2006
IEEE
International
Conference
on
Systems,
Man,
and
Cybernetics
October 8-11, 2006,
Taipei,
Taiwan
Optimal
Output Feedback Control for Linear Uncertain
Systems
Using LMI-Based Approach and Genetic Algorithm
Shinn-Homg Chen, Wen-Hsien
Ho
and
Jyh-Homg Chou,
Senior
Member,
IEEE
Abstract This paper considers the robust-optimal design problems of output feedback controllers for linear systems with both time-varying elemental (structured) and norm-bounded (unstructured) parameter uncertainties. A new sufficient condition is proposed in terms of linear matrix inequalities (LMIs) for ensuring that the linear output feedback systems with both time-varying elemental and norm-bounded parameter uncertainties are asymptotically stable, where the mixed quadratically-coupled parameter uncertainties are directly considered in the problem formulation. A numerical example is given to show that the presentedsufficient condition is less conservative than the existing one reported recently. Then, by integrating the hybrid Taguchi-genetic algorithm (HTGA) and the proposed LMI-based sufficient condition, a new integrative approach is presented to find the output feedback controllers of the linear systems with both time-varying elemental and norm-bounded parameter uncertainties such that the control objective of minimizing a quadratic integral performance criterion subject to the stability robustness constraint is achieved. A design example of the robust-optimal output feedback controller for the AFTI/F-16 aircraft control system with the time-varying elemental parameter uncertainties is given to demonstrate the applicability of the proposed new integrative approach.
1. INTRODUCTION
In general, a mathematical description is only an approximation of the actual physical system and deals with fixed nominal parameters. Usually, these parameters are not known exactly due to imperfect identification or measurement, aging of components and/or changes inenvironmental conditions. Thus, it is almost impossible to get an exact model for the system due to the existence of various parameter uncertainties. Here, we consider linear state-space systems with time-varying uncertain parameters in the system matrix, input matrix, and output matrix. Because the output feedback controller design is usually based on the nominal values of these system matrices, it is interesting to know whether the closed-loop system remains asymptotically stable in the presence of time-varying Manuscript received March 30, 2006. This work wassupported bythe National Science Council, Taiwan,Republic ofChina, under grant number NSC 94-221 8-E327-001.
S. H. Chen is with the Department ofMechanical Engineering, National Kaohsiung University of Applied Sciences, 415 Chien-Kung Road, Kaohsiung 807, Taiwan, R.O.C. (e-mail:
shchen@cc.kuas.edu.tw).
W. H. Ho is with the Institute of Engineering Science and Technology, National Kaohsiung First University of Science and Technology, 1 University Road, Yenchao, Kaohsiung 824, Taiwan,
R.O.C.(e-mail: wenhsienl
102gyahoo.com.tw).
J. H. Chou is with the Institute of Engineering Science and Technology, National Kaohsiung First University of Science and Technology, I University Road, Yenchao, Kaohsiung 824, Taiwan,
R.O.C. (phone: +886-7- 6011000; fax: +886-7-6011066; e-mail: choujh(ccms.nkfust.edu.tw).
uncertain parameters. Applying those existing robust stability analysisresults [1]-[13] tosolve this problem is not easy, in that after output feedback, there will be coupled terms of parameters in the closed-loop system matrix because of the uncertainparameters in bothinput and output matrices [14]-[15]. Though we may regard thesecoupled terms as new independentparameters ifwe insist on using thoseexisting robuststability analysis results, literatures [14] and [16]have showed that a conservative analysis conclusion may be reached. Therefore, literatures [14] and [16] investigated the robust stability problem of linear systems with constant output feedback in the presence oftime-varying elemental (structured) parameteruncertainties by directly considering the coupled terms in the problem formulation. Su and Fong [14] used the Lyapunov method to analyze the robust stability of linear continuous-time systems with quadratically-coupledelemental uncertainties. Tseng etal. [16]applied thestructuredsingularvaluetechniquetosolve the robuststability analysis problem of linearcontinuous-time systems withquadratically-coupledelemental uncertainties. Inaddition,it is well known that an approximate system model is always used in practice and sometimesthe approximationerrorshouldbecovered by introducing both elemental (structured) and norm-bounded
(unstructured) parameter uncertainties in control system analysis
anddesign [17]. That is,it is not unusual that at times we haveto
dealwith a system simultaneously consistingof two parts: onepart has only the elemental parameteruncertainties, and the other part has the norm-bounded parameter uncertainties. Therefore, very recently, based on the Lyapunov approach and some essential properties of matrix measures, Chen and Chou [15] proposed a sufficient conditionto studytheproblem ofstabilityrobustness for linear output feedback systems with bothtime-varying elemental
and norm-boundedparameteruncertainties bydirectly considering
the mixed quadratically-coupled uncertainties (i.e., quadratically-coupled elemental, quadratically-coupled
norm-bounded, and coupled elemental and norm-bounded uncertaintiesappearingtogether)intheproblemformulation. Here it
shouldbe noted thatonlythe articles in[14]-[16] studiedthe robust
stability of linear continuous-time systems with output feedback controllers and time-varying uncertain parameters by
directly
considering thecoupled terms intheproblem formulation. Thatis,
the research on the stability robustness of linear continuous-time systemswith outputfeedback controllersandtime-varying uncertain
parametersbydirectlyconsideringthecoupledtermsintheproblem
formulation is considerably rare and almost embryonic. Besides,
here it should be also notedthat, for the case ofonly considering
time-varyingelementalparameteruncertainties,ChenandChou[15] haveshown that their sufficient condition is less conservative than those in [14] and[16].
On the other hand, only robust stability is often notenoughin control design. Theoptimaltrackingperformance is also considered in many practical control engineering applications. Hence, one of the most important study issues for linear continuous-time output feedback control systems withtime-varying uncertain parameters is to find the output feedback controllers that minimize an
H,
performanceindex (i.e., the integral of the squared error(ISE) or the integral of thetime-weighted squared error (ITSE)) subject to the stability robustness constraint, where the quadratically-coupled
parameteruncertaintiesaredirectlyconsidered in the robuststability
problem formulation. But, to the authors' best knowledge, the mentioned-abovestudyissuehasnotbeen discussed in the literature. Therefore, the purpose of this paper is to propose a newmethodfor
findingthe outputfeedback controllersofthe linear continuous-time systems with both time-varying elemental and norm-bounded
parameter uncertainties such that the control objective of minimizing an
H,
performance index subject to the stabilityrobustness constraintisachieved. Theproposednewmethodisan
integrative approach which integrates the linear-matrix-inequality (LMl)technique and thehybridTaguchi-genetic algorithm (HTGA),
where the LMI technique is used to derive the robust stability condition and the mixed quadratically-coupled parameter
uncertaintiesaredirectly consideredinthe robuststability problem formulation. The reasonwhythe HTGA isappliedinthis paper is thatChou andhisassociateshave shownthat the HTGA mayobtain better andmorerobust results than thoseexisting improved genetic algorithms reportedin theliterature[1 8]-[19].
This paper isorganizedasfollows.InSection 2, an LMI-based
robuststabilitycondition ispresentedfor the linear continuous-time systems with output feedback controllers as well as both
time-varying elementaland norm-bounded parameteruncertainties by directly considering the mixed quadratically-coupled
uncertainties (i.e., quadratically-coupled elemental, quadratically-coupled norm-bounded, and coupled elemental and
norm-bounded uncertainties appearing together) in the problem
formulation. A numericalexampleand the conservatismcomparison
between the proposed LMI-based sufficient condition and the
non-LMI-basedsufficient conditionproposedin[15] arealsogiven
inthissection. The HTGA forthe mixed H2I LMI robust-optimal
output feedback controllers design is described in Section 3. A
design example of the AFTI/F-16 aircraft control system is also given in this section for demonstrating the applicability of the
proposed new integrative method. Finally, Section 4 offers some
conclusions.
Il.
STABILITY ROBUSTNESS ANALYSISConsider thelinearuncertain system withthestate-spacemodel
x~(t)
=A(t)x(t)+B(t)u(t)
,(1)
y(t)
=Q(tx(t
'(2)
where
x(t)
cR' isthe statevector, y(t)ERP isthe output vector, u(t) ERq is theinput
vector,andin in
A(t)
A +s(t)A
+A(t),
B(t)=B +,cj
(t)Bj
+B(t),j=l j=l
m
and C(t)=
CO
+±ZCj (t)Cj
+C(t) (3)Jl
are the system matrix, the input matrix and the output matrix, respectively, in which
8j
(t)(
ej<8j
(t)<£ / , and j=1,2,...,
m)are the time-varying elemental uncertainties;
Aj
,Bi
andCJ
(
j=1,2,...,m)
are,respectively,
thegiven
nxn, nxq and pxn constantmatrices whichareprescribed priortodenotethe linearlydependent information on time-varying elemental uncertainties
8£/(t)
S; m is the number ofindependent time-varying
uncertain parameters. The time-varying norm-bounded uncertain matricesA(t), B(t) and
C(t)
areassumedtobebounded,
i.e.,
A(t) <
A3,
||B(t)||<,2
, and ||C(t) <p,8
(4)where /3, /32' and /3 are non-negative real constant number, and
111
denotes any matrixnorm.In this paper, we only discuss the static output feedback controllers. Thus, the closed-loop system equation of the linear uncertain system canbe expressed as
x(t)
LA
+ Z_,(t)Ej
+ZjZi (t)Ck(t)Eok
+F(t)
x(t),
(5)j=i j=l k=l
where K denotes the outputfeedback gain matrix,
AO AO
+BoKC0,
Ej
=ABj
+B(KCj
+±BKC0,
Elk =-(B KC +
BkKCj),
andF(t)
A(t)
+BoKC(t)
+B(t)KC,
+B(t)KC(t)
+
Z£j
(t)(B1KC(t)+B(t)KC ).j=l
From Eqs. (4)and (6), we can get that
||F(t)II
<K
, inwhich(6)
(7)
m
18
=A,9
+
9211KCO
11
+
9311BOKI|
+,92,3
II|K||I
+
J:X
(3
JIB.,KII
+
8211KC ||),
and jmax{|e, j} Therefore, the norm-bounded uncertainmatrix F(t) canbe expressed by
F(t)=
A(t)
(8)where
A(t)
is an unknown matrix function which is bounded byA(t)
EQ:={A(t)IIA(t)II
<1,theelementDfA(t)areLebesgurmeasurabl4The robuststability problemtobe studiedcanbedescribedas:
given an output feedback gain matrix K for making
A.
a stablematrix, determine the condition such that the linear closed-loop uncertainsystemin(5)isstill asymptotically stable.
Inwhatfollows, under the assumption thatanoutputfeedback gain matrix K has been specified to make
AO
a stable matrix inadvance, by using the Lyapunov approach, we present a new
sufficient condition interms of linearmatrix inequalities (LMIs), which canbe efficiently solved by using thenumerically efficient
convex programming algorithms [20] for ensuring that the linear
closed-loop uncertainsystemin (5)remainsasymptotically stable.
Theorem:
The linear closed-loop uncertain system in (5) remains
asymptotically stable, if, for the specified output feedback gain
matrix K and for a specified constant a with 0<a
<1,
thereexist asymmetric positive definite matrix P andaconstant r>0
suchthatthe following LMIsaresimultaneously satisfied:
UP±PU ±TI
mi P <0,
(h
and thesamematrix P makes the
following inequality
hold:(10)
Is m mn
)
"Ej(t)(0
+I
I
- (t)c,(t),4j,
<I XJi=ll j=l k=l
where
i-1,2,...,2;
A=aAo
A, (1-a)A,;
Ui
= ;(t)Ej,8(t)=,/j
or8j;4137
REFERENCES
[1]
K. Zhou and P. P.Khargonekar,
1987,"Stability
robustness bounds for linearstate-space
models with structureduncertainties",
IEEE Trans. on AutomaticControl, Vol.32,
pp.621-623.
[2]
T. H.Hopp
and W. E.Schmitendorf, 1990,
"Design
ofalinear controller for robusttracking
and modelfollowing",
ASME J. ofDynamic
Systems,
Measurement andControl,
Vol.112,
pp.552-558.
[3]
A. Weinmann, 1991, Uncertain Models and RobustControl,
NewYork:
Springer-Verlag.
[4]
Z.Gaoand P. J.Antsaklis,
1993."Explict asymmetric
bounds for robuststability
of continuousand discrete-timesystems",
IEEE Trans.onAutomaticControl, Vol.38,
pp.332-335.
[5]
J. H.Chou, 1994,
"Robuststability
bounds on structuredtime-varying independent
perturbations
for linear state-spacemodels",
JSME InternationalJournal,
SeriesC,
Vol.37,
pp.733-738.
[6]
C. L. Chen and S. K.Yang,
1997, "Robuststability
andperformance
analysis
for the system with lineardependent
perturbations",
J. of MathematicalAnalysis
andApplications,
Vol.2
13,
pp.583-600.
[7]
G.Schreier,
J.Ragot,
P. M. Frank and G. L.Gissinger,
1998,
"Bounds ofparameteruncertainties forrobust
stability
oflinear uncertainsystems",
Int.J.ofSystems Science, Vol.29,
pp.85-90.
[8]
J. D.Chen,
1999, "Robuststability
criterion for structuredperturbations
continuoussystems withLQ
regulators",
Proc. ofthe 1999Automatic Control
Conference,
Taiwan,
pp.748-750.
[9]
J. H. Chou and S. H.Chen,
2000,"~Stability
robustnessanalysis
of linear time-variantsystemswith both
time-varying
structured and unstructuredperturbations",
JSME InternationalJournal,
Series
C,
Vol.43,
pp.98-103.
[10]
E. F. Costa and V. A.Oliveira, 2002,
"On thedesign
ofguaranteed
cost controllers for a class of uncertain linearsystems", Systems
and ControlLetters,Vol.46,
pp.1 7-29.[11I]
L. Yu,2002,
Robust Control: Linear MatrixInequality
Approach, Beijing: Tsinghua University
Press.[12]
J. H.Chou,
S. H.Chen and L. A.Zheng,
2003,
"LMIconditionforrobust
stability
oflinear systems with bothtime-varying
elemental and norm-bounded
uncertainties",
Journal of the ChineseInstitute ofEngineers,
Vol.26,
pp.100-105.[13
]
C. C.Sun,
H. Y.Chung
and W. J.Chang,
2005, "H-,
/H,
robust staticoutputfeedback control
design
viamixedgeneticalgorithm
and linear matrixinequalities",
ASME J.ofDynamic
Systems,
MeasurementandControl,
Vol.127,pp.715-722.
[14]
J. H. Su and I. K.Fong,
1993, "Robuststability analysis
oflinear continuous/discrete time systems with output feedback
controllers",
IEEE Trans. on AutomaticControl, Vol.38,
pp.1 154-1158.
[15]
5. H. Chen and J. H.Chou, 2003,
"Robuststability
oflinearcontinuous/discrete-time output feedback systems with both
time-varying
structured and unstructureduncertainties",
JSMEInternational
Journal,
SeriesC,Vol.46,
pp.705-712.
[16]
C. L.Tseng, I.
K.Fong
and J. H. Su, 1994,"Analysis
andapplications
of robustnonsingularity problem using
the structuredsingular
value",
IEEE Trans.onAutomaticControl.Vol.39,
pp.21
18-2122.[17]
K. Zhou and G. Gu, 1992, "Robuststability
ofmultivariablesystems with both real parametric and norm bounded
uncertainties",
IEEE Trans. on AutomaticControl,
Vol.37,
pp.1533-1537.[18]
J. H.Chou,
W. H. Liao and J. J.Li,
1998,"Application
ofTaguchi-genetic
method todesign
optimal grey-fuzzy
controller ofa constant
turning
forcesystems",
Proc. of the 15th CSME AnnulConference,
Taiwan,
pp.313-38.
[19]
J. T.Tsai,
T. K. Liu and J. H. Chou.2004,
"Hybrid
Taguchi-genetic algorithm
forglobal
numericaloptimization",
IEEETrans.on
Evolutionary Computation,
Vo.8,
pp.365-377.
[20]
P.Gahinet,
A.Nemirovski,
A. J. Laub and M.Chilali, 1995,
LMIControl
Toolbox,
Massachusetts:The Math Works Inc.[21]
M. Gen and R.Cheng,
1997,
GeneticAlgorithms
andEngineering Design,
NewYork: JohnWiely
and Sons.[22]
G.Taguchi,
S.Chowdhury
and S.Taguchi,
2000,
RobustEngineering,
NewYork: McGraw-Hill.[23]
Y.Wu,2000,
Taguchi
MethodsforRobustDesign,
NewYork: The AmericanSociety
ofMechanicalEngineers.
[24]
K. S. Narendra and S. S.Tripathi,
1973, "Identification andoptimization
of aircraftdynamics",
J. ofAircraft, Vol.10,
pp.193-199. 0.7-S,0.66 a) 8 0.64 0.58 F-15 Generatiton 20 25 3
Figure 1. Convergence of H2 tracking performance (J for the
AFTI/F-1 6control systeminExample2.
2.5 2 1.5 rw 0.5 05 105 0 2 4 6 10 12 14 16 18 20 Time(sec)
Figure 2. Desired output response (dash line) and actual output
response (solid line)ofthe uncertain AFTI/F-1 6 aircraft controlsysteminExample2.
4141
---U.bb'