Finite-time H
∞
filtering of Markov jump
systems with incomplete transition
probabilities: a probability approach
ISSN 1751-9675
Received on 10th September 2014
Revised on 16th April 2015
Accepted on 16th May 2015
doi: 10.1049/iet-spr.2014.0376
www.ietdl.org
Mouquan Shen
1,2,3
✉
, Shen Yan
1
, Ze Tang
4
, Zhou Gu
3,5
1
College of Electrical Engineering and Control Science, Nanjing Technology University, Nanjing 211816, People's Republic of China
2
Key Laboratory of Advanced Control and Optimization for Chemical Processes, Shanghai 200237, People’s Republic of China
3
School of Automation, Southeast University, Nanjing 210096, People’s Republic of China
4
Department of Electrical Engineering, Yeungnam University, 280 Daehak-Ro, Kyonsan 712-749, Republic of Korea
5
College of Mechanical & Electronic Engineering, Nanjing Forestry University, Nanjing 210037, People’s Republic of China
✉ E-mail: mouquanshen@gmail.com
Abstract: This paper concerns the finite-time H
∞
filtering of discrete Markov jump system with inco mplete transition
probabilities which cover the cases of known, uncertain and unknown. To include all possible cases, with the
probability viewpoint, a truncated Gaussian distribution is employed to describe them. To ensure the filtering error
systems to be finite-time stochastic stable with a prescribe d noise attenuation level, sufficient conditions for the H
∞
filter design are yielded in terms of solvability of a set of linear matrix inequalities. A numerical example is given to
illustrate the effectiveness of the proposed method.
1 Introduction
Over the past decades, an increasing research activity has been
devoted to Markov jump systems (MJSs). Many results about
stability, stabilisation, particle filtering, synchronisation control and
sliding mode control have been scattered in the literature, see [1–
19]. Paralleling to these fruitful theoretical results, many efforts
have also been made on the widespread use of this kind of
systems, such as aircraft control systems, robotic manipulator
systems, power systems and transportation systems [1, 2].
Different from linear systems, the dynamic process of MJSs is
governed by a Markov chain, namely, the mode transited from one
to another is determined by its transition probabilities which has
direct influence on system stability and performance. Under the
assumption that the transition probabilities are completely known,
a lot of achievements have been reported on stability analysis,
controller design and optimisation, see [3–19]. Nevertheless, in
engineering, it is difficult or costly to catch the full information of
all transition probabilities, which leads that the results established
by the idea assumption is restrictive. Taking vertical take-off
landing helicopter system in the aerospace industry as an example,
the airspeed variation involved in the system matrices are
modelled as a Markov chain. Due to the limitation of the
measurement equipment, not all the probabilities of the jumps
among multiple airspeeds are readily to be measured.
Consequently, the obtained transition probabilities may be
inaccurate. To conquer this difficulty, some tentative methods have
been developed in [20–27]. Concretely speaking, [20, 21] utilise
the robust methodology to deal with transition probabilities with
norm bounded or polytopic uncertainties. Considering a more
realistic situation that the practical transition probability matrix
may be partly known, the stability, stabilisation and filtering
problems are studied by the authors of [22–26]. With the
stochastic viewpoint, an alternative representation of partly known
transition probabilities is given in [27–29].
On the other hand, an important problem in the process of control
system design is its stability. For MJSs, stochastic stability, moment
stability, exponential stability and almost sure stability are well
investigated by the authors of [3, 4] over an infinite-time interval.
However, the stochastic stability in infinite-time interval may
cause large values of the states which are not acceptable in the
presence of saturations. To deal with this transient performance of
control dynamics, finite-time stability or short-time stability for
linear system is proposed in [30–35]. By extending them to
stochastic scenario, the finite-time stochastic stability is adopted to
study the behaviour of MJSs during a fixed finite-time interval
[36–40]. Specially, the
finite-time H
∞
filtering of time-delay
stochastic jump systems with unbiased estimation is investigated in
[36]. On the basis of dynamic observer-based state feedback
method, the finite-time H
∞
fuzzy controller design for non-linear
jump system with time delays is presented in [37]. Finite-time H
∞
estimation of discrete-time MJSs with time-varying transition
probabilities subject to average dwell time switching is discussed
in [40]. With the consideration of partly known transition
probabilities, the finite-time stochastic stability and stabilisation of
discrete MJSs and the finite-time H
∞
filtering non-linear stochastic
systems are studied in [38, 39], respectively.
This paper further considers the finite-time filtering of discrete
MJSs with incomplete transition probabilities which are assumed
to be known, uncertain and unknown. To present these transition
probabilities in a unified framework, a stochastic description is
proposed in terms of the truncated Gaussian method. On the basis
of this description, the main focus is concentrated to the finite-time
filter design. To make the filtering error system be stochastically
stable with a prescribed H
∞
performance index, on the basis of the
stochastic Lyapunov theory and the transition probability property,
sufficient conditions for the desired filter design are developed in
terms of solvability of a set of linear matrix inequalities. A
numerical example is provided to demonstrate the effectiveness of
the proposed approach.
Notation: Throughout this paper, M
T
represents the transpose of
matrix M. ℤ
+
denotes the set of positive integers and N ∈ ℤ
+
. The
notation X ≤ Y(X < Y) where X and Y are symmetric matrices,
means that X − Y is negative semi-definite (negative definite),
respectively. I and 0 represent identity matrix and zero matrix,
respectively. L
2
denotes the space of square integrable vector
functions of a given dimension over [0, ∞), with norm
E ||x||
2
2
= E
1
k=0
x(k)
T
x(k)dt
, 1. w denotes the entries of
matrices implied by symmetry. η
min
and η
max
denote the smallest
and the largest eigenvalue of matrix P, respectively.
4
l
ij
and
4
u
ij
IET Signal Processing
Review Article
IET Signal Process., 2015, Vol. 9, Iss. 7, pp. 572–578
572
&
The Institution of Engineering and Technology 2015