Skip to main content

Full text of "DTIC ADA173748: Solution to the Algebraic Riccati Equation for Parabolic Systems."

See other formats


AD-A17J 748 SOLUTION TO THE ALGEBRAIC RICCATI EQUATION FOR 

PARABOLIC SVSTEHS(U) JOHNS HOPKINS UNIV BALTIHORE HD 
DEPT OF ELECTRICAL ENGINEERIN. . L R RIDDLE ET AL. 
UNCLASSIFIED 22 OCT 86 JHU/EE-86/21 NB0814-85-K-B255 F/G 12/1 















































•■■m r- 










Solution to the Algebraic 
Riccati Equation for Parabolic Systems * 

Laurence R. Riddle and Howard L. Weinert 

Department of Electrical Engineering 
The Johns Hopkins University 
Baltimore, MD 21218 

ABSTRACT 

This paper presents an analytical solution to the operator alge¬ 
braic Riccati equation (ARE) for selfadjoint parabolic systems. 
The solution to the operator ARE is important in the design of the 
steady-state, on-line filter for estimating the system’s states. This 
analytical solution is derived by considering the operator analog of 
Potter’s method of using the Hamiltonian system’s eigenvectors 
and eigenvalues to solve a finite-dimensional ARE. As an example 
of using this analytical solution, the steady-state filtering error 
covariance for the 2-D heat equation is studied. 

This work supported by the Office of Naval Research under Contract N00014-85-K- 









V. W-'\ .1 i.1 A -.1 ,V„Tf A’■J\k 


1. Introduction 



In this paper, we derive an analytical solution to the operator algebraic 
Riccati equation (ARE) on a Hilbert space H: 

AP+PA'+Q =PRP (1.1) 

when —A is a strongly positive (coercive) selfadjoint operator that generates a 

continuous semigroup, Q is positive-definite, bounded and commutes with A, 
and R is bounded and nonnegative-definite. We also assume that the domain 
of A is dense in H, and is contained in the domain of A , and that the positive 
square root of Q commutes with A. In the context of distributed parameter 
filtering and control, one needs a solution to Eq. (1.1) that is bounded, nonnega¬ 
tive, selfadjoint and that maps the domain of A into itself. Gibson [3] has 
shown that under the above assumptions, Eq. (1.1) has a unique solution with 
these properties. 

An example where one would solve Eq.(l.l) with the above assumptions on 
A , Q and R is the design of a steady-state filter to estimate a process governed 
by the heat equation: 


u(x,t) — V u(x,t)+e(x,t ) , x G H C R 3 , <>0 (1.2) 

dt 


- u(x,t) = 0 , x G dCl 

dn 

d 

where denotes the outward normal derivative, and the observations are 
dn 

Vj{t) = f Cj(x)u(x,t)dx+Wj(t) , j = 1, • • • ,N 
n 

In Eq (1.2), e(x,J) is a random process that is white in space and time with 







constant intensity Q, w-{t) is a scalar white noise process with intensity R~ , 
and A — V 2 . Several applications using this model to describe the tempera¬ 
ture distribution of heated metals have been reported [l],[9]. Numerical exam¬ 
ples of calculating the steady-state filtering error covariance P for this model 
are given in [ll]. 


2. Solving the ARE 

In this section we will solve Eq. (l.l) by using the operator analog to the 
Potter method [10] of solving the ARE in finite dimensions. The Potter method 
of solving the ARE is summarized as follows: let the Hamiltonian matrix L be 


defined as 


A Q 


R -A 


and diagonalize L via 


where 


LM = MA 


A = diag(\, ■ ■ ■ ,\ n ) , * • • \ > 0 


M = 


■M n M 12 


M 2 i A/gj 


Then 


p = m 11 m 2 " 1 


is a nonnegative symmetric solution to the ARE , assuming A/ 21 is invertible. 













To justify the formal solution given by Eq. (2.3) we need to show that P is 
bounded, nonnegative, selfadjoint, and maps the domain of A into itself. Since 
—A > 0 it follows that A +R has a unique positive square root that is selfad¬ 
joint and that the domain of A 2 is dense in the domain of (A 2 -kR)*[7, p. 281]. 
Therefore —A +(A +R ) is strongly positive and selfadjoint with dense domain: 


D{-A+{A 2 +Rf) =D{-A) nD({A 2 +R)*)CZD{A 2 )CD(A) (2.4) 

where D( ) denotes domain; so that P x as defined in Eq. (2.2) is bounded [ 4, p. 

209 ]. Furthermore, P l is selfadjoint and nonnegative because (—A+(A 2 +R) k ) 

is [ 7, p. 272]. We will prove that P x maps the domain of A into the domain of 









..1 VlV 


The range of F, is contained in the domain of A by Eq. (2.4), and so it is 
sufficient to show that the domain of P { contains the domain of A. Since P x is 
bounded and necessarily closed (being selfadjoint), it follows [7, p. 269] that the 
domain of is H, and hence contains the domain of A. 

Since we have assumed that the operator S commutes with A, if x G D(A) 
then Sx £ D (A), hence P in Eq. (2.3) also maps D(A) into D(A). We have 
thus shown that the operator defined in Eq. (2.3) is the correct solution to Eq. 
(1.1) needed for filtering and control applications. 


3. Numerical Considerations and Example 

The analytical approach to solving the operator ARE presented in this 
chapter provides a computationally faster way of implementing the optimal 
gain for filtering and control applications. In our approach, finite-dimensional 
approximations to the operators in Eq (1.1) are done after the analytical solu- 
tion has been obtained, resulting in a computational complexity of 0(6n ), 
where n is the size of the matrix that approximates the operator A , whereas in 
other techniques [2] one approximates Eq (l.l) by a matrix ARE and then solves 
this equation using algorithms developed for the finite-dimensional case which 
require approximately 0(75n ) operations [8]. We remark that a similar 
approach (with the same computational complexity 0(6n ) ) to the finite¬ 
dimensional ARE has been considered [5],[6]. 

Another important difference between the analytical approach used in this 
chapter and those using a high order finite-dimensional approximation to Eq. 
































(1.1), is that the implementation of Eq. (2.3) calls for approximations of A and 

2 2 
A separately, whereas in approximating Eq. (1.1) one is implicitly using X , 

where X is a finite-dimensional approximation to A, as an approximation to A . 

The fact that X may not be a good approximation to A suggests that our 

approach may be more accurate. 

The steady-state filtering error covariance for estimating the temperature 
profile of a heated square aluminum slab was calculated using the results 
obtained in the previous section. These calculations were based on the follow¬ 
ing model for the variation in temperature u (x,y,t) above an assumed known 
ambient temperature T Q : 

d 2 

— u(x,y,t) — <*V u(a:,y,t)+e(a:,y,t) 0 < x , y < L 
dt 

with boundary conditions 


«(0 ,y,t) — u(L,y,t) — u(x,0,t) = u(x,L,t) = 0 

where 


■—5 2 / 

L — Imeter , a = 5 X 10 meter /second 
and the observations are 


y{t) = u{.5,.5,t)+w{t ) 

The input e(x,y,t) is assumed to be white in space and time with intensity 

Q — 1 (degreeC) /{meter ) second 
and the observation noise w(t) is white with intensity 

R~ X = 10 ~ 2 {degreeC) 2 /second 


In Eq. (2.3) ,a 10 X 10 finite-difference approximation to the operators A and 








































A was used. Figure 1 shows a cross-section of the variances of the tempera¬ 
ture variations u(x,.5,t) and the filtering error It is evident from Fig¬ 

ure 1 that the unobserved temperatures more that 10 cm away from the sensor 
are not being estimated very effectively. This result places some doubt on the 
possibility of estimating, with few sensors, the unobserved temperature distribu¬ 
tion of heated metals in the presence of spatially white process noise. 

References 

[1] M. B. Ajinkya, M. Kohne, M. F. Mader, and W. H. Ray, ‘ The experimen¬ 
tal implementation of a distributed parameter filter,’ Automatica, vol 11, 
pp. 571-577, 1975. 

[2] H.T. Banks and K. Kunisch, ‘The linear regulator problem for parabolic 
systems,’ SIAM J. Contr. Opt. ,vol. 22, pp 684-695, 1984. 

[3] J.S. Gibson, ‘The Riccati integral equations for optimal control problems 
on Hilbert spaces,’ SIAM J. Contr. Opt. vol. 17, pp 537-565, 1979. 

[4] K. E. Gustafson, Introduction to partial differential equations and Hilbert 
space methods, Wiley, New York, 1980. 

[5] M. Hamidi, ‘A method of optimal actuator and sensor placement’, Proc. 
Fourth IFAC Symp. Contr. Dist. Parm. Syst., Los Angeles, 1986. 

[6] F. Incertis, ‘ An extension on a new formulation of the algebraic Riccati 
equation problem’, IEEE Trans. Auto. Contr., vol AC-28, pp. 235-237, 1983. 



































[7] T. Kato, Perturbation theory for linear operators, Springer-Verlag, New 


York, 1966. 

[8] A.J. Laub, ‘A Schur method for solving algebraic Riccati equations,’ IEEE 
Trans. Auto. Contr., vol. AC-24, pp. 913-922,1979. 

[9] G. K. Lausterer, W. H. Ray and H. R. Martens, ‘The real time application 
of distributed parameter state estimation theory to a two dimensional 
heated ingot,’ pp. 451-467 of Proc. Second IF AC Symp. Contr. Dist. Parm. 
Syst., S. P. Banks, A. J. Pritchard Eds., Peragamon Press, Oxford,1978. 

[10] J.E. Potter, ‘Matrix quadratic solutions,’ SIAM J. Appl. Math., vol. 14, pp 
496-501, 1966. 

[11] K. Watanabe, M. Iwasaki, ‘A fast computational approach in optimal dis¬ 
tributed parameter state estimation,’ J. Dynamical Systems, Measurement 
and Contr., vol. 105, pp. 1-10, 1983. 











































location 




































UNC LASSIFIED _ 

SECUR'TV C-ASS'E'CATlQN QP THIS PAGE 



I, RERO^T SECURITY C LASS I P * C A T1 ON 

UNCLASSIFIED 


U- SECURITY CLASSIFICATION AUTHORITY 


3b OECLASSIPICATION/OOWNGRAOING SCHEDULE 


4 PERPCRMING ORGANIZATION REPORT NUMBERlS) 

JHU/EE-86/21 


REPORT DOCUMENTATION PAGE 


lb. RESTRICTIVE MARKINGS 


3. OISTRIBUTION/AVAILABILITY OP REPORT 

UNLIMITED 


5. MONITORING ORGANIZATION REPORT NUMBERlS) 


6*. NAME OP PERFORMING ORGANIZATION Bb. OPPlCE SYMBOL 7#. NAME OF MONITORING ORGANIZATION 

(If applicable! 

THE JOHNS HOPKINS UNIVERSITY OFFICE OF NAVAL RESEARCH 


6c. AOORESS tCity. Stott and ZIP Codtl 

CHARLES AND 34TH STREETS 
BALTIMORE, MARYLAND 21218 


S«. NAME OP PUNOING/SPONSORING 
ORGANIZATION 


8c AOORESS iCity. Stair and ZIP Codtl 


7b. AOORESS ICity. Stall and ZIP Codtl 

800 N. QUINCY ST. 
ARLINGTON, VA 22217 


8b. OFFICE SYMBOL 9. PROCUREMENT INSTRUMENT IDENTIFICATION NUMBER 
(If applicable) 

N00014-85-K-0255 


10. SOURCE OF FUNOING NOS. 


PROGRAM 
ELEMENT NO. 


1 1 TITLE (Include Security Clarification) SOLUTION TO THE ALGEBRAIC 
RICCATI EQUATION FOR PARABOLIC SYSTEMS (UNCLASSIFIED) 


12. PERSONAL AUTHOR(S) 

RIDDLE, L.R. AND WEINERT, H.L. 


134. TYPE OF REPORT 

INTERIM 


PROJECT 

TASK 

WORK UNIT 

NO. 

NO. 

NO- 


NR661-019 



rrrr« 

mMmmsMM 


14. OATE OF REPORT (Yr . Mo.. Day) 15. PAGE COUNT 

OCTOBER 22, 1986 9 



COSATl COOES 


18. SUBJECT TERMS (Continue on rwuerte if neceuary and identify by block number) 

OPERATOR RICCATI EQUATION, PARABOLIC SYSTEMS, DISTRIBUTED 
PARAMETER FILTERING AND CONTROL. 


19. ABSTRACT iContinue on reverte if necerary and identify by block number / 

This paper presents an analytical solution to the operator algebraic Riccati equation 
(ARE) for selfadjoint parabolic systems. The solution to the operator ARE is important 
in the design of the steady-state, on-line filter for estimating the system's states. 

This analytical solution is derived by considering the operator analog of Potter's method 
of using the Hamiltonian system's eigenvectors and eigenvalues to solve a finitei-dimension- 
al ARE. As an example of using this analytical solution, the steady-state filtering error 
covariance for the 2-D heat equation is studied. 


20 OISTRIBUTION/AVAILABILITY OP ABSTRACT 
UNCLASSIPIEO/UNLlMlTED £2 SAME AS RPT Z OTIC USERS □ 


22*. NAME OP RESPONSIBLE INDIVIDUAL 

DR. NEIL L. GERR 


DD FORM 1473, 83 APR EDITION OF 1 JAN 


21 ABSTRACT SECURITY CLASSIFICATION 

UNCLASSIFIED 


22b TELEPHONE NUMBER 
iI nclude Area Code) 

(202)696-4321 


EOITION OF 1 JAN 73 iS OBSOLETE. 


22c OFF ICE SYMBOL 


UNCLASSIFIED _ 

SECURITY CLASSIFICATION OF Tm-S PAGE