Published in IET Control Theory and Applications

Published in IET Control Theory and Applications

Received on 13th September 2008

Revised on 12th August 2009
doi: 10.1049/iet-cta.2008.0415
ISSN 1751-8644
Delay-dependent robust H
11111
control of
time-delay systems
M. Sun
1
Y. Jia
2
1
Department of Automation and Engineering, Northeastern University at Qinhuangdao, Qinhuangdao 066004 Hebei,
Peoples Republic of China
2
Department of Systems and Control, Beihang University, Beijing 100083, Peoples Republic of China
E-mail: [email protected]
Abstract: In this study, a delay-dependent H
1
performance criterion that possess decoupling structure is derived for a
class of time-delay systems. It is then extended to H
1
state-feedback synthesis for time-delay systems with polytopic
uncertainty and multichannel H
1
dynamic output-feedback synthesis for time-delay systems. All the conditions are
given in terms of the linear matrix inequalities. In some previous descriptor methods, the products of controller
matrices and Lyapunov matrices are completely separated in performance analysis, whereas it is not the case in
controller synthesis. However, with the method in the paper, the weakness is eliminated. Numerical examples
illustrate the effectiveness of our solutions as compared to results obtained by other methods.
1 Introduction
Application of a single Lyapunov function to the analysis and
design is investigated for systems with polytopic uncertainty
and multichannel constraints [1]. The strict requirement of a
single Lyapunov function for all admissible uncertainties and
all multichannel constraints can lead to conservative results.
To reduce the conservativeness, researchers turn to using
parameter-dependent Lyapunov functions and have obtained
many results [27]. Among these works, an effective idea is
to separate the products of Lyapunov matrices and controller
matrices in the given LMIs by introducing auxiliary slack
variables. A breakthrough towards this direction is the work
for discrete systems in [2] that is extended to continuous time
case in [37] by different methods. However, time delay is
not considered in the above references. Time delay is a source
of instability in many cases. Therefore the stability and
performance analysis for time-delay systems are of theoretical
and practical importance [812]. Static output-feedback
control problems are investigated in [11, 12]. However, time-
delay systems considered are with norm-bounded uncertainty
but not with polytopic uncertainty. Although stability and
stabilisability conditions with parameter-dependent Lyapunov
variables for time-delay system are presented in [13, 14].
These works are restricted to state-feedback and static output-
feedback control, which are easier to solve than dynamic
output-feedback control. In some situations, there is a strong
need to construct a dynamic output-feedback controller
instead of a static one in order to obtain a better performance
and dynamical behaviour of the state response. The dynamic
output-feedback control is considered in [15]; however, the
multichannel H
1
output-feedback synthesis still remains an
open problem and the products of controller matrices and
Lyapunov matrices are not completely separated in [15], and
this motivates the present paper.
In this paper, we solve dynamic output-feedback H
1
controller design problems for time-delay systems. We
present new delay-dependent H
1
performance conditions
based on the Lyapunov method. Then we derive the H
1
performance criterion, which decouples controller matrices
and Lyapunov matrices. It can reduce conservatism inherent
in the conventional method in solving both robust control
problems of polytopic systems and multichannel synthesis
problems by providing a parameter-dependent Lyapunov
function. Then we design a robust H
1
state-feedback
controller for polytopic systems and a dynamic H
1
output-
feedback controller for systems with multichannel constraints.
The advantages of the results over the conventional methods
are shown by two numerical examples. For simplication, we
use the symbol Sym{
.
} to denote Sym{X} =
def
X + X
T
, the
symbol

to denote the symmetric part.
1122 IET Control Theory Appl., 2010, Vol. 4, Iss. 7, pp. 11221130
& The Institution of Engineering and Technology 2010 doi: 10.1049/iet-cta.2008.0415
www.ietdl.org
2 H
1
performance analysis
Consider the following time-delay system
x(t) = Ax(t) + A
d
x(t h) + Bw(t)
z(t) = Cx(t) + Dw(t)
x(t) = 0, t [ [h, 0]
(1)
where, x(t) [ R
n
is the state, w(t) [ R
n
w
is the disturbance
signal of nite energy in the space L
2
[0, 1), z(t) [ R
n
z
is the
exogenous output signal, and A, B, A
d
, C, D are constant
matrices of appropriate dimensions. The time-delay h . 0
is assumed to be known. For a prescribed scalar g . 0, we
dene the performance index
J (w) =
_
1
0
(z
T
z g
2
w
T
w) dt (2)
Lemma 1 [16]: For any 2n 2n symmetric and positive-
denite matrix R . 0 and matrix M, the following inequality
holds
2
_
t
th
b
T
(a)a(a) da
_
t
th
a(a)
b(a)
_ _
T
R RM
(2, 2)
_ _
a(a)
b(a)
_ _
da
for a(a) [R
2n
, b(a) [R
2n
given for s [[t h, t]. Here
(2, 2) =(M
T
R+I )R
1
(RM +I ).
Lemma 2 [13]: Let f, a and b be given matrices of
appropriate dimension. Then the two statements are equivalent:
(i) f, a and b satisfy f , 0 and f + ab
T
+ ba
T
, 0.
(ii) f, a and b are such as the following LMI is feasible in the
variable G.
f a +bG
T
G G
T
_ _
=
f a
0
_ _
+Sym
0
I
_ _
G
b
T
I
_ _
_ _
,0
We rst provide delay-dependent H
1
performance
conditions for the time-delay system (1).
Theorem 1: For prescribed positive scalars g . 0, h . 0
and scalar l, system (1) is stable and the cost function (2)
achieves J(w) , 0 for all non-zero w [ L
2
[0, 1), if there
exist symmetric and positive-denite matrices P . 0,
S . 0, R
1
. 0, R
3
. 0 and a matrix R
2
such that (see (3))
where (1, 1) = Sym{P(A + A
d
+ lA
d
)} + S, (1, 3) = h(A+
(1 + l)A
d
)
T
A
T
d
R
3
.
The proof can be referred to in the appendix, and hence is
omitted here.
Corollary 1: For prescribed positive scalars g . 0, h . 0
and scalar l, system (1) is stable and the cost function (2)
2
[0, 1), if there
exist symmetric and positive-denite matrices Q . 0,
S . 0,

R
1
. 0,

R
3
. 0 and a matrix

R
2
such that (see (4))
where
(1, 1) = Sym{(A + A
d
+ lA
d
)Q}
(1, 3) = hQ(A + (1 + l)A
d
)
T
A
T
d
Proof: Pre- and post-multiplying both sides of the LMI (3)
by diag{Q,

S,

R
3
, I , I ,

R
3
,

R
3
}, denoting Q = P
1
,
S = S
1
,

R
1
= R
1
3
R
1
R
1
3
,

R
2
= R
1
3
R
2
R
1
3
,

R
3
= R
1
3
,
making a series of congruence transformations and
following the Schur complement Lemma, then we can
obtain Corollary 1 immediately. A
3 H
1
state-feedback synthesis
In this section, the results developed in the previous section
are extended to the state-feedback synthesis problem for
(1, 1) lPA
d
(1, 3) PB C
T
h(l + 1)P h(l + 1)P
S lhA
T
d
A
T
d
R
3
0 0 0 0
hR
3
hR
3
A
d
B 0 0 h
2
(l + 1)R
3
A
d
g
2
I D
T
0 0
I 0 0
hR
1
hR
2
hR
3
, 0 (3)
(1, 1) lA
d

S (1, 3) B QC
T
h(l + 1)
R
3
h(l + 1)
R
3
Q

S lh
SA
T
d
A
T
d
0 0 0 0 0
h
R
3
hA
d
B 0 0 h
2
(l + 1)A
d

R
3
0
g
2
I D
T
0 0 0
I 0 0 0
h
R
1
h
R
2
0
h
R
3
0

, 0 (4)
IET Control Theory Appl., 2010, Vol. 4, Iss. 7, pp. 11221130 1123
doi: 10.1049/iet-cta.2008.0415 & The Institution of Engineering and Technology 2010
www.ietdl.org
time-delay systems. Consider the following system
x(t) = Ax(t) + A
d
x(t h) + B
1
w(t) + B
2
u(t)
z(t) = Cx(t) + D
11
w(t) + D
12
u(t)
(5)
where u(t) [ R
n
u
and the other signals are the same with
system (1), D
12
is constant matrix of appropriate dimension.
We seek a state-feedback control law
u(t) = Kx(t) (6)
that will asymptotically stabilise the system and achieve
J (w) , 0 for all non-zero w [ L
2
[0, 1).
Applying the state-feedback control law (6) to system (5),
we can obtain the closed-loop system
x
c
(t) = A
c
x
c
(t) + A
d
x
c
(t h) + B
1
w(t)
z
c
(t) = C
c
x
c
(t) + D
11
w(t)
(7)
where
A
c
= A + B
2
K, C
c
= C + D
12
K (8)
Then we present the following delay-dependent condition
under which there exists a state-feedback H
1
controller for
system (7).
Theorem 2: For prescribed positive scalars g . 0, h . 0,
1 . 0 and scalar l, there exists a state-feedback controller such
that the closed-loop system (7) is asymptotically stable and the
cost function achieves J(w) ,0 for all non-zero w [ L
2
[0, 1)
if there exist symmetric and positive-denite matrices Q . 0,
S . 0,

R
1
. 0,

R
3
. 0 and matrix

R
2
, V, U such that (see (9))
where
(1, 6) = (1, 7) = h(l + 1)R
3
(1, 9) = Q + (A + A
d
+ lA
d
)V +
1
2
1V + B
2
U
(3, 7) = h
2
(l + 1)A
d

R
3
(3, 9) = hA
d
(A + A
d
+ lA
d
)V + hA
d
B
2
U
If the above condition holds, a desired state-feedback
controller can be given by u(t) = UV
1
x(t).
omitted here.
Remark 1: The descriptor system method is applied in [15].
From [15, Theorem 2.1], we can see that the products of
system matrices and Lyapunov matrices are completely
separated. However, from [15, Theorem 3.1], we can see
that the controller matrix is coupled with a symmetric and
positive-denite matrix Q
1
. Therefore, with the method in
[15], the products of controller matrices and Lyapunov
matrices are completely separated in performance analysis
while it is not the case in controller synthesis. However,
with our method, the weakness is eliminated. Therefore it
is expected to obtain less conservative results.
The LMI in Theorem 2 is afne in the system matrices.
It can thus be applied also to the case where these matrices
are uncertain and are known to reside within a given
polytope. Consider system (5) and denote
V =
A B
1
B
2
A
d
C D
11
D
12
_ _
we assume that V [ C
o
{V
i
, i = 1, . . . , N}, namely
V =

N
j=1
f
j
V
j
for some 0 f
j
1,

N
j=1
f
j
= 1, where
the N vertices of the polytope are described by
V
i
=
A
i
B
1i
B
2i
A
di
C
i
D
11i
D
12i
_ _
We obtain the following corollary.
Corollary 2: Consider system (5), where the system
matrices reside within the polytope V. For a prescribed
g . 0, the state-feedback law of (6) achieves, J(w) , 0 for
all non-zero w [ L
2
[0, 1) and for all the matrices in V if
for prescribed scalars 1 . 0, l, there exist symmetric and
positive-denite matrices Q
i
. 0,

S
i
. 0,

R
1i
. 0,

R
3i
. 0
and matrices

R
2i
, V, U that satisfy LMIs (9) for
i = 1, . . . , N, where the matrices
A
i
, A
di
, B
1i
, B
2i
, C
i
, D
11i
, D
12i
, Q
i
,

S
i
,

R
1i
,

R
2i
,

R
3i
are taken with the index i. The state-feedback gain is then
given by K = UV
1
.
1Q lA
d

S 0 B 0 (1, 6) (1, 7) 0 (1, 9)

S lh
SA
T
d
A
T
d
0 0 0 0 0 0
h
R
3
hA
d
B 0 0 (3, 7) 0 (3, 9)
g
2
I D
T
11
0 0 0 0
I 0 0 0 CV + D
12
U
h
R
1
h
R
2
0 0
h
R
3
0 0

S V
V V
T
, 0 (9)
www.ietdl.org
Example 1: We consider the following time-delay system
x(t) = Ax(t) + A
d
x(t h) + B
1
w(t) + B
2
u(t)
z(t) = Cx(t) + D
12
u(t)
(10)
where
A =
0 e
f 0.8
_ _
, A
d
=
0.1 0.02
0.03 0.1
_ _
,
B
1
=
0
0.2
_ _
, B
2
=
1
1
_ _
, C = 1 0
_ _
, D
12
= 1 (11)
where 0 e 0.1, 0.5 f 0.4, therefore, the plant
can be described as a polytope with four vertices. Our goal
is to nd a state-feedback gain K minimising the worst case
H
1
cost for all possible values of the parameters e and f.
From [15, Theorem 3.1], for h 0.5, a minimum value of
g 1.8069 is obtained with a corresponding state-feedback
gain of K = 8.3629 9.8057
_ _
10
4
. For g 3, the
gain of K = 398.5485 475.2630
_ _
is obtained.
Applying Corollary 2 in this paper, we obtain, for the
same h and for l 20.1, 1 100, a minimum g of
1.7698 with corresponding state-feedback gain of K =
[ 381.5342 446.1074 ]. For g 3, the gain of K =
[ 175.0746 210.5423 ] is obtained.
It can be seen that the method in this paper achieves
improvement of the H
1
upper bounds. It also shows that
the method in this paper arrives at much lower gain than
that of the method in [15]. Therefore the method in this
paper is less conservative.
4 H
1
output-feedback synthesis
In this section, we use the criterion derived above to provide a
new more accurate method for output-feedback controller
synthesis with multichannel constraints. Let us consider the
following time-delay system
x(t) = Ax(t) + A
d
x(t h) + B
1
w(t) + B
2
u(t)
z(t) = C
1
x(t) + D
11
w(t) + D
12
u(t)
y(t) = C
2
x(t) + D
21
w(t)
(12)
where y(t) [ R
n
y
is the measured output signal and the other
signals are the same with system (5).
Denote the output-feedback controller by
x(t) = A
K
x(t) + B
K
y(t)
u(t) = C
K
x(t)
(13)
Applying the controller (13) to (12) will result in the
following closed-loop system
x
c
(t) = A
c
x
c
(t) + A
dc
x
c
(t h) + B
1c
w(t)
z
c
(t) = C
c
x
c
(t) + D
c
w(t)
(14)
where
x
c
=
x
x
_ _
, A
c
=
A B
2
C
K
B
K
C
2
A
K
_ _
, A
dc
=
A
d
0
0 0
_ _
B
1c
=
B
1
B
K
D
21
_ _
, C
c
= C
1
D
12
C
K
_ _
, D
c
= D
11
(15)
Then we present a sufcient condition under which there
exists an output-feedback H
1
controller of form (13) for
the closed-loop system (14).
Theorem 3: Dene G =
W
T
11
U
T
I V
11
_ _
. For prescribed
positive scalars g . 0, h . 0, 1 . 0 and scalar l, there exists
a dynamical output-feedback controller such that the closed-
loop system (14) is asymptotically stable and the cost function
2
[0, 1) if there
exist symmetric and positive-denite matrices T . 0, S . 0,
R =
R
1
R
2
R
3
_ _
. 0 and matrices V
11
, V
21
, W
11
, U,

A,

B,
C such that the following LMI holds (see (16))

1T (1, 2) (1, 3) 0 (1, 5) G
(2, 2)
0 0
0 0
hA
d
B
1
0
0 0
(2, 4)
lA
T
d
lA
T
d
V
11
0 0
0 0
0 0
0

g
2
I D
T
11
I
_ _
0
B
T
1
B
T
1
V
11
+ D
T
21

B
T
0 0
_ _
0
(4, 4) h(l + 1)
I V
11
0 V
21
I V
11
0 V
21
0
G G
T
0
(6, 6)
, 0 (16)
www.ietdl.org
where
(1, 2) = h
0 0 W
T
11
(A +(1 +l)A
d
)
T
A
T
d
+

C
T
B
T
2
A
T
d
0
0 0 (A +(1 +l)A
d
)
T
A
T
d
0
_ _
(1, 3) =
0 W
T
11
C
T
1
+

C
T
D
T
12
0 C
T
1
_ _
(see equation at the bottom of the page)
(2, 2) =
S
lhA
T
d
A
T
d
0
0 0
_ _
2hI +hR
3
(2, 4) =
0 0 0 0
0 0 0 0
0 0 h
2
(l +1)A
d
0
0 0 0 0
(4, 4) = h
R
1
R
2
R
3
_ _
, (6, 6) =
2I V
11
V
21
+V
T
21
_ _
+S
After we obtain the solution of the LMI (16), the
corresponding controller gain matrices A
K
, B
K
and C
K
can
be obtained by the following formulas
A
K
= V
T
21
(
A V
T
11
(A +A
d
+lA
d
)W
11

BC
2
W
11
V
T
11
B
2

C)W
1
21
B
K
= V
T
21

B, C
K
=

CW
1
21
omitted here.
Then we design an output-feedback controller with the
form (13), which meets a family of input output
specications. One such set of specications is, for
instance, L
1
T
zw
(s)J
1
1
, g
1
, L
2
T
zw
(s)J
2
1
, g
2
.
Matrices L
i
, J
i
are selection matrices that specify which
channel is involved in the corresponding constraint. With
each channel is associated an LMI constraint of the form
encountered in Theorem 3. The desired characterisation for
output-feedback synthesis with multichannel specications
can be derived in the following three steps: (i) introduce
different Lyapunov variables T
i
, S
i
, R
i
for each channel; (ii)
introduce a variable V common to all channels; (iii)
perform adequate congruence transformations and use
linearising changes of variables to end up with LMI
synthesis condition.
Corollary 3: For prescribed positive scalars g . 0, h . 0,
1 . 0 and scalar l, there exists a dynamical output-
feedback controller such that the closed-loop system (14) is
asymptotically stable and the cost function achieves
J(w) . 0 for all non-zero w [ L
2
[0, 1) if there exist
symmetric and positive-denite matrices T
i
. 0, S
i
. 0,
R
i
. 0 and matrices V
11
, V
21
, W
11
, U,

A,

B,

C satisfy the
LMI (16) for i = 1, . . . , N, where the matrices T
i
, S
i
, R
i
are taken with the index i. If the above conditions hold, the
output-feedback gain can be obtained similarly from
Theorem 3.
Remark 2: In contrast with earlier results, a different
Lyapunov function is employed for each channel. Hence far
better results can generally be expected.
Example 2: Consider the system (12) with multichannel
constraints, where the system matrices are as follows
A =
0 1
0.8 0.6
_ _
, A
d
=
0.1 0.02
0.03 0.1
_ _
,
B
1
=
0 0.2
0.5 0
_ _
, B
2
=
1
1
_ _
C
1
= 1 0
_ _
, C
2
= 1 1
_ _
, D
12
= 2, D
11
= D
21
= 0
Then we design an output-feedback controller with the form
(13) which meets L
1
T
zw
(s)J
1
1
,g
1
, L
2
T
zw
(s)J
2
1
,g
2
.
Let L
1
= 2, L
2
= 2, J
1
=
0.8 1
1 0.2
_ _
, J
2
=
0.6 0.8
0 1
_ _
,
g
1
= 2.
Applying Corollary 3 we obtain, for h 0.5, l 20.9,
1 300, a minimum g
2
of 1.3142. And we can obtain the
gain matrices of the dynamic output-feedback controller
(13) for system (12)
A
K
=
2.6882 2.5190
2.7187 2.5479
_ _
10
6
, B
K
=
5.3824
5.4325
_ _
10
3
C
K
= 693.6611 683.7420
_ _
Then we use the output feedback controller design method
derived directly from the LMI (3) without carrying out the
process similarly to the proof of Theorem 2 (we omit the
design process for simplication). Because it cannot
separate the Lyapunov matrices and controller matrices, it
should introduce the same Lyapunov function for each
channel. We can obtain the minimum of g
2
is 2.0981.
And the corresponding gain matrices of the dynamic
output-feedback controller are as follows
A
K
=
1.9870 1.6742
1.7365 1.7590
_ _
10
7
, B
K
=
2.0871
2.2245
_ _
10
4
C
K
= 879.4389 902.6732
_ _
(1, 5) = T +
W
T
11
(A + A
d
+ lA
d
)
T
+

C
T
B
T
2

A
T
(A + A
d
+ lA
d
)
T
(A + A
d
+ lA
d
)
T
V
11
+ C
T
2

B
T
_ _
+
1
2
1G
www.ietdl.org
It can be seen that the method in Corollary 3 achieves better
H
1
upper bound and lower gain because it introduces a
different Lyapunov function for each channel.
5 Conclusions
A new H
1
performance criterion is proposed based on LMI
approach for a class of time-delay systems with parameter-
dependent Lyapunov variables. It possesses a decoupling
structure which is then used to solve the multichannel
output-feedback control problems and synthesise the
polytopic uncertain systems. Unfortunately, the scalar 1 non-
linearly appears in the LMI conditions. Further results,
therefore, includes how to eliminate the non-linear
inuence. Numerical results show that the proposed method
does provide a further improvement in reducing
conservativeness for time-delay systems with polytopic
uncertainty.
6 Acknowledgment
This work was supported by Central College Basic Research
Fund under Grant N090223001.
7 References
[1] BOYD S., GHAOUI L.E., FERON E., BALAKRISHNAN V.: Linear matrix
inequality in systems and control theory (SIAM,
Philadelphia, 1994)
[2] DE OLIVEIRA M.C., BERNUSSOU J., GEROMEL J.C.: A new discrete-
time robust stability condition, Syst. Control Lett., 1999, 37,
(4), pp. 261265
[3] APKARIAN P., TUAN H.D., BERNUSSOU J.: Continuous-time
analysis, eigenstructure assignment, and H
2
synthesis with
enhanced linear matrix inequalities (LMI) characterizations,
IEEE Trans. Autom. Control, 2001, 46, (12), pp. 19411946
[4] JIA Y.: Alternative proofs for improved LMI
representations for the analysis and the design of
continuous-time systems with polytopic type uncertainty:
a predictive approach, IEEE Trans. Autom. Control, 2003,
48, (8), pp. 14131416
[5] OLIVEIRA R.C.L.F., PERES P.L.D.: A convex optimization
procedure to compute H
2
and H
1
norms for uncertain
linear systems in polytopic domains, Optim. Control Appl.
Method, 2008, 29, pp. 295312
[6] HE Y., WU M., SHE J.: Improved bounded-real-lemma
representation and H
1
control of systems with polytopic
uncertainties, IEEE Trans. Circuits Syst., 2005, 52, (7),
pp. 380383
[7] OLIVEIRA R.C.L.F., PERES P.L.D.: Stability of polytopes of
matrices via afne parameter-dependent Lyapunov
functions: asymptotically exact LMI conditions, Linear
Algebra Appl., 2005, 405, pp. 209228
[8] RICHARD J.: Time-delay systems: an overview of some
recent advances and open problem, Automatica, 2003,
39, (10), pp. 16671694
[9] BASER U., KIZILSAC B.: Dynamic output feedback H
1
control problem for linear neutral systems, IEEE Trans.
Autom. Control, 2007, 52, (6), pp. 11131118
[10] LIN P., JIA Y., LI L.: Distributed robust H
1
consensus
control in directed networks of agents with time-delay,
Syst. Control Lett., 2008, 57, (8), pp. 643653
[11] PARLAKCI M.N.A.: Improved robust stability criteria and
design of robust stabilizing controller for uncertain linear
time-delay systems, Int. J. Robust Nonlinear Control,
2006, 16, (13), pp. 599636
[12] PARLAKCI M.N.A.: An LMI approach to static output
feedback stabilization problem for linear uncertain
systems. Proc. Ninth IASTED International Conf. Control
and Applications, 2007, pp. 249254
[13] SAADNI M.S., MEHDI D.: Stability and stabilizability of a class
of state retarded uncertain polytopic systems. IEEE Int. Conf.
Systems, Man and Cybernetics, 2004, pp. 42444248
[14] FRIDMAN E., SHAKED U.: An improved stabilization method
for linear time-delay systems, IEEE Trans. Autom. Control,
2002, 47, (11), pp. 19311937
[15] FRIDMAN E., SHAKED U.: A descriptor system approach to
H
1
control of linear time-delay systems, IEEE Trans.
Autom. Control, 2002, 47, (2), pp. 253270
[16] PARK P.: A delay-dependent stability criterion for
systems with uncertain time-invariant delays, IEEE Trans.
Autom. Control, 1999, 44, (4), pp. 876877
8 Appendix
8.1 Proof of Theorem 1
Proof: Consider a LyapunovKrasovskii functional
candidate V(t) as follows
V(t) = V
1
(t) + V
2
(t) + V
3
(t) (17)
where
V
1
(t) = x
T
(t)Px(t)
V
2
(t) =
_
t
th
x
T
(a)Sx(a) da
V
3
(t) =
_
0
h
_
t
t+u
x
T
(a)A
T
d
R
3
A
d
x(a) dadu
www.ietdl.org
Note the following identity (LeibnizNewton):
_
b
a
v(t) dt = v(b) v(a). Then the time derivative of V(t)
along the solution of (1) is given by

V(t) =

V
1
(t)
+

V
2
(t) +

V
3
(t), where
V
1
(t) = 2x
T
(t)P (A + A
d
)x(t) A
d
_
t
th
x(a) da + Bw(t)
_ _
= 2x
T
(t)P(A + A
d
)x(t) 2
_
t
th
x
T
(t)PA
d
x(a) da
+ 2x
T
(t)PBw(t)
V
2
(t) = x
T
(t)Sx(t) x
T
(t h)Sx(t h)
V
3
(t) =
_
t
th
x
T
(a)A
T
d
R
3
A
d
x(a) da + h x
T
(t)A
T
d
R
3
A
d
x(t)
Let
b = R
3
A
d
x(t), j
T
=
x
T
(t) x
T
(t h) b
T
_ _
R =
R
1
R
2
R
3
_ _
. 0, d =
P 0 0
P 0 hA
T
d
_ _
.
In order to obtain an LMI we have to restrict ourselves to the
case of M lI in Lemma 1, where l , R is a scalar
parameter. And let a(a) =
0 A
T
d
_ _
T
x(a), b(a) = dj in
Lemma 1. By Lemma 1, we can obtain
2
_
t
th
x
T
(t)PA
d
x(a) da
= 2
_
t
th
j
T
d
T
0
A
d
_ _
x(a) da + 2h
_
t
th
b
T
A
d
A
d
x(a) da
_
t
th
x
T
(a)
0 A
T
d
_ _
R
0
A
d
_ _
x(a) da
+ 2l
_
t
th
j
T
d
T
0
A
d
_ _
x(a) da
+ (l + 1)
2
_
t
th
j
T
d
T
R
1
dj da
+ 2h
_
t
th
b
T
A
d
A
d
x(a) da
=
_
t
th
x
T
(a)A
T
d
R
3
A
d
x(a) da
+ 2l
_
t
th
x
T
(t)PA
d
x(a) da
+ 2(l + 1)h
_
t
th
b
T
A
d
A
d
x(a) da
+ (l + 1)
2
_
t
th
j
T
d
T
R
1
dj da
Note that
2(l + 1)h
_
t
th
b
T
A
d
A
d
x(a) da + h x
T
(t)A
T
d
R
3
A
d
x(t)
= 2hb
T
A
d
(A + (1 + l)A
d
)x(t) 2lhb
T
A
d
A
d
x(t h)
hb
T
R
1
3
b + 2hb
T
A
d
Bw(t)
Then substituting
_
t
th
x(a) = x(t) x(t h) and
A
d
x(t h) = x(t) Ax(t) Bw(t) into the above
equation, we obtain
dV(t)
dt
+ z
T
(t)z(t) g
2
w
T
(t)w(t) = r
T
Vr + z
T
(t)z(t)
(18)
where
r
T
=
j
T
w
T
(t)
_ _
V=
C
1
lPA
d
h(A+A
d
+lA
d
)
T
A
T
d
S hlA
T
d
A
T
d
hR
1
3
+V
1
PB
0
hA
d
B
g
2
I
V
1
=h(l+1)
2
d
T
R
1
d
C
1
=Sym{P(A+A
d
+lA
d
)}+S
Integrate (18) in t from 0 to 1. Because
V(0) =0, V(1) 0 and
_
1
0
z
T
(t)z(t) =
_
1
0
(x
T
(t)C
T
+
w
T
(t)D
T
)(Cx(t)+Dw(t)). From Schur complement
Lemma, we can obtain that J(w), 0 (and

V ,0) if the
following LMI holds (see equation at the bottom of the
page)
Then we can obtain the LMI (3) by Schur complement
lemma. A
C
1
lPA
d
h(A + A
d
+ lA
d
)
T
A
T
d
PB C
T
h(l + 1)P h(l + 1)P
S hlA
T
d
A
T
d
0 0 0 0
hR
1
3
hA
d
B 0 0 h
2
(l + 1)A
d
g
2
I D
T
0 0
I 0 0
hR
1
hR
2
hR
3
, 0
www.ietdl.org
Proof: Applying Corollary 1 to the closed-loop system (7),
we can obtain (see (19))
where
(1, 1) = Sym{(A
c
+ A
d
+ lA
d
)Q}
(1, 3) = hQ(A
c
+ (1 + l)A
d
)
T
A
T
d
It is obvious that there exists a positive scalar 1 such that the
following equation holds (see (20))
Note that the LMI (19) can be rewritten as
P + ab
T
+ ba
T
, 0 (21)
where
a = Q 0 0 0 0 0 0 0
_ _
T
b =
(A
c
+ A
d
+ lA
d
)
T
+
1
2
1I 0
_
h(A
c
+ A
d
+ lA
d
)
T
A
T
d
0C
T
c
00I
_
T
Substitute (8) into (21) and dene U = KV. From
Lemma 2, we can obtain that the LMIs (20) and (21) are
equivalent to the LMI (9). A
Proof: Apply Theorem 1 to the closed-loop system (14) and
carry out the process similar to the proof of Theorem 2, then
we can obtain (see (22))
where
(1, 3) = h(A
c
+ (1 + l)A
dc
)
T
A
T
dc
R
3
(1, 8) = P + (A
c
+ (1 + l)A
dc
)
T
V +
1
2
1V
(3, 7) = h
2
(1 + l)R
3
A
dc
Partition V and its inverse V
1
in the LMI (22) as
V =
V
11
V
12
V
21
V
22
_ _
, W = V
1
=
W
11
W
12
W
21
W
22
_ _
(23)
From VW I, dene
F
1
=
W
11
I
W
21
0
_ _
, F
2
=
I V
11
0 V
21
_ _
(24)
(1, 1) lA
d

S (1, 3) B QC
T
c
h(l + 1)
R
3
h(l + 1)
R
3
Q

S lh
SA
T
d
A
T
d
0 0 0 0 0
h
R
3
hA
d
B 0 0 h
2
(l + 1)A
d

R
3
0
g
2
I D
T
11
0 0 0
I 0 0 0
h
R
1
h
R
2
0
h
R
3
0

, 0 (19)
P =
1Q lA
d

S 0 B 0 h(l + 1)
R
3
h(l + 1)
R
3
0

S lh
SA
T
d
A
T
d
0 0 0 0 0
h
R
3
hA
d
B 0 0 h
2
(l + 1)A
d

R
3
0
g
2
I D
T
11
0 0 0
I 0 0 0
h
R
1
h
R
2
0
h
R
3
0

, 0 (20)
1P 0 (1, 3) 0 C
T
c
0 0 (1, 8) I
S lhA
T
dc
A
T
dc
R
3
0 0 0 0 lA
T
dc
V 0
hR
3
hR
3
A
dc
B
1c
0 0 (3, 7) 0 0
g
2
I D
T
11
0 0 B
T
1c
V 0
I 0 0 0 0
hR
1
hR
2
h(l + 1)V 0
hR
3
h(l + 1)V 0
V V
T
0
S
1
, 0 (22)
www.ietdl.org
And also we dene U = V
T
21
W
21
+ V
T
11
W
11
. It is always
possible to nd the invertible matrices V
21
and W
21
satisfying
V
T
21
W
21
= U V
T
11
W
11
. This is veried by assuring
the invertibility of U V
T
11
W
11
, which is proved below.
After performing the linearising congruence
transformations and change of variables, we can obtain
F
T
1
(V
T
+ V)F
1
= F
T
1
V
T
F
1
+ F
T
1
VF
1
=
W
11
+ W
T
11
I + U
T
V
11
+ V
T
11
_ _
. 0
The above inequality ensures the invertibility of the three
matrices F
T
1
V
T
F
1
, W
11
and V
11
. With W
11
, we can dene
the invertible matrix P =
I I
0 W
11
_ _
. The product of
the two matrices F
T
1
VF
1
and P leads to
F
T
1
V
T
F
1
P =
W
11
I
U V
T
11
_ _
I I
0 W
11
_ _
=
W
11
0
U U V
T
11
W
11
_ _
(25)
which assures that U V
T
11
W
11
is invertible.
Perform the following linearising changes of variables:
A = V
T
11
(A + A
d
+ lA
d
)W
11
+ V
T
21
B
K
C
2
W
11
+ V
T
11
B
2
C
K
W
21
+ V
T
21
A
K
W
21
B = V
T
21
B
K
,

C = C
K
W
21
, T = F
T
1
PF
1
(26)
Pre- and post-multiply the LMI (22) by
diag{F
T
1
, I , R
1
3
, I , I , I , I , F
T
1
, F
T
1
V
T
} and its inverse,
respectively, substitute (15), (23), (24) into the obtained
equation. Obviously, F
T
1
V
T
S
1
VF
1
F
T
1
V
T
VF
1
+
S, R
1
2I + R. Using the above equations, the LMI
(16) can be derived.
After we obtain the solution of the LMI (16), the
corresponding controller of form (13) will be constructed as
follows: (i) compute W
21
from U V
T
11
W
11
; (ii) utilising
the matrices A
j
, B
j
, C
j
, V
11
, W
11
, V
21
and W
21
obtained
above, compute the controller gain by reversing (26). A
www.ietdl.org

Uploaded by

Copyright:

Available Formats

Uploaded by

Copyright:

Available Formats

Published in IET Control Theory and Applications

Received on 13th September 2008

C such that the following LMI holds (see (16))

You might also like