G. Wang et al. / Applied Mathematics and Computation 314 (2017) 98–109 99
principle. Instead, the newest control signal may arrive at the destination before the older one. Packet disorder could
increase the difficulty of system modeling and make its analysis and synthesis complicated. It also has negative impacts on
system performance and could lead to waste of resources. Up to now, there are very few results to study this issue, where
some new interesting and challenging problems are introduced. By exploiting a packet disordering compensation method,
some LMI conditions were presented in reference [46] . Based on transforming the underlying system into a discrete-time
system with multi-step delays, the stability and H
∞
control problems of NCSs with packet disordering were considered in
[47,48] , while some less conservative results were given in [49] . Based on the average dwell-time method, a kind of packet
reordering method was proposed in [50] . By investigating these references, it is seen that the originally studied systems are
all without any time delay. To our best knowledge, very few results are available to design a disordered controller for delay
systems. All the facts motivate the current research.
In this paper, the stabilization problem of stochastic delay systems closed by a disordered controller is considered. The
main contributions of this paper are summarized as follows: 1) Different from references [7,8,11,13] where no disorder
occurs, a generally kind of stabilizing controller experiencing a disorder between control gains and system states is proposed
for stochastic delay systems; 2) Compared with the disordered stabilization results [46–50] , the original system considered
in this paper is more general and has time delay. Moreover, the disorder considered here takes place between system states
and control gains, which is different from the above references; 3) In contrast to the above methods dealing with disorder,
such a disorder is modeled into a controller with special uncertainties and handled by applying a robust approach. More
importantly, the switching probability between such uncertainties is also taken into account in the controller design; 4)
Because of the results presented with LMI forms, they could be applied to many different and complicated situations such
as discrete-time systems, signal estimation, and so on.
Notation: R
n
denotes the n -dimensional Euclidean space, R
m ×n
is the set of all m × n real matrices. E {·} means the math-
ematical expectation of [ · ]. · refers to the Euclidean vector norm or spectral matrix norm. In symmetric block matrices,
we use “
∗
”as an ellipsis for the terms induced by symmetry, diag { } for a block-diagonal matrix, and (M )
M + M
T
.
2. Problem formulation
Consider a kind of stochastic delay systems described as
d x (t) = (Ax (t) + A
τ
x (t − τ ) + Bu (t)) dt + (Cx (t ) + C
τ
x (t − τ ) + Du (t )) dω(t )
x (t ) = φ(t) , t ∈ [ −τ, 0]
(1)
where x (t) ∈ R
n
is the system state, u (t) ∈ R
m
is the control input, and ω( t ) is an one-dimensional Brownian motion or
Wiener process. Matrices A , A
τ
, B , C , C
τ
, and D are known matrices of compatible dimensions. Time delay τ satisfies τ ≥ 0.
φ( t ) is a continuous function and defined from [ −τ, 0] to R
n
. It is known that the traditional state feedback controllers for
delay systems are commonly as follows:
u (t ) = Kx (t) (2)
u (t ) = K
τ
x (t − τ ) (3)
u (t ) = Kx (t) + K
τ
x (t − τ ) (4)
where K and K
τ
are control gains to be determined. It is said that controller (4) is more general and has some advantages.
The main reason is both delay and non-delay states are taken into account. However, the action of controller (4) needs an
assumption that the control gains and theirs related states should be available in a right sequence. Unfortunately, due to
some practice constraints, this assumption may be very hard satisfied. In this paper, a kind of controller experiencing a
disorder phenomenon is proposed and described by
u (t ) =
Kx (t) + K
τ
x (t − τ ) , no disordering
K
τ
x (t ) + Kx (t − τ ) , disordering occuring
(5)
It is rewritten to be
u (t ) = [ α(t) K + (1 − α(t)) K
τ
)] x (t) + [ α(t) K
τ
+ (1 − α(t)) K] x (t − τ )
= [ K
τ
+ α(K − K
τ
) + (α(t) − α)(K − K
τ
)] x (t) + [ K + α(K
τ
− K) + (α(t) − α)(K
τ
− K)] x (t − τ ) (6)
Here, α( t ) is the Bernoulli variable and assumed to take values in a finite set S { 0 , 1 } , one has
Pr { α(t) = 1 } = α, Pr { α(t) = 0 } = 1 − α
Pr { α(t) − α} = 0 , Pr { (α(t) − α)
2
} = α(1 − α) (7)
Then, controller (6) is equivalent to
u (t ) = [
ˆ
K + (α(t) − α)
ˆ
K ] x (t) + [
ˆ
K
τ
+ (α(t) − α)
ˆ
K
τ
] x (t − τ ) (8)