Sliding Mode Control for Mismatched Uncertain Markov Jump

Systems
Youguo He1 , Yaliang An2 ,Yugang Zhang2 , Yuanwei Jing3
1. Faculty of Information Engineering, Shenyang University, Shenyang 110044, China
E-mail: hyg197715@163.com
2. Department of Electric Detection, Shenyang Artillery Academy, Shenyang 110044, China
E-mail: anyl@163.com, zhyg@163.com
3. Faculty of Information Science and Engineering, Northeastern University, Shenyang 110004, China
E-mail: ywjjing@mail.neu.edu.cn

Abstract: The sliding mode control problem of a class of mismatched uncertain Markov jump systems is considered.
On the assumption that the system uncertainties are known upper bounds, the sufficient conditions are proposed respec-
tively via linear matrix inequalities to guarantee the mean exponential stability of reduced-order sliding mode system.
Corresponding to mismatched system uncertainties, a robust sliding mode controller is designed, which is proved able to
guarantee that the system trajectory reach the sliding surface in a finite time interval and keep up here thereafter. Simula-
tion results demonstrate that the proposed controller has good stability and robustness for uncertain hybrid linear Markov
jumping systems.

Key Words: Sliding mode control, Markov jump systems, Mismatched uncertainty, Linear matrix inequalities

1 INTRODUCTION (SMC) is an effective method of robustness control[8−10] .

A lot of dynamical systems have variable structures subject
to random abrupt changes, which may result from abrupt
phenomena such as random failures and repairs of the com-
ponents, changes in the interconnections of subsystems,
sudden environment changes, modification of the operat-
ing point of a linearized model of a nonlinear systems,etc.
Systems with this character may be modelled as hybrid
ones, that is, to the continuous state variable, a discrete ran-
dom variable called the mode, or regime, is appended. The
mode describes the random jumps of the system parame-
ters and the occurrence of discontinuities.A special class of
hybrid systems is Markov jump systems (MJSs). In MJSs,
the random jumps in system parameters are represented by
a Markov process taking values in a finite set. This class
of systems may represent a large variety of processes in-
cluding those in the aircraft flight control systems, manu-
facturing systems, communication systems, and economic
systems,etc[1−3] . Over the past decade, many important is-
sues have been extensively studied for Markov jump linear
systems (MJLSs).More recently,H∞ control for uncertain
Markovian jump systems has been considered in [4]; sta-
bilization of Billinear uncertain time-delay stochastic sys-
tems can be found in [5] ;stability and output feedback sta-
bilization of Markov jump systems can be found in [6];and
Guaranteed cost control of a Markov jump linear uncertain
system has been discussed in [7].
In another active research area, sliding mode control
This work is supported by the National Natural Science Foundation
of China under Grant 60274009 and Specialized Research Fund for the
Doctoral Program of Higher Education under Grant 20020145007.
SMC has a good performance that is the system behavior
is insensitive to the internal parameter variations and exter-
nal disturbances.Thought a lot of work has been conducted
on Markon jumping systems,very few results are available
for sliding mode control problem of Markov jump sys-
tems.Recently,the problem of design of sliding mode con-
trol for Markon jumping systems without system uncertain-
ties has appeared in [11].However,the problem of design of
sliding mode control for Markonian jumping systems with
system uncertainties is still of interest.This motivated us to
study the sliding mode control problem of a class of uncer-
tain hybrid linear systems with Markov jump parameter.
In this note ,the sliding mode control problem of a class
of uncertain Markov jump systems is considered. Mis-
matched uncertainties are introduced into Markov jump
systems.Concept of mean exponentially stability for un-
derlying system is proposed.The sliding surface and reach-
ing motion controller for system will be designed.Above
problems are solved in terms of linear matrix inequali-
ties(LMIs).
2 PROBLEM FORMULATION AND PRELIM-
INARIES
Consider a class of uncertain hybrid linear systems with
Markovian jump parameter in the following form:
ẋ(t) = (A(ηt )+ΔA(ηt ))x(t)+B(ηt )(u(t)+G(ηt )w(t)).
(1)
where x(t) ∈ Rn is the state vector, u(t) ∈ R is the
control input vector, w(t) ∈ R is the disturbance, ηt is a
finite-state Markovian process having a state space S =

5234

978-1-4244-1734-6/08/$25.00 
c 2008 IEEE
  
{1, 2, ....., N },generator(πij ) with transition probability   Σ(ηt )
B(ηt ) = U1 (ηt ) U2 (ηt ) ΓT (ηt ).
from mode i at time t to mode j at time t + σ, i, j ∈ S: 0(n−1)×1
pij = P (ηt+σ = j|ηt = i) where Σ(ηt ) ∈ R, and Σ(ηt ) > 0, Γ(ηt ) = 1.By the
state transformation z = T (ηt )x , system(1) has the regular
πij σ + o(σ), if i = j, form
={ (2)
1 + πii σ + o(σ), if i = j. ż(t) = (Ā(ηt ) + ΔĀ(ηt ))z(t)
 

N 0(n−1)×1
where πii = − πij , πij ≥ 0 ∀i, j ∈ S, i = j, σ > + (u(t) + G(ηt )w(t)) (3)
j=1,j=i B2 (ηt )1×1
0, and limσ↓0 o(σ)/σ = 0.
For each possible value ηt ∈ S, A(ηt ) ∈ Rn×n , B(ηt ) ∈ where Ā(ηt ) = T (ηt )A(ηt )T −1 (ηt ),
Rn×1 are known real constant matrices, G(ηt ) are ΔĀ(ηt ) = T (ηt )ΔA(ηt )T −1 (ηt ).
known real constant scalar with appropriate dimen-
sions, ΔA(ηt ) ∈ Rn×n is parameter uncertainty matrix. Assumption 2: System uncertainties satisfies the following
For the sake of simplicity,we will denote the system matrix mismatched condition:
associated with mode ηt = i by M (ηt ) = M (i).
For the convenience of proof,the definitions ,lemma and as- ΔĀ(ηt ) = H(ηt )F (ηt )E(ηt ).
sumption are given as follows:  
Definition 1[12] : For stochastic Lyapunov function H1 (ηt )
whereH(ηt ) = , E(ηt ) = [E1 (ηt ) E2 (ηt )],
V (x(t), ηt = k), the weak infinitesimal operator Ψ of the H2 (ηt )
Markov process(x(t), ηt ), t ≥ 0 is given by H1 (ηt ), H2 (ηt ) E1 (ηt ), E2 (ηt ), are known real constant
ΨV (x(t), ηt ) = lim 1/σ[E {V (x(t + σ), ηt+σ )|x(t), ηt } matrices, F (ηt ) ∈ R is unknown time-varying matrix, but
σ→0 satisfied F (ηt )T F (ηt ) ≤ I ∀t ≥ 0, ηt ∈ S.
−V (x(t), ηt )] System(3) can be rewritten as:
ż1 (t) = [Ā11 (ηt ) + H1 (ηt )F (ηt )E1 (ηt )]z1 (t)
Definition 2[12] : System (1) with u(t) = 0, w(t) = +[Ā12 (ηt ) + H1 (ηt )F (ηt )E2 (ηt )]z2 (t)
0, ΔA(ηt ) = 0 is said to be mean exponentially sta-
ble(MES) if there
 exist constants  > 0, a > 0, b >
(4)
0 such that E x(t)2  ≤ bx(0)2 e−at when x(0) <
, where x(0)(t = 0, ηt = η0 , η0 ∈ S) is the initial condi- ż2 (t) = [Ā21 (ηt ) + H2 (ηt )F (ηt )E1 (ηt )]z1 (t)
tion. +[Ā22 (ηt ) + H2 (ηt )F (ηt )E2 (ηt )]z2 (t)
+B2 (ηt )[u(t) + G(ηt )w(t)].
Lemma 1: For any matrices D ∈ Rn×nf , E ∈
Rnf ×n ,and F ∈ Rnf ×nf with P > 0 , F  ≤ 1 , and (5)
scalar ε > 0 ,we have:
DF E + E T F T DT ≤ ε−1 DDT + εE T E where z1 ∈ R(n−1)×1 ,z2 ∈ R.
The sliding surface can be chosen as follows:
Assumption 1: G(i)w(t) ≤ g(i), i ∈ S , where g(i), i ∈  
S are positive scalars. s(t) = C(ηt )z(t) = C1 (ηt ) C2 (ηt ) z(t) (6)

The objective of this paper is where C1 (ηt ) ∈ R1×(n−1) ,C2 (ηt ) ∈ R ,C2 (ηt ) = 0 for
1)The sliding motion is mean exponentially stable; any ηt ∈ S. so on the sliding surface we have
2)System(1) is mean exponentially stable with the reaching s(t) = C1 (ηt )z1 (t) + C2 (ηt )z2 (t) = 0 and
control law u(t). z2 (t) = −C2 (ηt )−1 C1 (ηt )z1 (t).
Let K(ηt ) = C2 (ηt )−1 C1 (ηt ), and substitute z2 (t) =
3 MAIN RESULTS −K(ηt )z1 (t) to (4) gives the sliding motion
In this section, the design of sliding surface and reaching ż1 (t) = [Ā11 (ηt ) − Ā12 (ηt )K(ηt ) + H1 (ηt )F (ηt )E1 (ηt )
motion controller will be presented. Let us first consider −H1 (ηt )F (ηt )E2 (ηt )K(ηt )]z1 (t)
the problem of sliding surface design.
(7)
In order to obtain a regular form of system(1),we can
choose a nonsingular matrix
 T (ηt ) such
 that Let
0(n−1)×1
T (ηt )B(ηt ) = . Ã1 (ηt ) = [Ā11 (ηt )− Ā12 (ηt )K(ηt )+H1 (ηt )F (ηt )E1 (ηt )
B2 (ηt )1×1
−H1 (ηt )F (ηt )E2 K(ηt )]
where B2 (ηt ) = 0.
then the sliding motion(7) can be rewritten
For convenience[13]
 ,letT us partition

U2 (ηt ) ż1 (t) = Ã1 (ηt )z1 (t)
T (ηt ) = T . (8)
U1 (ηt )
n×1
where U1 (ηt ) ∈ R and U2 (ηt ) ∈ Rn×(n−1) are two Theorem 1:The reduced order system(8)is mean exponen-
sub-blocks of a unitary matrix resulting from the singular tially stable if there exist symmetric positive-definite ma-
value decomposition of B(ηt ), that is trix P (i) ∈ R(n−1)×(n−1) , and general matrix K(i) ∈

2008 Chinese Control and Decision Conference (CCDC 2008) 5235

 R1×(n−1) , i ∈ S,such that the following inequalities hold where Â1 (i) = Ā11 − Ā12 K(i),
for all i ∈ S Ê1 (i) = E1 (i) − E2 (i)K(i).
  Pre- and post-multiplying above inequalities by P (i)−1 on
Π11 (i) Π12 (i) both
T <0 (9) ⎛ sides,we have ⎞
Π12 (i) Π22 (i) P (i)−1 Ã1 (i)T + Ã1 (i)P (i)−1
⎜ +H1 (i)N (i)Ê1 (i)P (i)−1 ⎟
where ⎜ ⎟
⎜ −1 T ⎟
X(i) = P (i)−1 , ⎜ +(H1 (i)N (i)Ê1 (i)P (i) ) ⎟
⎜  ⎟ < 0.
Π11 (i) = Â1 (i)X(i) + X(i)ÂT1 (i) + πii X(i) ⎜ −1
N
−1 ⎟
⎜ +P (i) ( πij P (j))P (i) ⎟
+ε(i)−1 H1 (i)H1 (i)T , ⎝ j=1,j=i ⎠
Â1 (i) = Ā11 − Ā12 K(i), +πii P (i)−1
1/2 1/2
Π12 (i) = [ X(i)Ê1 (i)T πi1 X(i) πi2 X(i) From lemma i,there exist a scalar ε(i) > 0 satisfied
1/2
... πiN X(i) ], [H1 (i)N (i)Ê1 (i)P (i)−1 ]T + H1 (i)N (i)Ê1 (i)P (i)−1
Ê1 (i) = E1 (i) − E2 (i)K(i), ≤ ε(i)−1 H1 (i)H1T (i)+ε(i)[Ê1 (i)P (i)−1 ]T [Ê1 (i)P (i)−1 ].
Π22 (i) = −diag{ ε(i)−1 I X(1) X(2) ... X(N ) }. (11)
Let X(i) = P (i)−1 ,
Proof:Take the stochastic Lyapunov function as applying Schur complement formula, we can have inequal-
ities (9).
V (z1 (t), ηt , t) = z1T (t)P (ηt )z1 (t). (10)
Next, the result of designing reaching motion controller is
Let the mode at time t be i, then derived in the following theorem.
ΨV (z1 (t), i, t) = z1T (t)P (i)ż1 (t) + ż1T (t)P (i)z1 (t) Theorem 2:Assume the condition in theorem 1

N holds,i.e.,inequalities (9) have solutions X(i), and
+ πij V (j) the linear sliding surface is given by (6), and there
j=1
exist selected Ξ(i) ,i ∈ S such that the inequalities
= z1T (t)P (i)Ã1 (i)z1 (t) + z1T (t)ÃT1 (i)P (i)z1 (t) 
N

N −Ω(i)Ξ(i) − Ξ(i)T Ω(i) + πij Ω(i) < 0 have definite
+z1T (t) πij P (j)z1 (t) j=1
j=1 positive matrix solutionsΩ(i).Then the following con-

N
troller makes the sliding surface mean exponentially stable
= z1T (t)[P (i)Ã1 (i) + ÃT1 (i)P (i) + πij P (j)]z1 (t).
j=1 and globally attractive in finite time:
−1
N u(t) = − (C2 (i)B2 (i)) {[C1 (i) C2 (i)]Āz(t)
Let Θ(i) = P (i)Ã1 (i) + ÃT1 (i)P (i) + πij P (j).
j=1
+C(i)H(i)E(i)z(t)
Then ΨV < 0 when Θ(i) < 0. +C2 (i)B2 (i)[Ω(i)s(t)ζ(i)
so we have +g(i)]sign(Ω(i)s(t))}

ΨV z1 (t)T Θ(i)z1 (t) (12)

V = z1T (t)P (i)z1 (t)
≤ −γ,

where γ = mini∈S λmin (−Θ(i))

> 0. where C1 (i), C2 (i) are designed in theorem 1.
λmax P (i)
which Proof:Take the stochastic Lyapunov function as
 implies

t
E 0 ΨV (z1 (ξ), ηξ , ξ)dξ V (s(t), ηt , t) = sT (t)Ω(ηt )s(t) (13)


t
≤ −γE 0 V (z1 (ξ), ηξ , ξ)dξ .
Let the mode at time t be i, then
Using Dynkin’s formula and Gronwall-Bellman  lemma,
E [V (z1 (t), ηt , t)] ≤ e−γt E z1T (0)P (0)z1 (0) N

Moreover,it is easy check that ΨV (s(t), i, t) = sT (t)Ω(i)ṡ(t)+ṡT (t)Ω(i)s(t)+ πij V (j).
E[z1 (t)2 ] ≤ mini∈S {λ1min P (i)} E[z1T (i)P (i)z1 (i)] j=1
(14)
≤ mini∈S {λ1min P (i)} E[V (z1 (t), i, t)] Substitute (6),(12)to(13),we have
≤ mini∈S {λ1min P (i)} e−γt E[z1T (0)P (0)z1 (0)] ΨV (s(t), i, t) ≤ s(t)T {−Ω(i)Ξ(i) − Ξ(i)T Ω(i)
≤ maxi∈S {λmax P (0)} −γt
z1 (0)2 . N
mini∈S {λmin P (i)} e + πij Ω(i)}s(t)
j=1
From the definition 2,the reduced order system(8)is mean +2Ω(i)s(t)[C(i)H(i)N (i)E(i)z(t)
exponentially stable if there exist symmetric positive- −C(i)H(i)E(i)z(t)
definite matrix P (i) ∈ R(n−1)×(n−1) , i ∈ S,such that +C2 (i)B2 (i)G(i)w(t)
Θ(i) < 0 for all i ∈ S. −C2 (i)B2 (i)g(i)
Θ(i) < 0 is equivalent to −C2 (i)B2 (i)Ω(i)s(t)ζ(i)]
⎛ ⎞
P (i)(Ã1 (i) + H1 (i)N (i)Ê1 (i)) 
N
when −Ω(i)Ξ(i) − Ξ(i)T Ω(i) + πij Ω(i) < 0.
⎜ +(Ã1 (i) + H1 (i)N (i)Ê1 (i)) P (i) ⎟
T
⎜ ⎟ < 0. j=1
⎝ N ⎠ ΨV (s(t), i, t) ≤ −2C2 (i)B2 (i)Ω(i)s(t)2 ζ(i)
+ πij P (j)
j=1 Because

5236 2008 Chinese Control and Decision Conference (CCDC 2008)

Ω(i)s(t)2 = [Ω(i)s(t)]T [Ω(i)s(t)] = s(t)T Ω(i)2 s(t)
0.5
≥ λmin (Ω(i)2 )s(t)T s(t), 0.45
x1
x2

so we have 0.4

ΨV (s(t), i, t) ≤ −s(t)T βs(t). 0.35

where β = 2(i)C2 (i)B2 (i)λmin (Ω(i)2 ) > 0. 0.3

x1−x2
0.25
so
ΨV (s(t),i,t) s(t)T βs(t)
V (s(t),i,t) ≤ − s(t)T Ω(i)s(t) ≤ −γ.
0.2

0.15

where γ = maxi∈S {λβmax (Ω(i)} > 0. 0.1

From the proof of theorem 1,we can know the sliding sur- 0.05

0
face is mean exponentially stable and globally attractive in 0 0.5 1 1.5 2
time(s)
finite time.
Figure 1: Responses of system state variables (x1,x2)
4 EXAMPLE
In this section ,for purpose of illustrating the usefulness of 0.18
the theory developed in this note,we present a simulation 0.16
example. 0.14

Let us consider the following system with generator for 0.12

Markov process
 governing
 the mode switching
 being: 0.1

s(t)
π11 π12 −4 4 0.08

= = .
π21 π22 3 −3 0.06

For the two operating modes,the associated data are 0.04

0.02

0
mode 1   −0.02
−3 −2 0 0.5 1 1.5 2
A(1) = , time(s)
1 0
  Figure 2: Responses of system sliding motion s
0.1  
ΔA(1) = sin(t) 0.3 0.4
0.2
 
0
B(1) = , G(1) = 1, w(t) = 0.1sin(t).
1 0.2

mode 2  
−0.2

−1 0 −0.4
A(2) = ,
2 −2 −0.6
 
 
u(t)

0.3
−0.8

ΔA(1) = sin(t) 0.1 0.2 −1

0.4
  −1.2
1
B(2) = , G(2) = 1, w(t) = 0.1sin(t). −1.4

0
   
−1.6

1 0 0 1 −1.8
Taking T (1) = , T (2) = , using The- 0 0.5 1 1.5 2

0 1 1 0
time(s)

orem 1 and LMI method,we have Figure 3: Responses of system control input u
P (1) = 3.2044, K(1) = −0.1569,
P (2) = 3.2103, K(2) = −0.1572.
ε(1) = 4.1808, ε(2) = 4.3643.
2
Then
C1 (1) = −0.0392, C2 (1) = 0.25, 1.8

C1 (2) = −0.0707, C2 (2) = 0.45.

markov jump

1.6
Taking Ξ(1) = 0.8, Ξ(2) = 0.54, by theorem 2 and LMI
method,we have Ω(1) = 1.9658, Ω(2) = 2.1037. 1.4

Taking g(1) = g(2) = 1, ζ(1) = 1, ζ(2) = 1, the initial

 T 1.2
value x(0) = 0.5 0.5 , η0 = 1, we have the follow-
ing simulation results. 1

0 5 10 15 20 25 30 35 40
time(0.05ms)
From Figs.1 and 2,we can see that trajectories of the system
state and sliding motion converge to the origin quickly and Figure 4: Markov jumping parametrs
will be in the stead state.

2008 Chinese Control and Decision Conference (CCDC 2008) 5237

5 CONCLUSIONS
In this note,the robust sliding mode control problem of a
class of mismatched uncertain hybrid linear systems with
Markov jump parameter is considered.By nonsingular state
transformation,system has been transformed the regular
form.In term of LMIs,the sufficient conditions are proposed
to guarantee the mean exponentially stability of reduced-
order sliding mode system for known upper bounds mis-
matched system uncertainties.Then, corresponding reach-
ing motion controller is designed such that the resulting
closed-loop system can be driven onto sliding surface in a
finite time.The simulation resulting has shown the sliding
mode control in this note is robust stable to the internal pa-
rameter uncertainties and external disturbances.
REFERENCES
[1] D D Sowrder, R O Rogersk. An LQ-solution to a con-
trol problem associated with asolar thermal central receiver,
IEEE Trans on Automatic control, vol.28, No.10, pp 971-
977, 1983.
[2] D D Moerder, N Halyo, J R Broussard. Application of pre-
computed control laws in a reconfigurable aircraft flight con-
trol system, J. Guid., Control, Dyna, vol.12, No.3, pp 325-
333, 1989.
[3] E K Boukas, Q Zhang, G Yin. Robust production and main-
tenance planning in stochastic manufacturing systems, IEEE
Trans on Automatic control, vol.40, No(6), pp 1098-1102,
1995.
[4] P Shi, E K Boukas. H∞ control for Markovian jumping lin-
ear systems with parametric uncertainty, J. Optim. Theory
Appl., vol. 95, pp 75-99, 1997.
[5] Z D Wang, H Qiao, K J Burnham. On Stabilization of
Billinear Uncertain Time-Delay Stochastic Systems with
Markovian Jumping Parameters, IEEE Trans on Automatic
control, vol.47, No.4, pp 640-646, 2002.
[6] Boukas E K, Liu Z K, Shi P. Delay-dependent stability and
output feedback stabilization of Markov jump systems with
time-delay, IEE Proc Control Theory Appl, vol. 149(5), 2002,
pp 379-386.
[7] E K Boukas, Z K Liu, F Al-Sunni. Guaranteed cost con-
trol of a Markov jump linear uncertain system using a time-
multiplied cost function, J.Optim.Theory Appl., vol.116,
No.1, pp 183-204, 2003.
[8] H H Choi. On the existence of linear sliding surface for a
class of uncertain dynamic systems with mismatched uncer-
tainties, Automatica, vol.35, No.10, pp 1707-1715, 1999.
[9] Y Niu, J Lam, X Wang and et al. Sliding-mode control
for nonlinear state-delayed systems using neural network
approximation, IEE Proc., Control Theory Appl., vol.150,
No.3, pp 233-239, 2003.
[10] J Hu, J Chu, H Su. SMVSC for a class of time-delay un-
certain systems with mismatching uncertainties, IEE Proc.,
Control Theory Appl., vol.147, pp 687-693, 2000.
[11] P Shi, Y Q Xia, G P Liu and et al. On Designing of Sliding
Mode Control for Stochastic Jump Systems ,IEEE Trans on
Automatic control, vol.51, No.1, pp 97-103, 2006.
[12] M Mariton, Jump linear systems in automatic control, New
York: Marcel Dekker, NJ; 1990.
[13] K S Kim,Y Park, S H Oh. Designing of robust sliding
surfaces for parametric uncertain systems:A Riccati appo-
rach,Automatica, vol.36, pp 1041-1048, 2000.

5238 2008 Chinese Control and Decision Conference (CCDC 2008)