0/P ≤ 0 the matrix P is positive/negative semi-definite
P>0/P < 0 the matrix P is positive/negative definite
λ
max
(
P
)
(respectively, λ
min
(
P
)
)
the largest (respectively, smallest) eigenvalue of symmetric matrices P
I/0 identity matrix / zero matrix with appropriate dimension
M
T
/M
−1
the transpose/inverse of the matrix M
l
2
[
0, ∞
)
the space of square-summable infinite vector sequences over
[
0, ∞
)
E
{
·
}
the expectation operator with respect to some probability measure P