Information Sciences 179 (2009) 3041–3058
Contents lists available at ScienceDirect
Information Sciences
journal homepage: www.elsevier .com/locate / ins
H1 control design for fuzzy discrete-time singularly perturbed systemsvia slow state variables feedback: An LMI-based approach
Jiuxiang Dong, Guang-Hong Yang *
College of Information Science and Engineering, Northeastern University, Shenyang 110004, PR ChinaKey Laboratory of Integrated Automation of Process Industry (Ministry of Education), Northeastern University, Shenyang 110004, PR China
a r t i c l e i n f o
Article history:Received 15 May 2008Received in revised form 6 February 2009Accepted 16 March 2009
Keywords:Takagi–Sugeno (T–S) fuzzy systemsDiscrete-time systemsNonlinear singularly perturbed systemsH1 controlLinear matrix inequalitiesReduced-order control
0020-0255/$ - see front matter � 2009 Elsevier Incdoi:10.1016/j.ins.2009.03.012
* Corresponding author. Address: College of Info13504182968; fax: +86 24 83681939.
E-mail addresses: [email protected](G.-H. Yang).
a b s t r a c t
This paper addresses the H1 control problem via slow state variables feedback for discrete-time fuzzy singularly perturbed systems. At first, a method of evaluating the upper boundof singular perturbation parameter � with meeting a prescribed H1 performance boundrequirement is given. Subsequently, two methods for designing H1 controllers via slowstate variables feedback are presented in terms of solutions to a set of linear matrixinequalities (LMIs). In particular, one of them can be used to improve the upper boundof the singular perturbation parameter �. Finally, two numerical examples are given toillustrate the effectiveness of the proposed methods.
� 2009 Elsevier Inc. All rights reserved.
1. Introduction
Slow and fast dynamic phenomena in control systems often occur due to the presence of small ‘‘parasitic” parameters,such as motor control systems, electronic circuits, magnetic-ball suspension systems, and so on. In a state space framework,such systems are commonly modeled by a mathematical description of singular perturbations, where a small parameter � isexploited to determine the degree of separation between slow and fast parts of the dynamical system. Due to the very smallsingular perturbation parameter �, the analysis and synthesis approaches for normal systems often lead to ill-conditionedresults. Therefore, a so-called reduction technique with a two-step design methodology [17] is widely adopted for overcom-ing the difficulty. Firstly, through the separate stabilization of two lower dimensional subsystems in two different timescales, a composite stabilizing controller is synthesized from separate stabilizing controllers of the two subsystems, wherethe controller could be determined without the knowledge of the small singular perturbation parameter. In the past severaldecades, many control problems of singularly perturbed systems have attracted considerable attentions, see the survey pa-per [23] and the references therein.
In control theory, a well-known convex optimization technique, i.e., linear matrix inequality (LMI) technique, has beenextensively exploited to solve control problems [3]. In contrast to Riccati approaches, linear matrix inequalities (LMIs)can be formulated as convex optimization problems that are amenable to compute solution and can be solved effectively[3]. Another good feature of LMIs is their ability of adding constraints to the parametrical optimization problem provided
. All rights reserved.
rmation Science and Engineering, Northeastern University, Shenyang 110004, PR China. Tel.: +86
, [email protected] (J. Dong), [email protected], [email protected]
3042 J. Dong, G.-H. Yang / Information Sciences 179 (2009) 3041–3058
they are themselves linear with respect to unknowns [12]. In particular, for H1 synthesis, it has the merit of eliminating theregularity restrictions attached to the Riccati-based solutions [11]. Motivated by the merits of the LMI formulations, someLMI-based controller design approaches for singularly perturbed systems have been developed in [7–10,28] recently.
On the other hand, there has been a great deal of interest in using Takagi–Sugeno (T–S) fuzzy models to approximate non-linear systems, and many control problems of nonlinear systems have been widely studied based on T–S fuzzy systems, see[14,15,26,29,30] and the references therein. In particular, the controller design conditions for the state feedback [19], staticoutput feedback [21], and dynamic output feedback [2] cases are exploited in terms of solutions of LMIs for fuzzy singularlyperturbed systems. Moreover, the methods of designing H1 state feedback controllers with pole placement constraints aregiven in [1]. In most cases, the ‘‘fast” dynamics of singularly perturbed systems are not adequately modeled and are, therefore,neglected in order to simplify the design [24,6]. In practice the fast variables are not directly measurable sometimes, for exam-ple, the flexible variables (modeled as fast variables) of the flexible link manipulators [22]. Therefore, the study on the prob-lem of designing state feedback controllers by only using slow state variables of singularly perturbed systems is necessary.
Because most of the existing synthesis techniques for singularly perturbed systems are independent of � for avoiding toobtain ill-conditioned results, it is of great importance to find the bound of � for ensuring the stability of the closed-loopsystems. As a result, it has attracted increasing interest in the past several decades. In [5], an approach to characterizeand compute the stability bound is presented for continuous-time singularly perturbed systems. By considering criticalstability criteria with a bialternate product, systematic approaches to determine the exact stability bound of discrete-timesingularly perturbed systems are given in [13,18]. Moreover, an algorithm for finding the upper bound of the singular per-turbation parameter for D-stability is presented in [16]. However, by the authors’ knowledge, the topic of evaluating theupper bounds of singular perturbation parameters for nonlinear discrete-time singularly perturbed systems with meetingH1 performance requirements has not been studied. In this paper, the topic will be addressed by using the two lemmas thatare proposed for linear singularly perturbed systems in [8], which are given in Appendices A and B.
In this paper, a method of evaluating the upper bound of the singular perturbation parameter � for discrete-time fuzzysingularly perturbed systems with meeting a prescribed H1 performance bound requirement is given. Furthermore, twoH1 controller design methods via slow state variables feedback are presented in terms of solutions to a set of LMIs. In contrastto the conventional design methods [23], the new design methods are with twofold advantages. One is that the two designmethods are based on LMIs, which can eliminate the regularity restrictions attached to the Riccati-based solution. The otheris that one of the two methods can be used to improve the upper bound of singularly perturbed parameter � at the stage ofcontrol design, which implies that the tradeoff between the H1 performance index and the upper bound of the singularperturbation parameter � is considered in the design. Thus, the new controller design method can overcome the disadvantagethat the allowable upper bound of the singular perturbation parameter of the closed-loop system with the controllersdesigned by the existing ones is too small to be used. This paper is organized as follows. Section 2 presents system descriptionand some preliminaries. In Section 3, a sufficient condition is derived for evaluating the upper bound �� of � subject to a pre-scribed H1 performance constraint. Moreover, two new LMI-based H1 controller design methods are presented. In particular,one of them can improve the upper bound of the singular perturbation parameter � by designing controllers. The validity ofthese approaches is illustrated by two numerical examples in Section 4. Finally, Section 5 concludes the paper.
Notation: Rn denotes the set which consists of real n-vectors (n� 1 matrices). For a symmetric block matrix, (*) is used forthe blocks induced by symmetry, for example,
M11 � �M21 M22 �M31 M32 M33
264375 ¼ M11 MT
21 MT31
M21 M22 MT32
M31 M32 M33
264375
The superscript T stands for matrix transposition and the notation M�T denotes the transpose of the inverse matrix of M.
2. System description and some preliminaries
A class of nonlinear singularly perturbed systems under consideration are described by the following fuzzy system model:
Plant Rule i :
IF v1ðkÞ is Mi1 and v2ðkÞ is Mi2; . . . ; vpðkÞ is Mip; THEN
x1ðkþ 1Þ ¼ Ai11x1ðkÞ þ �Ai
12x2ðkÞ þ Biw1wðkÞ þ Bi
u1uðkÞx2ðkþ 1Þ ¼ Ai
21x1ðkÞ þ �Ai22x2ðkÞ þ Bi
w2wðkÞ þ Biu2uðkÞ
zðkÞ ¼ Ciz1x1ðkÞ þ �Ci
z2x2ðkÞ þ DizwwðkÞ þ Di
zuuðkÞ
ð1Þ
where i ¼ 1;2; . . . ; r, r is the number of IF–THEN rules. Mil, 1 6 i 6 r, 1 6 l 6 p are fuzzy sets. v iðkÞ are the premise variables,x1ðkÞ 2 Rn1 and x2ðkÞ 2 Rn2 are respectively the slow and fast state vectors, uðkÞ 2 Rnu is the control input, wðkÞ 2 Rnw is thedisturbance, zðkÞ 2 Rnz is the controlled output, the matrices Ai
11, Ai12, Ai
21, Ai22, Bi
w1, Biw2, Bi
u1, Biu2, Ci
z1, Ciz2, Di
zw and Dizu are of
appropriate dimensions. � > 0 is a singular perturbation parameter, which determines the degree of separation betweenthe ‘‘slow” and ‘‘fast” modes of the system [1].
J. Dong, G.-H. Yang / Information Sciences 179 (2009) 3041–3058 3043
Denote
-iðvðkÞÞ ¼Yp
j¼1
lijðv jðkÞÞ
lijðv jðkÞÞ is the grade of membership of v jðkÞ in Mij, where it is assumed that
Xri¼1
-iðvðkÞÞ > 0; -iðvðkÞÞP 0 i ¼ 1;2; . . . ; r
Let aiðvðkÞÞ ¼-iðvðkÞÞPri¼1-iðvðkÞÞ
, then
0 6 aiðvðkÞÞ 6 1; andXr
i¼1
aiðvðkÞÞ ¼ 1 ð2Þ
aiðvðkÞÞ, i ¼ 1; . . . ; r are said to be normalized membership functions. Then, the T–S fuzzy model of (1) is inferred as follows:
x1ðkþ 1Þ ¼Xr
i¼1
aiðvðkÞÞ Ai11x1ðkÞ þ �Ai
12x2ðkÞ þ Biw1wðkÞ þ Bi
u1uðkÞ� �
x2ðkþ 1Þ ¼Xr
i¼1
aiðvðkÞÞ Ai21x1ðkÞ þ �Ai
22x2ðkÞ þ Biw2wðkÞ þ Bi
u2uðkÞ� �
zðkÞ ¼Xr
i¼1
aiðvðkÞÞ Ciz1x1ðkÞ þ �Ci
z2x2ðkÞ þ DizwwðkÞ þ Di
zuuðkÞ� �
ð3Þ
which can be rewritten as follows:
xðkþ 1Þ ¼ AðaðkÞÞE�xðkÞ þ BwðaðkÞÞwðkÞ þ BuðaðkÞÞuðkÞzðkÞ ¼ CzðaðkÞÞE�xðkÞ þ DzwðaðkÞÞwðkÞ þ DzuðaðkÞÞuðkÞ
ð4Þ
where
AðaðkÞÞ ¼Xr
i¼1
aiðvðkÞÞAi; BwðaðkÞÞ ¼
Xr
i¼1
aiðvðkÞÞBiw;
BuðaðkÞÞ ¼Xr
i¼1
aiðvðkÞÞBiu; CzðaðkÞÞ ¼
Xr
i¼1
aiðvðkÞÞCiz;
DzwðaðkÞÞ ¼Xr
i¼1
aiðvðkÞÞDizw; DzuðaðkÞÞ ¼
Xr
i¼1
aiðvðkÞÞDizu
xðkÞ ¼x1ðkÞx2ðkÞ
� �E� ¼
In1�n1 00 �In2�n2
� �Ai ¼
Ai11 Ai
12
Ai21 Ai
22
" #Bi
w ¼Bi
w1
Biw2
" #
Biu ¼
Biu1
Biu2
" #Ci
z ¼ ½Ciz1Ci
z2�
ð5Þ
In this paper, the concept of parallel distributed compensation (PDC) is used to design fuzzy controllers, i.e., the designedfuzzy controller shares the same fuzzy sets with the fuzzy model in the premise parts. The more details can be found in[25]. For the fuzzy model (1), the following slow state feedback controller is adopted, where Ki
a, 1 6 i 6 r are the parametersto be designed.
Control Rule i :
IF v1ðkÞ is Mi1 and v2ðkÞ is Mi2; . . . ;vpðkÞ is Mip
THEN uðkÞ ¼ Kiax1ðkÞ ð6Þ
Because the control rules are the same as the plant rules, the fuzzy controller can be obtained as follows:
uðkÞ ¼Xr
i¼1
aiðvðkÞÞKiax1ðkÞ ¼
Xr
i¼1
aiðvðkÞÞKi x1ðkÞx2ðkÞ
� �ð7Þ
where
Ki ¼ Kia 0
� �ð8Þ
3044 J. Dong, G.-H. Yang / Information Sciences 179 (2009) 3041–3058
Combining (7) and (4), then the resulting closed-loop system is given as follows:
xðkþ 1Þ ¼ AðaðkÞÞE� þ BuðaðkÞÞKðaðkÞÞð ÞxðkÞ þ BwðaðkÞÞwðkÞzðkÞ ¼ CzðaðkÞÞE� þ DzuðaðkÞÞKðaðkÞÞð ÞxðkÞ þ DzwðaðkÞÞwðkÞ
ð9Þ
where
KðaðkÞÞ ¼Xr
i¼1
aiðvðkÞÞ Kia 0
� �ð10Þ
The replacement of �x2 by n2 in system (9) will result in the equivalent system
nðkþ 1Þ ¼ E� AðaðkÞÞ þ BuðaðkÞÞKðaðkÞÞð ÞnðkÞ þ E�BwðaðkÞÞwðkÞzðkÞ ¼ CzðaðkÞÞ þ DzuðaðkÞÞKðaðkÞÞð ÞnðkÞ þ DzwðaðkÞÞwðkÞ
ð11Þ
where
nðkÞ ¼x1ðkÞn2ðkÞ
� �
The H1 norm in [20] for nonlinear discrete-time systems is applicable for nonlinear discrete-time singularly perturbed sys-tems. The definition is given as follows:Definition 1 [20]. Given a real number c > 0, it is said that the H1 norm of the closed-loop system (11) is less than or equalto c (i.e., the exogenous signals are locally attenuated by c) if there exists a neighborhood U of x ¼ 0 such that for everypositive integer N and for every w 2 l2ð½0;NÞ;Rnw Þ for which the state trajectory of the closed-loop system (11) startingxð0Þ ¼ 0 remains in U for all k 2 ½0;N�, the response z 2 l2ð½0;N�;Rnz Þ of (11) satisfies
XNi¼0
kzkk26 c2
XN
i¼0
kwkk2; for all N
In this paper, the following problems will be addressed.
2.1. Evaluation of the upper bound of � with meeting stability and H1 performance bound requirement
Let c > 0 be a given constant and the gains Ki be given. Find an �� > 0 as big as possible such that the system (1) with (7) isasymptotically stable and its H1-norm is less than or equal to c for any � 2 ð0; ���.
2.2. H1 controller designs without the consideration of improving the upper bound of �
Let c > 0 be a given constant. Find gains Ki ði ¼ 1; . . . ; rÞ, and there exists a positive scalar ��, such that the system (1) with(7) is asymptotically stable and its H1-norm is less than or equal to c for any � 2 ð0; ���.
2.3. H1 controller design with the consideration of improving the upper bound of �
Let c > 0 be a given constant and �� > 0 be a prescribed upper bound of the singular perturbation parameter �. Find gainsKi
a ði ¼ 1; . . . ; rÞ and an �� > 0 such that the system (1) with (7) is asymptotically stable and its H1-norm is less than or equalto c for any � 2 ð0; ���.
The following lemmas will be used in this sequel.
Lemma 2. If there exists a symmetric positive-definite matrix PðaðkÞÞ such that the following LMIs hold,
U11ðk; kþ 1Þ �U21ðk; kþ 1Þ U22ðk; kþ 1Þ
� �< 0; f or � 2 ð0; ��� ð12Þ
where AðaðkÞÞ, BwðaðkÞÞ, BuðaðkÞÞ, CzðaðkÞÞ, DzwðaðkÞÞ and DzuðaðkÞÞ are the same as in (5), KðaðkÞÞ is the same as in (10), and
U11ðk; kþ 1Þ ¼ �PðaðkÞÞ þ AðaðkÞÞ þ BuðaðkÞÞKðaðkÞÞð ÞT
� E�Pðaðkþ 1ÞÞE� AðaðkÞÞ þ BuðaðkÞÞKðaðkÞÞð Þ
þ 1c
CzðaðkÞÞ þ DzuðaðkÞÞKðaðkÞÞð ÞT
� CzðaðkÞÞ þ DzuðaðkÞÞKðaðkÞÞð ÞU21ðk; kþ 1Þ ¼ BT
wðaðkÞÞE�Pðaðkþ 1ÞÞE� AðaðkÞÞ þ BuðaðkÞÞKðaðkÞÞð Þ
þ 1c
DTzwðaðkÞÞ CzðaðkÞÞ þ DzuðaðkÞÞKðaðkÞÞð Þ
U22ðk; kþ 1Þ ¼ BTwðaðkÞÞE�Pðaðkþ 1ÞÞE�BwðaðkÞÞ þ
1c
DTzwðaðkÞÞDzwðaðkÞÞ � cI
ð13Þ
J. Dong, G.-H. Yang / Information Sciences 179 (2009) 3041–3058 3045
Then for each singular perturbation parameter � 2 ð0; ���, the closed-loop system (11) is asymptotically stable and its H1 norm isless than or equal to c.
Proof. See Appendix A. h
Lemma 3 [8]. For a given positive scalar ��, if the following conditions are satisfied,
a P 0 ð14Þa��2 þ b�� þ c < 0 ð15Þc < 0 ð16Þ
where a, b and c are constants, then
a�2 þ b�þ c < 0; f or � 2 ½0; ��� ð17Þ
Proof. See Appendix B. h
Lemma 4 [8]. For a given positive scalar �� and matrices T1, T2, T3, if following conditions are satisfied,
T1 P 0 ð18Þ��2T1 þ ��T2 þ T3 < 0 ð19ÞT3 < 0 ð20Þ
then
�2T1 þ �T2 þ T3 < 0; f or � 2 ½0; ��� ð21Þ
Proof. See Appendix C. h
3. Main results
In this section, a method of evaluating the upper bound of singularly perturbed parameter � subject to the stability of theclosed-loop system with meeting H1 performance bound requirements is presented. Moreover, two sufficient conditions fordesigning H1 controllers are given. In particular, one of them can improve the upper bound of the singular perturbationparameter � by designing controllers.
3.1. Computation of stability bound of � subject to an H1 performance bound constraint
Firstly, the following preliminary lemma is needed.
Lemma 5. If there exists a symmetric matrix
QðaðkÞÞ ¼ Q 11ðaðkÞÞ QT21ðaðkÞÞ
Q 21ðaðkÞÞ Q22ðaðkÞÞ
" #;
such that the following inequalities hold, " #
�20 00 Q 22ðaðkÞÞ
� �� � �
0 0 � �0 0 0 �0 0 0 0
2666664
3777775þ �0 Q T
21ðaðkÞÞQ21ðaðkÞÞ 0
� � �
0 0 � �0 0 0 �0 0 0 0
26666664
37777775
þ
W11ðk; kþ 1Þ � � �0 �cI � �
W31ðk; kþ 1Þ BwðaðkÞÞ �Qðaðkþ 1ÞÞÞ �W41ðk; kþ 1Þ DzwðaðkÞÞ 0 �cI
2666437775 < 0; f or � 2 ð0; ��� ð22Þ
where � �
W11ðkÞ ¼Q11ðaðkÞÞ 00 0
� SðaðkÞ;aðkþ 1ÞÞ � STðaðkÞ;aðkþ 1ÞÞ
W31ðkÞ ¼ AðaðkÞÞ þ BuðaðkÞÞKðaðkÞÞð ÞSðaðkÞ;aðkþ 1ÞÞW41ðkÞ ¼ CzðaðkÞÞ þ DzuðaðkÞÞKðaðkÞÞð ÞSðaðkÞ;aðkþ 1ÞÞ
ð23Þ
then for each singular perturbation parameter � 2 ð0; ���, the system (9) is asymptotically stable and its H1 norm is less than orequal to c.
3046 J. Dong, G.-H. Yang / Information Sciences 179 (2009) 3041–3058
Proof. See Appendix D. h
Based on Lemma 5, a method of evaluating the upper bound of the singular perturbation parameter � with a prescribedH1 performance bound constraint is given in the following theorem.
Theorem 6. For a given positive scalar ��, if there exist matrices Q i ¼ ðQiÞT , Sij, 1 6 i; j 6 r
Q i ¼Q i
11 ðQ i21Þ
T
Q i21 Q i
22
" #; Sij ¼
S11 0Sij
21 Sij22
� �
satisfying the following LMIsMiil < 0; 1 6 i; l 6 r ð24aÞ1
r � 1Miil þ
12ðMijl þMjilÞ < 0; 1 6 i – j 6 r; 1 6 l 6 r ð24bÞ
Kiil < 0; 1 6 i; l 6 r ð25aÞ1
r � 1Kiil þ
12ðKijl þKjilÞ < 0; 1 6 i – j 6 r; 1 6 l 6 r ð25bÞ
where
Mijl ¼
Q i11 0
0 0
" #� Sil � ðSilÞT � � �
0 �cI � �
Ai þ BiuKj
� �Sjl Bi
w �Ql �
Ciz þ Di
zuKj� �
Sjl Dizw 0 �cI
266666666664
377777777775
Kijl ¼ ��2
0 0
0 Q i22
" #0 0 0
0 0 0 0
0 0 0 0
0 0 0 0
266666664
377777775þ ��
0 ðQ i21Þ
T
Q i21 0
" #0 0 0
0 0 0 0
0 0 0 0
0 0 0 0
2666666664
3777777775
þ
Q i11 0
0 0
" #� Sil � ðSilÞT � � �
0 �cI � �
Ai þ BiuKj
� �Sjl Bi
w �Ql �
Ciz þ Di
zuKj� �
Sjl Dizw 0 �cI
266666666664
377777777775
then for each singular perturbation parameter � 2 ð0; ���, the system (9) is asymptotically stable and its H1 norm is less than orequal to c.Proof. See Appendix E. h
Remark 7. Theorem 6 presents a method of estimating the upper bound of singularly perturbed parameter � subject to thestability of the closed-loop system (9) while satisfying an H1 performance bound requirement. An upper bound of � can beobtained by solving the following optimization problem:
Minimize �� subject to ð24Þ and ð25Þ
which can be effectively solved by using the LMI Control Toolbox [11].
3.2. H1 controller design
In this subsection, two LMI-based methods of designing H1 controllers are given. The two methods are with the merit ofeliminating the regularity restrictions attached to the Riccati-based solutions. In particular, one of them can improve theupper bound of � by designing controllers.
J. Dong, G.-H. Yang / Information Sciences 179 (2009) 3041–3058 3047
3.2.1. Design without the consideration of improving the upper bound of �In the following, an LMI-based design method without considering the improvement of the upper bound of the singular
perturbation parameter � is given.
Theorem 8. If there exist matrices Qi ¼ ðQiÞT , Sil, Li, 1 6 i; l 6 r, with
Q i ¼Q i
11 ðQ i21Þ
T
Q i21 Q i
22
" #; Sil ¼
S11 0Sil
21 Sil22
� �; Li ¼ Li
a 0� �
ð26Þ
satisfying the following LMIs,
Caiil < 0; 1 6 i; l 6 r ð27aÞ1
r � 1Caiil þ
12ðCaijl þ CajilÞ < 0; 1 6 i – j 6 r; 1 6 l 6 r ð27bÞ
where
Caijl ¼
Q i11 0
0 0
" #� Sil � ðSilÞT � � �
0 �cI � �AiSjl þ Bi
uLj Biw �Q l �
CizSjl þ Di
zuLj Dizw 0 �cI
266666664
377777775 < 0; f or 1 6 i; j; l 6 r
then there exists a sufficient small �� > 0 such that for � 2 ð0; ���, the closed-loop system (9) with
Kia ¼ Li
aS�111 ; 1 6 i 6 r ð28Þ
is asymptotically stable and its H1 norm is less than or equal to c.
Proof. The proof is easily obtained from Theorem 6 and omitted. h
Remark 9. Theorem 8 presents a sufficient condition for designing H1 controllers for discrete-time fuzzy singularly per-turbed systems. The method is based on LMIs, and with the merit of eliminating the regularity restrictions attached tothe Riccati-based solutions. Example 12 in Section 4 will illustrate the effectiveness of the method. However, in the method,the issue of improving the upper bound �� of the singular perturbation parameter � is not addressed. As a result, the obtainedcontroller might give a very small stability bound so that the resulting closed-loop system is unstable for a practical singularperturbation parameter �, see Example 14 in Section 4. In order to overcome the difficulty, another new method with theconsideration of improving the upper bound of the singular perturbation parameter � while satisfying H1 performance con-straints will be proposed in the next subsection.
3.2.2. Design with the consideration of improving the upper bound of �In this subsection, a new LMI-based H1 controller design method with the consideration of improving the upper bound of
the singular perturbation parameter � is given as follows:
Theorem 10. For a given positive scalar ��, if there exist matrices Q i ¼ ðQiÞT , Sil, Li, 1 6 i; l 6 r, with
Q i ¼Q i
11 ðQ i21Þ
T
Q i21 Q i
22
" #; Sil ¼
S11 0Sil
21 Sil22
� �; Li ¼ Li
a 0� �
satisfying (27b) and the following LMIs,
Cbiil < 0; 1 6 i; l 6 r ð29aÞ1
r � 1Cbiil þ
12ðCbijl þ CbjilÞ < 0; 1 6 i – j 6 r; 1 6 l 6 r ð29bÞ
where
Cbijl ¼ ��2
0 00 Q i
22
� �0 0 0
0 0 0 00 0 0 00 0 0 0
2666664
3777775þ ��0 ðQi
21ÞT
Q i21 0
" #0 0 0
0 0 0 00 0 0 00 0 0 0
26666664
37777775þQ i
11 00 0
" #� Sil � ðSilÞT � � �
0 �cI � �AiSjl þ Bi
uLj Biw �Q l �
CizSjl þ Di
zuLj Dizw 0 �cI
266666664
377777775
Then for � 2 ð0; ���, the closed-loop system (9) with (28) is asymptotically stable and its H1 norm is less than or equal to c.3048 J. Dong, G.-H. Yang / Information Sciences 179 (2009) 3041–3058
Proof. The proof is easily obtained from Theorem 6 and omitted. h
Remark 11. In this section, Theorem 6 provides an LMI-based condition for estimating the upper bound of singularly per-turbed parameter � of discrete-time singularly perturbed systems subject to H1 performance bound constraints. In contrastto the existing techniques for estimating stability bounds, the new method can be used to estimate the bound of singularlyperturbed parameter � subject to H1 performance bound constraints. Moreover, two controller design methods are respec-tively given by Theorems 8 and 10. The two methods can eliminate the regularity restrictions attached the existing Riccati-based solution [23], the fact is shown by Example 12. In particular, the method given by Theorem 10 can be used to improvethe upper bound of singularly perturbed parameter � by designing controllers, which is illustrated by Example 14.
4. Example
In Example 12, a discrete-time fuzzy singularly perturbed system is obtained by discretizing a tunnel diode circuit, whichis borrowed from [2]. Because the example does not satisfy the regularity property, the conventional Riccati-based method isnot applicable. The new LMI-based method given by Theorem 8 will be applied to the example for illustrating itseffectiveness.
Moreover, Example 14 is given for better illustrating the effectiveness of the method given by Theorem 10 (i.e., the H1controller design method with the consideration of improving the upper bound of singularly perturbed parameter �).
Example 12. Consider a tunnel diode circuit (Fig. 1), where the diode current is iDðtÞ and the diode voltage vDðtÞ, and theysatisfy that
iDðtÞ ¼ �0:2vDðtÞ � 0:05v3DðtÞ
Moreover, C, R and L denote the capacitor, the resistance and the inductance, respectively. vCðtÞ, vRðtÞ and vLðtÞ are thecapacitor voltage, the resistance voltage and the inductor voltage, respectively. iCðtÞ, iRðtÞ and iLðtÞ are the capacitor current,the resistance current and the inductor current, respectively. uðtÞ is the input voltage, wðtÞ is the disturbance. ConsideringFig. 1 and applying the Kirchoff voltage and current law, we have that
CdðvCðtÞÞ
dt¼ icðtÞ; RiLðtÞ ¼ vRðtÞ
LdðiLðtÞÞ
dt¼ vLðtÞ; iCðtÞ ¼ iLðtÞ � iDðtÞ
vLðtÞ ¼ uðtÞ � vRðtÞ � vCðtÞ þwðtÞ
ð30Þ
Let x1ðtÞ ¼ vCðtÞ, x2ðtÞ ¼ iLðtÞ. Combining them and (30), then it follows that
C _x1ðtÞ ¼ 0:2x1ðtÞ þ 0:05x31ðtÞ þ x2ðtÞ
L _x2ðtÞ ¼ �x1ðtÞ � Rx2ðtÞ þ uðtÞ þwðtÞzðtÞ ¼ x1ðtÞ þ 0:1wðtÞ
ð31Þ
where zðtÞ is the controlled output. Assume that the parameters of the circuit are C ¼ 100 mF, L ¼ 1 mH and R = 20 X. Withthese parameters, (31) can be rewritten as follows:
Fig. 1. The tunnel diode circuit in Example 12.
J. Dong, G.-H. Yang / Information Sciences 179 (2009) 3041–3058 3049
_x1ðtÞ ¼ 2x1ðtÞ þ 0:5x31ðtÞ þ 10x2ðtÞ
� _x2ðtÞ ¼ �0:1x1ðtÞ � 2x2ðtÞ þ 0:1uðtÞ þ 0:1wðtÞzðtÞ ¼ x1ðtÞ þ 0:1wðtÞ
ð32Þ
where xðtÞ ¼ ½xT1ðtÞ; xT
2ðtÞ�T and � ¼ 10�4. Assume that jxðtÞj 6 3 and model the nonlinear system (32) by the sector nonlinear-
ity approach in [14], then the following T–S fuzzy model can be obtained:
E� _xðtÞ ¼X2
i¼1
aiðtÞ AicxðtÞ þ Bi
cwwðtÞ þ BicuuðtÞ
� �zðtÞ ¼
X2
i¼1
aiðtÞ CiczxðtÞ þ Di
czwwðtÞ� �
where
A1c ¼
2 10�0:1 �2
� �; A2
c ¼6:9 10�0:1 �2
� �; B1
cw ¼ B2cw ¼
00:1
� �B1
cu ¼ B2cu ¼
00:1
� �; C1
cz ¼ C2cz ¼ ½1 0�; D1
czw ¼ D2czw ¼ 0:1; E� ¼
1 00 �
� �
and aiðtÞ is the normalized time-varying fuzzy weighting function for each rule i ¼ 1;2, satisfies a1ðtÞ ¼ 1� x21ðtÞ9 ,
a2ðtÞ ¼ 1� a1ðtÞ.
From (32), it can be seen that x2ðtÞ, i.e., the inductor current iLðtÞ, changes very fast and exhibits a large range of variation.Therefore, it is very difficult to measure x2ðtÞ. On the other hand, the slow variable x1ðtÞ changes slowly and exhibits a smallrange of variation, it can be easily measured. Therefore, the slow state feedback controller is needed for this example.
Inhere, we discretize the model with a sampling period T ¼ 0:3 s and a zero-order holder, then the following discrete-time singularly perturbed model is obtained:
xðkþ 1Þ ¼X2
i¼1
aiðkÞ AidE�xðkÞ þ Bi
dwwðkÞ þ BiduuðkÞ
� �zðkÞ ¼
X2
i¼1
aiðkÞ CidzxðkÞ þ Di
dzwwðkÞ� �
where
A1d ¼
1:5683 7:8415�0:0784 �0:3921
� �; A2
d ¼6:8210 34:1039�0:3410 �1:7051
� �; B1
dw ¼0:18940:0405
� �B2
dw ¼0:45470:0273
� �; B1
du ¼0:18940:0405
� �; B2
du ¼0:45470:0273
� �; C1
dz ¼ C2dz ¼ ½1 0�;
D1dzw ¼ D2
dzw ¼ 0:1
Due to DdzwðaðkÞÞ– 0, the regularity property is not satisfied, the conventional Riccati-based methods [23] are not applicable.Applying Theorem 8 (the H1 controller design method without the consideration of improving the upper bound of the sin-gular perturbation parameter �), then the following results are obtained
Q 1 ¼0:4220 0:03080:0308 995:8465
� �; Q 2 ¼
0:4220 0:02930:0293 994:0820
� �;
L1a ¼ �4:4462; L2
a ¼ �6:5378;
K1a ¼ �10:5348; K2
a ¼ �15:4908; copt ¼ 0:51639
With the obtained controller gains, the upper bound of the singular perturbation parameter � of the tunnel diode circuit sys-tem are estimated as �� ¼ 0:1772 by using Theorem 6 under the H1 performance constraint c ¼ 0:5614. It is bigger than thepractical parameter � ¼ 10�4. Therefore, the designed controller can be used and some simulation results are shown in Figs.
2–4 with an initial state xð0Þ ¼ ½0:51�T , and wðkÞ ¼ 1; 5 6 k 6 100; others:
�From Figs. 2, and 3, it can be seen that the resulting closed-loop system is asymptotically stable and with a good H1 per-
formance, which further shows the effectiveness of the proposed condition in Theorem 8.
Remark 13. Note that Theorem 8 is applied for designing an H1 controller in Example 12 and the upper bound of singularlyperturbed parameter � for the resulting closed-loop system is more than the practical �, hence the controller is applicable.Since the controller design condition in Theorem 8 is without considering the tradeoff of the H1 performance bound c andthe stability bound ��, the obtained controller by Theorem 8 might be not applicable for some singularly perturbed systems.For solving the problem, Theorem 10 can be applied as an alternative, which is shown in Example 14.
0 5 10 15 20 25 30−0.4
−0.2
0
0.2
0.4
0.6
0.8
1
Time (sec)
x(k)
x1x2
Fig. 2. Trajectories of xðkÞ in Example 12.
0 5 10 15 20 25 30−0.5
0
0.5
1
1.5
2
2.5
3
3.5
Time (sec)
z(k)
, w(t)
zw
Fig. 3. Trajectories of zðkÞ and wðkÞ in Example 12.
Fig. 4. Membership functions aiðx1ðkÞÞ for Example 14.
3050 J. Dong, G.-H. Yang / Information Sciences 179 (2009) 3041–3058
J. Dong, G.-H. Yang / Information Sciences 179 (2009) 3041–3058 3051
Example 14. Consider a numerical example, which is described by (4) with
Table 1Upper b
c ¼ 1:5c ¼ 2c ¼ 2:5
A1 ¼
0:6 1 �2 10 2 0 �3
0:1 0 0:3 12:5 0:1 0:2 0:4
2666437775; A2 ¼
0:7 1 �3 0:80 1 0 �2
0:1 0 0:3 18 0:5 0:1 0:6
2666437775
B1w ¼
00:20
0:2
2666437775; B2
w ¼
00:30
0:1
2666437775; B1
u ¼
0:510
0:8
2666437775; B2
u ¼
0:310
0:4
2666437775
C1z ¼ C2
z ¼1 0 0 00 1 0 0
� �; D1
zw ¼01
� �; D2
zw ¼0
1:2
� �; D1
zu ¼ D2zu ¼
00:1
� �
E� ¼
1 0 0 00 1 0 00 0 � 00 0 0 �
2666437775
where � ¼ 0:05 and the membership functions aiðx1ðkÞÞ, i ¼ 1;2 are given in Fig. 4.
Applying Theorem 8 to the example (the H1 controller design method without the consideration of improving the upperbound of the singular perturbation parameter �), we can obtain the following controller gains and the optimal H1 perfor-mance bound.
K1a ¼ ½�0:1544 � 2:1685�; K2
a ¼ ½�0:0764 � 1:9145�copt ¼ 1:3334
ð33Þ
ounds of �.
Theorem 8 Theorem 10
0.0014 0.05270.0049 0.05560.0073 0.0576
0 5 10 15 20 25 30−3
−2
−1
0
1
2
3x 1015
Time (sec)
x(k)
x1x2
Fig. 5. Trajectories of xðkÞ via the gain (33) in Example 14.
3052 J. Dong, G.-H. Yang / Information Sciences 179 (2009) 3041–3058
Moreover, by using Theorem 10 with �� ¼ 0:05 (the H1 controller design method with the consideration of improving theupper bound of the singular perturbation parameter �), we can obtain
K1a ¼ ½�0:2233 � 2:1834�; K2
a ¼ ½�0:7358 � 1:8368�copt ¼ 1:4284
ð34Þ
Then the allowable upper bound of the singular perturbation parameter � of the closed-loop system with the controller gain(33) or (34) can be estimated by using Theorem 6. The obtained results are shown in Table 1.
From Table 1, it can be seen that,for the larger H1 performance indices,the larger upper bounds of the singular perturba-tion parameter � are achieved by using Theorem 10, which shows that the method of Theorem 10 is effective for improvingthe upper bound of the singular perturbation parameter �.
0 5 10 15 20 25 30−6
−4
−2
0
2
4
6
8
Time (sec)
x(k)
x1x2
Fig. 6. Trajectories of xðkÞ via the gain (34) in Example 14.
0 5 10 15 20 25 300
5
10
15
20
25
30
Time (sec)
The
ratio
1.3572
Fig. 7.ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiPk
i¼0zT ðiÞzðiÞ=Pk
i¼0wT ðiÞwðiÞq
with the gain (34) in Example 14.
J. Dong, G.-H. Yang / Information Sciences 179 (2009) 3041–3058 3053
Note that � ¼ 0:05, then for the case of a1ðkÞ ¼ 0, a2ðkÞ ¼ 1, the eigenvalues of
AðaðkÞÞE� þ BuðaðkÞÞKðaðkÞÞ ¼ A2E� þ B2uK2 ¼ A2E� þ B2
u½K2a0�
with the controller gain (33) designed by using Theorem 8 are �1.1190, 0.6411, 0.4496, �0.1641, which implies that theresulting closed-loop system is unstable. Therefore,the controller cannot be applied. On the other hand, Table 1 showsthe achieved upper bounds of � of the closed-loop system with the controller gains (34) under different H1 performancerequirements. The upper bounds are more than � ¼ 0:05. The fact shows that Theorem 10 can be used to effectively improvethe upper bound of the singular perturbation parameter � at the stage of controller design.
What it follows, some simulation results will be given in order to further validate the effectiveness of Theorem 10. As-sume that the initial state xð0Þ ¼ ½�2 8 0 0�T and the disturbance
wðkÞ ¼2; 3 6 k 6 100; others
�
Figs. 5 and 6 show the trajectories of the state xðkÞ of the closed-loop system with the controller gains (33) and (34), respec-tively. Fig. 7 shows the ratio offfiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiPk
i¼0zTðiÞzðiÞ=Pk
i¼0wTðiÞwðiÞq
in the simulation.From Fig. 5, it can be seen that the closed-loop system with the controller gain (33), which is obtained by Theorem 8, is
unstable. It can be seen that from Figs. 6 and 7 that the controller gain (34), which is obtained by Theorem 10, can guaranteethe stability and meet the H1 performance requirement. These simulation results further show the advantage of the methodgiven by Theorem 10, i.e., Theorem 10 can be used to effectively improve the upper bound of the singular perturbationparameter � at the stage of controller design.
5. Conclusion
In this paper, the H1 control problem via slow state variables feedback for discrete-time fuzzy singularly perturbed systemshas been investigated. Two LMI-based methods for designing H1 controllers via slow state variables feedback are presented,and one of them can be used to improve the upper bound of the singular perturbation parameter �, which can overcome thedisadvantage in the conventional design methods where the designed controller might not be used because the resulting allow-able upper bound of the singular perturbation parameter is too small. Moreover, a method of evaluating the upper bound of asingular perturbation parameter �with meeting a prescribed H1 performance bound requirement is given in terms of solutionsto a set of LMIs. The effectiveness of the proposed methods has been illustrated by the numerical examples.
Acknowledgements
This study was supported in part by the Funds for Creative Research Groups of China (No. 60821063), the State Key Pro-gram of National Natural Science of China (Grant No. 60534010), National 973 Program of China (Grant No. 2009CB320604),the Funds of National Science of China (Grant No. 60674021), the 111 Project (B08015) and the Funds of PhD program ofMOE, China (Grant No. 20060145019).
Appendix A. Proof of Lemma 2
Proof. Consider the system (11) (which is equivalent to the system (9)), and let PðaðkÞÞ > 0. Choose Lyapunov function
VðkÞ ¼ cnTðkÞPðaðkÞÞnðkÞ
then
Vðkþ 1Þ � VðkÞ þ zTðkÞzðkÞ � c2wTðkÞwðkÞ¼ c AðaðkÞÞ þ BuðaðkÞÞKðaðkÞÞð ÞnðkÞ þ BwðaðkÞÞwðkÞð ÞT E�Pðaðkþ 1ÞÞE� � AðaðkÞÞ þ BuðaðkÞÞKðaðkÞÞð ÞnðkÞðþ BwðaðkÞÞwðkÞÞ � cnTðkÞPðaðkÞÞnðkÞ þ CzðaðkÞÞ þ DzuðaðkÞÞKðaðkÞÞð ÞnðkÞð þ DzwðaðkÞÞwðkÞÞT � CzðaðkÞÞððþ DzuðaðkÞÞKðaðkÞÞÞnðkÞ þ DzwðaðkÞÞwðkÞÞ � c2wTðkÞwðkÞ
¼ cnðkÞwðkÞ
� �TU11ðk; kþ 1Þ UT
21ðk; kþ 1ÞU21ðk; kþ 1Þ U22ðk; kþ 1Þ
" #nðkÞwðkÞ
� �
where /11ðk; kþ 1Þ, /21ðk; kþ 1Þ, /22ðk; kþ 1Þ are the same as in (13).From (12) and the above equality, then it follows that
Vðkþ 1Þ � VðkÞ þ zTðkÞzðkÞ � c2wTðkÞwðkÞ 6 0; for � 2 ð0; ���
3054 J. Dong, G.-H. Yang / Information Sciences 179 (2009) 3041–3058
From the above inequality, we have that the system (11) with � 2 ð0; ��� is asymptotically stable in the disturbance-free case.For the initial condition xð0Þ ¼ 0 and every positive integer N, sum the above inequality from k ¼ 0 to N, then we can obtain
VðNÞ � Vð0Þ þX1k¼0
zTðkÞzðkÞ � c2X1k¼0
wTðkÞwðkÞ ¼ VðNÞ þX1k¼0
zTðkÞzðkÞ � c2X1k¼0
wTðkÞwðkÞ 6 0; for � 2 ð0; ���
which implies that
XNk¼0
zTðkÞzðkÞ 6 c2XN
k¼0
wTðkÞwðkÞ; for � 2 ð0; ���
Combining it and Definition 1, it follows that the H1 norm of the closed-loop system (11) is less than or equal to c. h
Appendix B. Proof of Lemma 3
Proof. Consider the following two cases:
(i): If a ¼ 0, then from (15) and (16), (17) obviously holds.(ii): If a > 0, we consider the following quadratic function of �,
yð�Þ ¼ a�2 þ b�þ c ð35Þ
Since a > 0, yð�Þ is convex function of � [4]. From (15) and (16), it follows that yð��Þ < 0 and yð0Þ < 0, which further impliesthat yð�Þ < 0 for � 2 ½0; ���, i.e., when a > 0, (17) holds. Thus, the proof is complete. h
Appendix C. Proof of Lemma 4
Proof. For all nonzero vector xðkÞ, pre- and post-multiplying (18)–(20) by xTðkÞ and its transpose, then we have
xTðkÞT1xðkÞP 0 ð36Þ
��2xTðkÞT1xðkÞ þ ��xTðkÞT2xðkÞ þ xTðkÞT3xðkÞ < 0 ð37Þ
xTðkÞT3xðkÞ < 0 ð38Þ
Denote axk¼ xTðkÞT1xðkÞ, bxk
¼ xTðkÞT2xðkÞ, cxk¼ xTðkÞT3xðkÞ. Substituting axk
, bxkand cxk
into (36)–(38), then it follows that
axkP 0 ð39Þ
axk��2 þ bxk
�� þ cxk< 0 ð40Þ
cxk< 0 ð41Þ
From (39)–(41) and applying Lemma 3, we can obtain axk�2 þ bxk
�þ cxk< 0 for � 2 ½0; ���, i.e., for all nonzero vector xðkÞ,
�2xTðkÞT1xðkÞ þ �xTðkÞT2xðkÞ þ xTðkÞT3xðkÞ < 0; for � 2 ½0; ���
which implies that matrix inequality (21) holds. Thus, the proof is complete. h
Appendix D. Proof of Lemma 5
Proof. (22) can be rewritten as follows,
W11ðkÞ � � �0 �cI � �
W31ðkÞ BwðaðkÞÞ �Qðaðkþ 1ÞÞ �W41ðkÞ DzwðaðkÞÞ 0 �cI
2666437775 < 0; for � 2 ð0; ���: ð42Þ
where W31ðkÞ, W41ðkÞ are the same as in (23) and
W11ðkÞ ¼ E�QðaðkÞÞE� � SðaðkÞ;aðkþ 1ÞÞ � STðaðkÞ;aðkþ 1ÞÞ
where E� is same as in (5).
J. Dong, G.-H. Yang / Information Sciences 179 (2009) 3041–3058 3055
From (42), we can obtain that W11ðkÞ < 0. Considering the block (3,3) of (42), then yields QðaðkÞÞ > 0. Combiningit and W11ðkÞ < 0, then we have SðaðkÞ;aðkþ 1ÞÞ þ STðaðkÞ;aðkþ 1ÞÞ > 0. Pre- and post-multiply (42) by
S�TðaðkÞ;aðkþ 1ÞÞ 0 0 00 I 0 00 0 I 00 0 0 I
26643775 and
S�1ðaðkÞ;aðkþ 1ÞÞ 0 0 00 I 0 00 0 I 00 0 0 I
26643775, respectively. Then it follows that
eW11ðkÞ � � �0 �cI � �
AðaðkÞÞ þ BuðaðkÞÞKðaðkÞÞ BwðaðkÞÞ �Qðaðkþ 1ÞÞ �CzðaðkÞÞ þ DzuðaðkÞÞKðaðkÞÞ DzwðaðkÞÞ 0 �cI
266664377775 < 0; for � 2 ð0; ��� ð43Þ
where
eW11ðkÞ ¼ S�TðaðkÞ;aðkþ 1ÞÞE�QðaðkÞÞE�S�1ðaðkÞ;aðkþ 1ÞÞ � S�1ðaðkÞ;aðkþ 1ÞÞ � S�TðaðkÞ;aðkþ 1ÞÞLet P�1ðaðkÞÞ ¼ E�QðaðkÞÞE�, therefore,
PðaðkÞÞ > 0 ð44Þ
thenS�TðaðkÞ;aðkþ 1ÞÞ � PðaðkÞÞ� �
P�1ðaðkÞÞ S�1ðaðkÞ;aðkþ 1ÞÞ � PðaðkÞÞ� �
P 0
which implies that
S�TðaðkÞ;aðkþ 1ÞÞP�1ðaðkÞÞS�1ðaðkÞ;aðkþ 1ÞÞ � S�1ðaðkÞ;aðkþ 1ÞÞ � S�TðaðkÞ;aðkþ 1ÞÞP �PðaðkÞÞ
Combining it with (43), we can obtain
�PðaðkÞÞ � � �0 �cI � �
AðaðkÞÞ þ BuðaðkÞÞKðaðkÞÞ BwðaðkÞÞ �E�1� P�1ðaðkþ 1ÞÞE�1
� �CzðaðkÞÞ þ DzuðaðkÞÞKðaðkÞÞ DzwðaðkÞÞ 0 �cI
2666437775 < 0; for � 2 ð0; ��� ð45Þ
Applying Schur complement lemma to (45), then we have
�PðaðkÞÞ � �0 �cI �
AðaðkÞÞ þ BuðaðkÞÞKðaðkÞÞ BwðaðkÞÞ �E�1� P�1ðaðkþ 1ÞÞE�1
�
264375
þ 1c
CTz ðaðkÞÞ þ KTðaðkÞÞDT
zuðaðkÞÞDT
zwðaðkÞÞ0
264375� CzðaðkÞÞ þ DzuðaðkÞÞKðaðkÞÞ DzwðaðkÞÞ 0½ �
¼� 11ðkÞ � �� 21ðkÞ � 22ðkÞ �
AðaðkÞÞ þ BuðaðkÞÞKðaðkÞÞ BwðaðkÞÞ �E�1� P�1ðaðkþ 1ÞÞE�1
�
2643750; for � 2 ð0; ���
where
� 11ðkÞ ¼ �PðaðkÞÞ þ 1c
CzðaðkÞÞ þ DzuðaðkÞÞKðaðkÞÞð ÞT
� CzðaðkÞÞ þ DzuðaðkÞÞKðaðkÞÞð Þ
� 21ðkÞ ¼1c
DTzwðaðkÞÞ CzðaðkÞÞ þ DzuðaðkÞÞKðaðkÞÞð Þ
� 22ðkÞ ¼ �cI þ 1c
DTzwðaðkÞÞDzwðaðkÞÞ
Applying Schur complement lemma to the above inequality, again, then it follows that
� 11ðkÞ � T21ðkÞ
� 21ðkÞ � 22ðkÞ
� �þ ATðaðkÞÞ þ KTðaðkÞÞBT
uðaðkÞÞBT
wðaðkÞÞ
� �E�Pðaðkþ 1ÞÞE� � AðaðkÞÞ þ BuðaðkÞÞKðaðkÞÞ BwðaðkÞÞ½ �
¼ U11ðk; kþ 1Þ UT21ðk; kþ 1Þ
U21ðk; kþ 1Þ U22ðk; kþ 1Þ
� �< 0; for � 2 ð0; ��� ð46Þ
where U11ðk; kþ 1Þ, U21ðk; kþ 1Þ and U22ðk; kþ 1Þ are the same as in (13).
3056 J. Dong, G.-H. Yang / Information Sciences 179 (2009) 3041–3058
Then applying Lemma 2 to (46), we have that, for each singular perturbation parameter � 2 ð0; ���, the system (11) isasymptotically stable and its H1 norm is less than c. Then the conclusion follows. h
Appendix E. Proof of Theorem 6
Proof. Considering the block (3,3) of (24a) and from (24a), it follows that
Qi > 0 1 6 i 6 r
which also implies that
Qi22 > 0 1 6 i 6 r ð47Þ
Let
T1i ¼
0 00 Q i
22
� �0 0 0
0 0 0 00 0 0 00 0 0 0
2666664
3777775
T2i ¼
0 ðQ i21Þ
T
Q i21 0
" #0 0 0
0 0 0 00 0 0 00 0 0 0
26666664
37777775T3il ¼ Miil
From (47), it follows that
T1i P 0 ð48aÞFrom (25a), we have
��2T1i þ ��T2i þ T3il < 0 ð48bÞ
From (24a), we haveT3il < 0 ð48cÞ
Applying Lemma 4 to (48a), then yields
�2T1i þ �T2i þ T3il < 0; for � 2 ½0; ���; 1 6 i; l 6 r
i.e.,
Kiil < 0; for � 2 ½0; ���; 1 6 i; l 6 r ð49aÞ
where
Kiil ¼ �2
0 00 Q i
22
� �0 0 0
0 0 0 00 0 0 00 0 0 0
2666664
3777775þ �0 ðQ i
21ÞT
Q i21 0
" #0 0 0
0 0 0 00 0 0 00 0 0 0
26666664
37777775þQ i
11 00 0
" #� Sil � ðSilÞT � � �
0 �cI � �Ai þ Bi
uKi� �
Sil Biw �Q l �
Ciz þ Di
zuKi� �
Sil Dizw 0 �cI
26666666664
37777777775
Similarly, from (47), (24b) and (25b), we can also obtain1r � 1
Kiil þ12ðKijl þKjilÞ < 0; for � 2 ½0; ���; 1 6 i – j 6 r; 1 6 l 6 r ð49bÞ
where
Kijl ¼ �2
0 00 Q i
22
� �0 0 0
0 0 0 00 0 0 00 0 0 0
2666664
3777775þ �0 ðQ i
21ÞT
Q i21 0
" #0 0 0
0 0 0 00 0 0 00 0 0 0
26666664
37777775þQ i
11 00 0
" #� Sil � ðSilÞT � � �
0 �cI � �AiSjl þ Bi
uKjSjl Biw �Q l �
CizSjl þ Di
zuKjSjl Dizw 0 �cI
266666664
377777775
J. Dong, G.-H. Yang / Information Sciences 179 (2009) 3041–3058 3057
Applying the parameterized linear matrix inequality (PLMI) technique in [27] to (49a), then yields
Xri¼1
Xr
j¼1
aiðkÞajðkÞKijl < 0; 1 6 l 6 r
Multiplying the above inequality by alðkþ 1Þ and summing them from l ¼ 1 to r, then we can obtain
Xri¼1
Xr
j¼1
Xr
l¼1
aiðkÞajðkÞalðkþ 1ÞKijl < 0 ð50Þ
Note that
Xr
j¼1
Xr
l¼1
ajðkÞalðkþ 1ÞKjSjl ¼Xr
j¼1
Xr
l¼1
ajðkÞalðkþ 1Þ KjaS11 0Sjl
21 Sjl22
" #¼
Prj¼1
ajðkÞKjaS11 0
Prj¼1
Prl¼1
ajðkÞalðkþ 1ÞSjl21
Prj¼1
Prl¼1
ajðkÞalðkþ 1ÞSjl22
2666437775
¼Prj¼1
ajðkÞKja 0
� ��
S11 0Prj¼1
Prl¼1
ajðkÞalðkþ 1ÞSjl21
Prj¼1
Prl¼1
ajðkÞalðkþ 1ÞSjl22
24 35¼ KðaðkÞÞ S11 0
S21ðaðkÞ;aðkþ 1ÞÞ S22ðk; kþ 1Þ
� �¼ KðaðkÞÞSðaðkÞ;aðkþ 1ÞÞ
where KðaðkÞÞ is the same as in (10) and
SðaðkÞ;aðkþ 1ÞÞ ¼S11 0Pr
j¼1
Prl¼1
ajðkÞalðkþ 1ÞSjl21
Prj¼1
Prl¼1
ajðkÞalðkþ 1ÞSjl22
24 35
Then Xri¼1
Xr
j¼1
Xr
l¼1
aiðkÞajðkÞalðkþ 1ÞBiuKjSjl ¼ BuðaðkÞÞKðaðkÞÞSðaðkÞ;aðkþ 1ÞÞ
Xr
i¼1
Xr
j¼1
Xr
l¼1
aiðkÞajðkÞalðkþ 1ÞDizuKjSjl ¼ DzuðaðkÞÞKðaðkÞÞSðaðkÞ;aðkþ 1ÞÞ
Therefore, (50) implies that (22) holds. Combining it and Lemma 5, it follows that for the singular perturbation parameter� 2 ð0; ���, the system (9) is asymptotically stable and its H1 norm is less than c. Then the conclusion follows. h
References
[1] W. Assawinchaichote, S.K. Nguang, H1 fuzzy control design for nonlinear singularly perturbed systems with pole placement constraints: an LMIapproach, IEEE Transactions on Systems, Man and Cybernetics, Part B 34 (1) (2004) 579–588.
[2] W. Assawinchaichote, S.K. Nguang, P. Shi, H1 output feedback control design for uncertain fuzzy singularly perturbed systems: an LMI approach,Automatica 40 (12) (2004) 2147–2152.
[3] S. Boyd, L.E. Ghaoui, E. Feron, V. Balakrishnan, Linear Matrix Inequalities in System and Control Theory, Society for Industrial and Applied Mathematics(SIAM), Philadelphia (PA), 1994.
[4] S. Boyd, L. Vandenberghe, Convex Optimization, Cambridge University Press, 2004.[5] L. Cao, H.M. Schwartz, Complementary results on the stability bounds of singularly perturbed systems, IEEE Transactions on Automatic Control 49 (11)
(2004) 2017–2021.[6] Y. Chen, Y. Liu, Summary of singular perturbation modeling of multi-time scale power systems, in: Transmission and Distribution Conference and
Exhibition, IEEE/PES, Asia and Pacific, 2005.[7] H.L. Choi, J.W. Son, J.T. Lim, Stability analysis and control of non-standard nonlinear singularly perturbed system, IEE Proceedings-Control Theory and
Applications 153 (6) (2006) 703–708.[8] J. Dong, G.H. Yang, Robust H1 control for standard discrete-time singularly perturbed systems, IET Control Theory and Applications 1 (4) (2007) 1141–
1148.[9] J. Dong, G.H. Yang, H1 control for nonstandard discrete-time singularly perturbed systems, Automatica 44 (2008) 1385–1393.
[10] E. Fridman, A descriptor system approach to nonlinear singularly perturbed optimal control problem, Automatica 37 (4) (2001) 543–549.[11] P. Gahinet, A. Nemirovski, A.J. Laub, M. Chilali, LMI Control Toolbox, The MathWorks, Inc., Natick, MA, 1995.[12] G. Garcia, J. Daafouz, J. Bernussou, The infinite time near optimal decentralized regulator problem for singularly perturbed systems: a convex
optimization approach, Automatica 38 (8) (2002) 1397–1406.[13] R. Ghosh, S. Sen, K.B. Datta, Method for evaluating stability bounds for discrete-time singularly perturbed systems, IEE Proceedings-Control Theory and
Applications 146 (2) (1999) 227–233.[14] K. Tanaka, H.O. Wang, Fuzzy Control Systems Design and Analysis: A Linear Matrix Inequality Approach, John Wiley & Sons Inc., New York, NY, USA,
2001.[15] C. Gong, B. Su, Robust L2—L1 filtering of convex polyhedral uncertain time-delay fuzzy systems, International Journal of Innovative Computing
Information and Control 4 (4) (2008) 793–802.[16] F.-H. Hsiao, J.-D. Hwang, S.-T. Pan, D-stability problem of discrete singularly perturbed systems, International Journal of Systems Science 34 (3) (2003)
227–236.
3058 J. Dong, G.-H. Yang / Information Sciences 179 (2009) 3041–3058
[17] P.V. Kokotovic, J. O’Reilly, H.K. Khalil, Singular Perturbation Methods in Control: Analysis and Design, Academic Press Inc., Orlando, FL, USA, 1986.[18] T.-H.S. Li, J.-S. Chiou, F.-C. Kung, Stability bounds of singularly perturbed discrete systems, IEEE Transactions on Automatic Control 44 (10) (1999)
1934–1938.[19] T.H.S. Li, K.-J. Lin, Stabilization of singularly perturbed fuzzy systems, IEEE Transactions on Fuzzy Systems 12 (5) (2004) 579–595.[20] W. Lin, C.I. Byrnes, H1-control of discrete-time nonlinear systems, IEEE Transactions on Automatic Control 41 (4) (1996) 494–510.[21] H. Liu, F. Sun, Y. Hu, H1 control for fuzzy singularly perturbed systems, Fuzzy Sets and Systems 155 (2) (2005) 272–291.[22] I. Lizarraga, V. Etxebarria, Combined PD-H1 approach to control of flexible link manipulators using only directly measurable variables, Cybernetics and
Systems 34 (1) (2003) 19–31.[23] D.S. Naidu, Singular perturbations and time scales in control theory and applications: an overview, Dynamics of Continuous Discrete and Impulsive
Systems-Series B-Applications and Algorithms 9 (2) (2002) 233–278.[24] V.R. Saksena, J. Cruz, Stabilization of singularly perturbed linear time-invariant systems using low-order observers, IEEE Transactions on Automatic
Control 26 (2) (1981) 510–513.[25] K. Tanaka, T. Ikeda, H.O. Wang, Fuzzy regulators and fuzzy observers: relaxed stability conditions and LMI-based designs, IEEE Transactions on Fuzzy
Systems 6 (2) (1998) 250–265.[26] C.-S. Ting, An observer-based approach to controlling time-delay chaotic systems via Takagi–Sugeno fuzzy model, Information Sciences 177 (20)
(2007) 4314–4328.[27] H.D. Tuan, P. Apkarian, T. Narikiyo, Y. Yamamoto, Parameterized linear matrix inequality techniques in fuzzy control system design, IEEE Transactions
on Fuzzy Systems 9 (2) (2001) 324–332.[28] J. Wang, H. Li, Nonlinear PI control of a class of nonlinear singularly perturbed systems, IEE Proceedings-Control Theory and Applications 152 (5)
(2005) 560–566.[29] Y. Wang, Z. Sun, H1 control of networked control system via LMI approach, International Journal of Innovative Computing Information and Control 3
(2) (2007) 343–352.[30] H.-N. Wu, K.-Y. Cai, Robust fuzzy control for uncertain discrete-time nonlinear Markovian jump systems without mode observations, Information
Sciences 177 (6) (2007) 1509–1522.