SENSITIVITY APPROACH TO OPTIMAL CONTROL FOR AFFINE NONLINEAR DISCRETE-TIME SYSTEMS

Similar documents
Chapter 3 Differentiation and Integration

A MODIFIED METHOD FOR SOLVING SYSTEM OF NONLINEAR EQUATIONS

Solving Fractional Nonlinear Fredholm Integro-differential Equations via Hybrid of Rationalized Haar Functions

OPTIMISATION. Introduction Single Variable Unconstrained Optimisation Multivariable Unconstrained Optimisation Linear Programming

Hongyi Miao, College of Science, Nanjing Forestry University, Nanjing ,China. (Received 20 June 2013, accepted 11 March 2014) I)ϕ (k)

A new Approach for Solving Linear Ordinary Differential Equations

36.1 Why is it important to be able to find roots to systems of equations? Up to this point, we have discussed how to find the solution to

Solution of Linear System of Equations and Matrix Inversion Gauss Seidel Iteration Method

Lecture 2 Solution of Nonlinear Equations ( Root Finding Problems )

A Hybrid Variational Iteration Method for Blasius Equation

: Numerical Analysis Topic 2: Solution of Nonlinear Equations Lectures 5-11:

Summary with Examples for Root finding Methods -Bisection -Newton Raphson -Secant

CHAPTER 4d. ROOTS OF EQUATIONS

CHAPTER 7 CONSTRAINED OPTIMIZATION 2: SQP AND GRG

CHAPTER 5 NUMERICAL EVALUATION OF DYNAMIC RESPONSE

EE 330 Lecture 24. Small Signal Analysis Small Signal Analysis of BJT Amplifier

Yong Joon Ryang. 1. Introduction Consider the multicommodity transportation problem with convex quadratic cost function. 1 2 (x x0 ) T Q(x x 0 )

ORDINARY DIFFERENTIAL EQUATIONS EULER S METHOD

General viscosity iterative method for a sequence of quasi-nonexpansive mappings

Chapter Newton s Method

Endogenous timing in a mixed oligopoly consisting of a single public firm and foreign competitors. Abstract

CISE301: Numerical Methods Topic 2: Solution of Nonlinear Equations

Finite Difference Method

Absorbing Markov Chain Models to Determine Optimum Process Target Levels in Production Systems with Rework and Scrapping

1 GSW Iterative Techniques for y = Ax

Solution for singularly perturbed problems via cubic spline in tension

Off-policy Reinforcement Learning for Robust Control of Discrete-time Uncertain Linear Systems

Irregular vibrations in multi-mass discrete-continuous systems torsionally deformed

Transfer Functions. Convenient representation of a linear, dynamic model. A transfer function (TF) relates one input and one output: ( ) system

An explicit solution to polynomial matrix right coprime factorization with application in eigenstructure assignment

Single Variable Optimization

Study on Active Micro-vibration Isolation System with Linear Motor Actuator. Gong-yu PAN, Wen-yan GU and Dong LI

A New Refinement of Jacobi Method for Solution of Linear System Equations AX=b

THE GUARANTEED COST CONTROL FOR UNCERTAIN LARGE SCALE INTERCONNECTED SYSTEMS

Inexact Newton Methods for Inverse Eigenvalue Problems

Numerical Differentiation

Convexity preserving interpolation by splines of arbitrary degree

On the Global Linear Convergence of the ADMM with Multi-Block Variables

Note 10. Modeling and Simulation of Dynamic Systems

Application of B-Spline to Numerical Solution of a System of Singularly Perturbed Problems

COEFFICIENT DIAGRAM: A NOVEL TOOL IN POLYNOMIAL CONTROLLER DESIGN

Numerical Solutions of a Generalized Nth Order Boundary Value Problems Using Power Series Approximation Method

EEE 241: Linear Systems

Newton s Method for One - Dimensional Optimization - Theory

Adjoint Methods of Sensitivity Analysis for Lyapunov Equation. Boping Wang 1, Kun Yan 2. University of Technology, Dalian , P. R.

Global Sensitivity. Tuesday 20 th February, 2018

Lecture 13 APPROXIMATION OF SECOMD ORDER DERIVATIVES

Department of Electrical and Computer Engineering FEEDBACK AMPLIFIERS

NON-CENTRAL 7-POINT FORMULA IN THE METHOD OF LINES FOR PARABOLIC AND BURGERS' EQUATIONS

On a direct solver for linear least squares problems

Operating conditions of a mine fan under conditions of variable resistance

A New Recursive Method for Solving State Equations Using Taylor Series

Interactive Bi-Level Multi-Objective Integer. Non-linear Programming Problem

Scroll Generation with Inductorless Chua s Circuit and Wien Bridge Oscillator

Numerical method for a class of optimal control problems subject to nonsmooth functional constraints

An L 2 Disturbance Attenuation Approach. to the Nonlinear Benchmark Problem. where is the (nondimensionalized) translational position

Assessment of Site Amplification Effect from Input Energy Spectra of Strong Ground Motion

Binomial transforms of the modified k-fibonacci-like sequence

829. An adaptive method for inertia force identification in cantilever under moving mass

risk and uncertainty assessment

Inverse Displacement Analysis of a General 6R Manipulator Based on the Hyper-chaotic Least Square Method

MATH 5630: Discrete Time-Space Model Hung Phan, UMass Lowell March 1, 2018

Septic B-Spline Collocation Method for the Numerical Solution of the Modified Equal Width Wave Equation

On the Interval Zoro Symmetric Single-step Procedure for Simultaneous Finding of Polynomial Zeros

Chapter 5 Function-based Monte Carlo

Chapter 6. Operational Amplifier. inputs can be defined as the average of the sum of the two signals.

Foundations of Arithmetic

Control of Uncertain Bilinear Systems using Linear Controllers: Stability Region Estimation and Controller Design

, rst we solve te PDE's L ad L ad n g g (x) = ; = ; ; ; n () (x) = () Ten, we nd te uncton (x), te lnearzng eedbac and coordnates transormaton are gve

Wavelet chaotic neural networks and their application to continuous function optimization

Linear Approximation with Regularization and Moving Least Squares

The Minimum Universal Cost Flow in an Infeasible Flow Network

Numerical Methods. ME Mechanical Lab I. Mechanical Engineering ME Lab I

System in Weibull Distribution

CRITICAL POINT ANALYSIS OF JOINT DIAGONALIZATION CRITERIA. Gen Hori and Jonathan H. Manton

New Exact Traveling Wave Solutions for Two Nonlinear Evolution Equations

Controller Design for Networked Control Systems in Multiple-packet Transmission with Random Delays

Chapter - 2. Distribution System Power Flow Analysis

IN the theory of control system, an optimal control problem

COMPARISON OF SOME RELIABILITY CHARACTERISTICS BETWEEN REDUNDANT SYSTEMS REQUIRING SUPPORTING UNITS FOR THEIR OPERATIONS

A CHARACTERIZATION OF ADDITIVE DERIVATIONS ON VON NEUMANN ALGEBRAS

Curve Fitting with the Least Square Method

Nonlinear Network Structures for Optimal Control

EXACT TRAVELLING WAVE SOLUTIONS FOR THREE NONLINEAR EVOLUTION EQUATIONS BY A BERNOULLI SUB-ODE METHOD

Module 3 LOSSY IMAGE COMPRESSION SYSTEMS. Version 2 ECE IIT, Kharagpur

Combined Wronskian solutions to the 2D Toda molecule equation

Shuai Dong. Isaac Newton. Gottfried Leibniz

Nonlinear n-th Cost Cumulant Control and Hamilton-Jacobi-Bellman Equations for Markov Diffusion Process

The Study of Teaching-learning-based Optimization Algorithm

Least-Squares Solutions of Generalized Sylvester Equation with Xi Satisfies Different Linear Constraint

College of Computer & Information Science Fall 2009 Northeastern University 20 October 2009

The Analysis of Convection Experiment

Iterative General Dynamic Model for Serial-Link Manipulators

Numerical Methods Solution of Nonlinear Equations

arxiv: v1 [math.co] 12 Sep 2014

Salmon: Lectures on partial differential equations. Consider the general linear, second-order PDE in the form. ,x 2

Remarks on the Properties of a Quasi-Fibonacci-like Polynomial Sequence

A Fast Computer Aided Design Method for Filters

ME 501A Seminar in Engineering Analysis Page 1

(Online First)A Lattice Boltzmann Scheme for Diffusion Equation in Spherical Coordinate

Transcription:

448 Asan Journal o Control Vol. 7 No. 4 pp. 448-454 December 25 Bre Paper SENSIIVIY APPROACH O OPIMAL CONROL FOR AFFINE NONLINEAR DISCREE-IME SYSEMS Gong-You ang Nan Xe Peng Lu ABSRAC hs paper deals wth the optmal control problem or a class o ane nonlnear dscrete-tme systems. By ntroducng a senstvty parameter epng the system varables nto a Maclaurn seres around t we transorm the orgnal optmal control problem or ane nonlnear dscrete-tme systems nto the optmal control problem or a sequence o lnear dscrete- tme systems. he optmal control law conssts o an accurate lnear term a nonlnear compensatng term whch s an nnte sequence o adjont vectors. In the present approach teraton s requred only or the nonlnear compensaton seres. By nterceptng a nte sum o the seres we obtan a suboptmal control law that reduces the complety o the calculatons. A numercal smulaton shows that the algorthm can be easly mplemented has a ast convergence rate. KeyWords: Dscrete-tme systems nonlnear systems optmal control senstvty approach. I. INRODUCION Physcal systems are nherently nonlnear n nature. he optmal control or nonlnear systems s one o the most mportant subjects o research n the eld o control theory. Lee studed the suboptmal local eedback control or a class o constraned nonlnear dscrete-tme control problems n []. Hager [2] Nagurka [3] Stojano [4] ang [5] also gave some mportant results. For the quadratc perormance nde the optmal control problem oten leads to a Hamlton-Jacob-Bellman (HJB) equaton or a nonlnear two-pont boundary value (PBV) problem. For the general optmal control problem or nonlnear systems however an analytcal soluton does not est. Manuscrpt receved December 8 23; accepted January 6 25. he authors are wth College o Inormaton Scence Engneerng Ocean Unversty o Chna Qngdao 2667 PRChna. Nan Xe s also wth School o Computer Scence echnology Shong Unversty o echnology Zbo 25549 P.R Chna. (emal: gtang@ouc.edu.cn). hs work was supported by the Natonal Natural Scence Foundaton o Chna (674) the Natural Scence Foundaton o Shong Provnce (Y2G2). hs has nspred researchers to propose approaches to obtanng an appromate soluton to the HJB equaton or the nonlnear PBV problem as well as to obtanng an appromate optmal control or nonlnear systems. One o these approaches s power seres appromaton (PSA). hs approach nvolves usng a power seres [6-9] or so-called Adoman decomposton [] to separate the nonlnear terms appromate the soluton o the HJB equaton. In order to obtan the appromate optmal control law n a seres orm ths approach needs an teratve soluton o a seres o HJB equatons (nonlnear matr derental equatons). Another approach s successve Galerkn appromaton (SGA) where an teratve process s used to nd a sequence o appromatons approachng the soluton o the HJB equaton [-3] whch s done by solvng a sequence o generalzed HJB equatons. However beng an teratve method SGA s dependent on the teratve ntal value. I t s not well chosen the method may converge very slowly or even not converge at all. Recently ang [4] presented a successve appromaton approach to optmal control or nonlnear systems wth a quadratc perormance nde. he optmal control law can be obtaned by an adjont vector sequence. Other approaches have been presented n [5-7].

G.-Y. ang et al.: Senstvty Approach to Optmal Control or Ane Nonlnear Dscrete-me Systems 449 he man objectve o ths artcle s to develop a new PSA optmal control algorthm based on the senstvty approach or dscrete-tme nonlnear systems. By ntroducng a senstvty parameter we transorm the optmal control problem or ane nonlnear dscrete-tme systems nto an optmal control problem or lnear dscrete-tme systems. In the present approach the optmal control law conssts o an accurate lnear term a nonlnear compensatng seres. Iteraton s requred only or the nonlnear compensaton seres.e. the soluton o a sequence o adjont vector derence equatons. By nterceptng a nte summaton o the seres we obtan a suboptmal control law that can reduce the complety o the calculatons. hs paper s organzed as ollows. In secton 2 by ntroducng a senstvty parameter epng the system varables nto a Maclaurn seres wth respect to that parameter we obtan a suboptmal control law. In secton 3 a numercal eample s gven to llustrate the proposed approach. Several conclusons are drawn n the last secton. II. DESIGN OF SUBOPIMAL CONROL Consder the ollowng ane nonlnear dscrete-tme systems: ( k+ ) = g( ( k)) + Bu( k) k N () = () where R n s the state vector u R r s the control vector B s an n r constant matr s the ntal state vector N = { 2 }. Assume that g C (R n U R n ) s a polynomal vector uncton that g() =. he nonlnear uncton g can be rewrtten n the ollowng orm: g( k ( )) = Ak ( ) + ( k ( )) (2) where s a nonlnear polynomal vector uncton o A = g()/. hereore system () can be rewrtten as ( k+ ) = A( k) + ( ( k)) + Bu( k) k N () =. (3) he man objectve o ths paper s to nd an optmal control law u (k) whch mnmzes the quadratc perormance nde N J = ( N) Q ( N) + [ ( k) Q( k) + u ( k) Ru( k)] 2 2 k = (4) subject to () where R R r r s a postve-dente matr Q Q R n n are sem-postve-dente matrces. It s well known that the optmal control problem n (3) (4) oten leads to the ollowng soluton o a nonlnear PBV problem: λ ( k) = Q( k) + A λ ( k + ) + ( ( k)) λ ( k + ) ( k+ ) = Ak ( ) + Buk ( ) + ( k ( )) k N (5a) wth the boundary condtons () = λ ( N) = Q ( N) (5b) where = / N = { 2 N }. he optmal control law s uk ( ) = R B λ ( k+ ) k N. (6) Unortunately n general an analytcal soluton or ths nonlnear PBV problem does not est. hereore t s necessary to nd appromaton approaches to solvng the optmal control problem or nonlnear dscrete-tme systems. We propose a senstvty approach that can smply the PBV problem n (5) get the optmal control law. Construct the ollowng PBV problem n whch a senstvty parameter s ntroduced: λ( k ) = Q( k ) + A λ ( k+ ) + ( ( k )) λ ( k+ ) k ( + ) = Ak ( ) + Buk ( ) +( k ( )) k N (7a) wth boundary condtons ( ) = λ( N ) = Q ( N) (7b) the control law uk ( ) = R Bλ ( k+ ) k N (8) where. In the ollowng we assume that u(k ) (k ) λ(k ) are nntely derentable wth respect to around =. Epng them nto a Maclaurn seres we obtan uk ( ) = u ( k) ( k ) = ( k) =! =! λ( k ) = λ ( k) (9)! = where the superscrpt () denotes the th-order dervatve o the seres wth respect to when =. Obvously the PBV problem n (7) control law (8) are equvalent to the orgnal problem n (5) (6) when = respectvely. Note that the Maclaurn seres ((k )) can be eped as

45 Asan Journal o Control Vol. 7 No. 4 December 25 = ( ) ( k ( )) ( ( k )) = ()! that ( ( k )) λ ( k+ ) can be eped as ( ( k )) λ ( k+ ) where j ( j) ( j) C k k = = [( )!] ( ) ( ( )) λ ( + ) ( ) j! k C λ ( ) = =. j!( j)! = = Substtutng (9) () () nto (7) we obtan = ( k+ )! ( ) = A ( k) ( ( k)) Bu ( k) + +! λ ( k)! j ( j) ( = Q ( k) + C ( ) ( ( k)) λ ( k + ) = + A λ ( k+ )! u () ( k ) ( ).! R B k = λ +! k N = = () (2) (3) From (3) we can easly obtan ( ) ( k + ) = A ( k) + ( ( k)) + Bu ( k) j ( j) λ ( k) = Q ( k) + C ( ) ( ( k)) ( ( ) λ ( k + ) + A λ ( k + ) u ( k) = R B λ ( k+ ) k N (4a) wth boundary condtons () () () = λ ( N) = Q ( N) () = λ ( N) = I + (4b) where I + = { 2 }. When = we get the th-order PBV problem () () () ( k + ) = A ( k) + Bu ( k) () () () λ ( k) = Q ( k) + A λ ( k+ ) () () u ( k) = R B λ ( k+ ) k N (5a) wth boundary condtons () () () = λ ( N) = Q ( N). (5b) Lettng () () λ ( k) = P( k) ( k) k N (6) substtutng (6) nto (5) we obtan the ollowng th-order optmal control law: () () u ( k) = S ( k + ) B P( k + ) A ( k) k N. (7) Here S(k) = R + B P(K) B P(k) s the postve-dente soluton o the ollowng Rccat matr derence equaton: Pk ( ) = A I Pk ( ) BS ( k ) B + + Pk ( + ) A+ Q PN ( ) = Q. k N (8) When = 2 we get the th-order PBV problem ( ) ( k+ ) = A ( k) + Bu ( k) + ( ( k)) λ ( k) = Q ( k) + A λ ( k+ ) j ( j) ( + C ( ) ( ( k)) λ ( k + ) u ( k) = R B λ ( k + ) k N (9a) wth boundary condtons () = λ ( N) =. (9b) Note that () j ( j) ((k)) C ( ) ( ( k)) ( j λ ) ( k + ) are known unctons whch have been obtaned n the ( )th teraton then the PBV problem n (9) becomes a lnear nonhomogeneous one. In order to solve ths problem we let λ ( k) = P( k) ( k) + g ( k) k N. (2) Substtutng (2) nto (9) we can decouple the PBV problem (9) get the th costate equaton

G.-Y. ang et al.: Senstvty Approach to Optmal Control or Ane Nonlnear Dscrete-me Systems 45 g ( k) = A I P( k + ) BS ( k + ) B g k+ + P k + k ( ) ( ) ( ) ( ( )) + C k λ k + k N j ( j) ( ( ) ( ( )) ( ) g ( N) = I +. (2) By solvng (2) we can get the soluton o g (k). Substtutng (2) nto (9) we can also get ( k + ) = I + BR B P( k+ ) ( ) A ( k) BR B g ( k ) ( ( k)) + + () = I + (22) ( ) u ( k) = S ( k + ) B P( k + ) A ( k) + ( ( k)) S ( k + ) B g ( k + ) I +. (23) Lettng g (k) substtutng (7) (9) (23) nto (9) we can get uk ( ) = S ( k+ ) B =! ( ) =! ( Pk ( + ) A ( k) + g( k+ )) + P( k + ) ( ( k)) (24) Summarzng the above we obtan the ollowng theorem. heorem. Consder the problem o mnmzng the cost unctonal (4) subject to system (3); the control law * u ( k) = S ( k+ ) B Pk ( + )( Ak ( ) + ( k ( ))) + = g ( k + )! (25) s optmal where P(k) g (k) are solved by (8) (2) respectvely. Remark. From optmal control law (25) we know that an accurate result o g ( k + ) (!) s nearly mpossble = to get. By replacng wth M n (25) we can get a suboptmal control law as ollows: u ( k) = S ( k+ ) B M M g ( ) ( )( ( ) ( ( ))) k+ Pk+ Ak + k +. =! (26) In the ollowng we present an teratve procedure or ndng the suboptmal control law. Denote g ( k+ ) h ( k) = S ( k+ ) B = 2 M (27)! [ ] u ( k) = S ( k+ ) B P( k+ ) A( k) + ( ( k)). (28) From (26) we obtan u ( k) = u ( k) + h( k) = 2 M. (29) Remark 2. In (2) we can calculate () ((k)) as ollows: j n d ( ) () k j j! = d ( ) () =. (3) j n d ( ) n () k j! d j n d 2 ( ( )) k k j! d Remark 3. When N the quadratc perormance nde (4) can be descrbed as J = ( k) Q( k) u ( k) Ru( k) 2 + (3) k = (8) (2) respectvely become APA P APBS BPA+ Q= (32) ( ) g( k) = A I PBS B g( k + ) + P ( ( k)) + C k λ k + j ( j) ( ( ) ( ( )) ( ) g ( N) = I + (33) where ( k+ ) = I + BR B P ( ) A ( k) BR B g ( k ) ( ( k)) + +

452 Asan Journal o Control Vol. 7 No. 4 December 25 () = k N. (34) hereore we obtan the suboptmal control law M g ( k + ) um ( k) = S B P( A( k) + ( ( k))) +. =! (35) Smlarly to (27)-(29) the teratve procedure or ndng the suboptmal control law n (35) s as ollows. Denote g ( k + ) h ( k) = S B! = 2 M (36) [ ] u ( k) = S B P A( k) + ( ( k)). (37) From (35) we obtan 2 ( k) 2 ( k) ( ) =.4 2 3( k) 2( k) ( k) 2( k) 3( k). (39) + + 4 ( k) he quadratc cost unctonal s chosen as J = k + k + k + k + u k 2 2 2 2 2 2 ( ( ) 2( ) 3( ) 4( ) ( )). (4) 2 k = When we choose the suboptmal control law when = 2 3 4 the values o the quadratc cost uncton are J =.86 J = 8.583 J 2 = 8.42 J 3 = 7.738 J 4 = 7.4765 respectvely. Lettng =.5 we can get (J 4 J 3 ) /J 4 =.34 < then we can stop teratng. Smulaton results are presented n the ollowng. u ( k) = u ( k) + h ( k) = 2 M. (38) o nd the suboptmal control law (26) we propose the ollowng algorthm or the system n (3) wth the quadratc perormance nde n (4). Algorthm. Step. Work out the value o P rom epresson (8) gven α > ; Step 2. obtan the th-order control law u (k) rom (5) calculate the value o J lettng = ; Step 3. work out g (k) rom (2); Step 4. obtan the th-order control law u (k) rom (29) calculate the value o J ; Step 5. ( J J )/J < α then let M = obtan the output u M (k) stop. Else let = + turn to step 3. Fg.. he curves o (k) when = 2 4. Remark 3. When N we can easly calculate P g (k) u (k) J rom (32) (33) (38) (3) respectvely. Usng the above algorthm we can get the suboptmal control law (35). III. A NUMERICAL EXAMPLE Consder the 4th-order dscrete-tme nonlnear system descrbed by (3) where..2.5 A= B = () =...2.2.5 Fg. 2. he curves o 2 (k) when = 2 4.

G.-Y. ang et al.: Senstvty Approach to Optmal Control or Ane Nonlnear Dscrete-me Systems 453 the suboptmal control law manly result rom g M. he weaker the nonlnear term s the smaller the errors wll be whch brngs the suboptmal control trajectory closer to the theoretc optmal control law reduces the number o teratons M. IV. CONCLUSION Fg. 3. he curves o 3 (k) when = 2 4. he man contrbutons o ths paper are summarzed as ollows. By ntroducng a senstvty parameter epng the system varables nto a seres we tranorm the optmal control problem or ane nonlnear dscrete-tme systems nto an optmal control problem or lnear dscrete- tme systems. A suboptmal control law has been obtaned that reduces the complety o the calculatons. An algorthm or solvng the suboptmal control law has also been gven. REFERENCES Fg. 4. he curves o 4 (k) when = 2 4. Fg. 5. he curves o u(k) when = 2 4. From the above gures t s clear that the larger the number o teratons the better the control precson. When = 4 the derence between the values o the quadratc cost uncton based on the 4th-order control law on the 3rd-order control law s small enough. hs ndcates that the 4th suboptmal control law u 4 s very close to the optmal control law u. It can be seen that the errors due to. Lee H.W.J. K.L. eo V. Rehbock Suboptmal Local Feedback Control or a Class o Constraned Dscrete me Nonlnear Control Problems Comput. Math. Appl. Vol. 36 pp. 33-48 (998). 2. Hager W.W. Multpler Methods or Nonlnear Optmal Control SIAM J Numer. Anal. Vol. 27 pp. 6-8 (99). 3. Nagurka M.L. V. Yen Fourer-Based Optmal Control o Nonlnear Dynamc Systems rans. ASME J. Dyn. Syst. Meas. Contr. Vol. 2 pp. 7-26 (99). 4. Stojano V.S. Optmal Dampng Control Nonlnear Ellptc Systems SIAM J. Contr. Optm. Vol. 29 pp. 594-68 (99). 5. ang G.-Y. Z.-W. Luo Suboptmal Control o Lnear Systems wth State me-delay Proc. IEEE Con. Syst. Man Cybern. okyo pp. 4-9 (999). 6. Garrard W.L D.F. Enns S.A. Snell Nonlnear Feedback Control o Hghly Manoeuvrable Arcrat Int. J. Contr. Vol. 56 No. 4 pp. 799-82 (992). 7. Nshkawa Y. N. Sannomya H. Itakura A Method or Suboptmal Desgn o Nonlnear Feedback Systems Automatca Vol. 7 pp. 73-72 (97). 8. ang G.-Y. H.-P. Qu Y.-M. Gao Senstvty Approach o Suboptmal Control or a Class o Nonlnear Systems J. Ocean Unv. Qngdao Vol. 32 pp. 65-62 (22) (n Chnese). 9. ang G.-Y. X.-H. Zhao Y.-J. Lu Senstvty Approach o Optmal Control or Lnear Dscrete Large-Scale Systems wth State me-delay Proc. World Cong. Intell. Contr. Autom. Vol. Hangzhou Chna pp. 5-9 (24).. Chanane B. Optmal Control o Nonlnear Systems:

454 Asan Journal o Control Vol. 7 No. 4 December 25 A Recursve Approach Comput. Math. Appl. Vol. 35 pp. 29-33 (998).. Beard R.W. G.N. Sards J.. Wen Galerkn Appromaton o the Generalzed Hamlton-Jacob- Bellman Equaton Automatca Vol. 33 pp. 259-277 (997). 2. Ral W.B. W.M. mothy Successve Galerkn Appromaton Algorthms or Nonlnear Optmal Robust Control Int. J. Contr. Vol. 7 pp. 77-743 (998). 3. Sards G.N. C.S.G. Lee An Appromaton heory o Optmal Control or ranable Manpulators IEEE rans. Syst. Man Cybern. Vol. 9 pp. 52-59 (979). 4. ang G.-Y. Suboptmal Control or Nonlnear Systems: A Successve Appromaton Approach Syst. Contr. Lett. Vol. 54 No. 5 (25). 5. Banks S.P. Eact Boundary Controllablty Optmal Control or a Generalsed Korteweg de Vres Equaton Nonln. Anal. Vol. 47 No. 8 pp. 5537-5546 (2). 6. Goh C.J. On the Nonlnear Optmal Regulator Problem Automatca Vol. 29 pp. 75-756 (993). 7. Huang C.S. S. Wang K.L. eo Solvng Hamlton-Jacob-Bellman Equatons by a Moded Method o Characterstcs Nonln. Anal. Vol. 4 pp. 279-293 (2).