sequential quadratic programming methods for optimal control problems with state constraints
TRANSCRIPT
Applied Mathematics
A J o u r n a l of Chinese Universit ies
Vol. 8 Ser. B No. 2 Dec. 1993
S E Q U E N T I A L Q U A D R A T I C P R O G R A M M I N G
M E T H O D S F O R O P T I M A L C O N T R O L P R O B L E M S
W I T H S T A T E C O N S T R A I N T S "
Xu C h e n g x i a n ( ~~,~)
(Department o f Mathematics Xi* an diaotong t nwersitg . 71 0 0 4 9 )
Jong de J. L.
(Technology ( n i v e r s i t y o f Eindhuvel~ , Itolland )
Abstract
A kind of direct methods is presented for the solution of optimal control problems with state
constraints. These methods are sequential quadrat ic programming methods. At every iteration a quadratic
programming which is obtained by quadratic approximation to Lagrangian function and linear
approximations to constraints is solved to get a search direction for a meri t function. The meri t function is
formulated by augmenting the Lagrangian function with a penalty term. A line search is carried out along
the search direction to determine a step length such that the meri t function is decreased. The methods
presented in this paper include continuous sequential quadrat ic programming methods and disereate
sequential quadratic programming methods.
Key Words Optimal Control Problems with State Constraints, Sequential Quadrat ic Programming,
Lagrangian Function, Meri t Function, Line Search.
l . Introduction
A state-constrained optimal control problem is to determine a control function )~ C- ( ' : ' ( R
R ~) . a s t a t e trajectory z E CZ( R ~ R' ) to minimize the functional
ho(x(O)) + [".ro(x(t). , , (n. t)dt + g~O.(r)) (1. 1)do
Received on May, 8. 1992.
164 Appl ied Mathemat ies - -A J o u r n a l o f C h i n e s e Unive r s i t i e s Vol. 8 Ser. B
s u b j e c t to f o l l o w i n g constraints
dx) 7 = f ( x ( t ) , u ( t ) , t ) , 0 ~ t ~ T , ( 1 . 2 )
D ( x ( O ) ) ---- 0 , ( 1 . . 3 )
E ( x ( T ) ) ---- O, ( 1 . 4 )
w h e r e h 0 : R ' - - ~ R , f 0 : R~ X Re X R--~ R , go: R'---~ R , f : R" X R"~ t R---~ R' , D: R'---~ Rp , E :
R ' --~ Rq a r e all t w i c e c o n t i n u o u s l y differentiable functions with respect to t h e i r a r g u m e n t s , a n d T is
the f i x e d f i n a l t i m e . The constraints are e q u a t i o n s o f the d y n a m i c b e h a v i o u r o f the s y s t e m
( e x p r e s s e d by ( 1 . 2 ) ) , initial state constraints ( g i v e n by ( 1 . 3 ) ) a n d t e r m i n a l s t a t e constraints ( i n
( 1 . 4 ) ) . This kind o f p r o b l e m s a r i s e in practice w h e n t h e r e is a d e m a n d to control a s y s t e m from
one state to another in some optimal s e n s e , for e x a m p l e , f l i g h t path optimization o f air p l a n e s a n d
s p a c e v e h i c l e s , econometrics or robotics.
N u m e r i c a l m e t h o d s for the solution of the state-constrained optimal control p r o b l e m s can be
d i v i d e d into t w o k i n d s d i r e c t a n d indirect m e t h o d s . The d i r e c t m e t h o d i s s t a r t e d with an initial
approximation to the solution a n d the approximation i s i m p r o v e d iteratively by minimizing the
objective f u n c t i o n a l , a u g m e n t e d with a p e n a l t y t e r m , a l o n g a s e a r c h direction w h i c h is o b t a i n e d by
approximation to the p r o b l e m . In d i r e c t m e t h o d s , u ( t ) i s treated a s v a r i a b l e s o f the minimization
p r o b l e m , x ( t ) can be t r e a t e d either a s a q u a n t i t y d e p e n d e n t on the control u ( t ) by u s i n g the
differential e q u a t i o n s o r a s v a r i a b l e s o f the p r o b l e m w h i l e the differential e q u a t i o n s are r e g a r d e d a s
e q u a l i t y constraints. I n d i r e c t methods use the optimality conditions to d e r i v e a multipoint b o u n d a r y
v a l u e p r o b l e m , a n d the n u m e r i c a l solution o f the multipoint b o u n d a r y v a l u e p r o b l e m y i e l d s a
c a n d i d a t e for the solution o f the optimal control p r o b l e m . F o r indirect m e t h o d s the convergence
r e g i o n is g e n e r a l l y s m a l l a n d the g e n e r a t e d n u m e r i c a l solution is a c c u r a t e . H o w e v e r , the indirect
methods need the k n o w l e d g e o f the s t r u c t u r e o f the solution. F o r d i r e c t m e t h o d s , a l t h o u g h the
generated n u m e r i c a l solution i s not so accutate a s for the indirect m e t h o d s , its convergence r e g i o n i s
l a rge a n d it does not n e e d the k n o w l e d g e o f the solution.
2. Optimality Conditions
Optimality conditions play the central r o l e s in a n y solution m e t h o d s for optimization. S i n c e the
state-constrained optimal c o n t r a l problems a r e the special c a s e s of the f o l l o w i n g a b s t r a c t optimization
p r o b l e m ,
minimize f ( x ) ,r E X
s u b j e c t to h ( x ) ~ 0 , ( 2 . 1)
w h e r e f . X ~ R , h : X --~ Z , a n d X , Z are B a n a c h s p a c e s , the optimality conditions for p r o b l e m
No. 2 Optimal Control Problems with State Constraints I 6 5
( 1 . 1 ) - - ( 1 . 4 ) c a n be d e r i v e d from those of the p r o b l e m ( 2 . 1 ) .
( 1 ) The f i r s t o r d e r necessary conditions for problem ( 1 . 1 ) - - ( 1 . 4 ) can be s t a t e d a s f o l l o w s :
Let ( z , u ) be a solution to p r o b l e m ( 1 . 1 ) - - ( 1 . 4 ) , then there exist a real n u m b e r p ~ 0 , a
vector f u n c t i o n ~,,: [ O , T ] --~ R ' , vectors o" E Rp a n d ~ E /i~ s u c h that
/.(t:) ~ -- £(t~)~ = - - ~" I6EtJdt, for all 0 ~ t, < t~ ~ T, ( 2 . 2 )dq
w h e r e
,~.(0) * = - - pho,E0-I - 3~D, E0~,
~'.(T) r = pgo,E0-] + ;/E, ET],
( 2 . 3 )
( 2 . 4 )
Hit] = p f o ( x ( t ) , u ( t ) , t ) + ~.(t)rf(x(t),u(t),t), ( 2 . 5 )
is a I4amiltonian f u n c t i o n , the notation [ t ] s t a n d s for ( x ( t ) , u ( t ) , t ) o r ( z ( t ) , t ) , a n d the subscript
x d e n o t e s the derivatives respect to z . F u r t h e r m o r e , if
rank ( D , [ 0 ] ) = p , rank ( E , [ T ] ) = q , ( 2 . 6 )
a n d t h e r e is a pair ( 6 x , 5 , ) such that
D,[O-]6,(O) = 0, (2. 7)
E, ET]6,(T) = O, ( 2 . 8 )
6Jc(t) = f , EtJ6,(t) + f : E t ] 6 , ( t ) , 0 ~ t ~ T , ( 2 . 9 )
then the r e g u l a r i t y constant/~ is not zero a n d hence p = [ . Conditions (2 . 6 ) - - ( 2 . 9 ) a r e c a l l e d
constrained qualifications.
In o r d e r to give the s e c o n d o r d e r conditions, d e f i n e the set S o f feasible directions
S = {(6,,6,,)]D,U_0]6,(0) = 0 , E~[T]6~(T) = O,
6 ~ ( t ) = f : l - t ] 6 , [ t ] ± f,,7#Ja,,(t), 0 ~ t ~ T}.
( 2 ) The s e c o n d o r d e r n e c e s s a r y conditions
Let ( x , u ) be a l o c a l solution to problem ( 1 . 1 ) - - ( 1 . 4 ) a n d constrained qualifications ( 2 . 6 )
- - ( 2 . 9 ) be satisfied. If there e x i s t vector function ).(t) : [ 0 , T ~ - + R° , a n d vectors <~ ~ R~ a n d / )
Rq satisfying the f i r s t o r d e r n e c e s s a r y conditions at ( ~ ' , u ) , then
~/2L(x,u,~;,~,~)(6,,6,,)(6,.6,,) ~ O. V (6,,&,) E S , (2. 10)
w h e r e L(z,u,o-,).,/~) is the L a g r a n g i a n functional d e f i n e d by
t*T= f o ( x ( t ) , u ( t ) ,t)dt + go(x(T)) + . 'TrD(J'(0))L(x,u,cr,).,~u) h 0 ( x ( 0 ) ) T Io
166 Applied Mathernaties--A Journal of Chinese Universities Voi. 8 Ser, B
(3) The second order sufficient conditions
I f the point ( x , u ) i s feasible to p r o b l e m ( 1 . ] ) - ( 1 . 4 ) a n d t h e r e e x i s t multipliers ~ E R ' , / ]
/ ~ a n d 5.: [ 0 , T ] - ~ R~ s u c h that the f i r s t o r d e r n e c e s s a r y conditions a n d constrained qualifications a r e
satisfied a t ( x , u ) . M o r e o v e r , if t h e r e is a fl ~ 0 s u c h that
V'L(~,,, ,~,; . ,~)(¢,~,)(¢,~,,) >~ ~(:11¢1l~ + I1~,,11~) for all ( 6 . , 6 , , ) E S, (2. 12)
then ( x , u ) is a l o c a l solution to problem ( 1 . 1 ) - - ( 1 . 4 ) .
3. Continuous Sequential Quadratic Programming Method
B a s e d on the i d e a w h e n a n approximation is near the s o l u t i o n , a better approximation c a n be
o b t a i n e d by s o l v i n g a s u i t a b l e q u a d r a t i c p r o g r a m m i n g w h i c h i s o b t a i n e d by a q u a d r a t i c
approximation to the L a g r a n g i a n functional a n d l i n e a r approximations to the constraints. T h e r e f o r e ,
S Q P - m e t h o d s iteratively s o l v e a s e q u e n c e of q u a d r a t i c s u b p r o g r a m m i n g p r o b l e m s u n t i l a satisfactory
approximation to the solution is o b t a i n e d .
Let ( ~ , u ) be a n approximation to the solution ( x , u ) . The q u a d r a t i c s u b p r o g r a m m i n g o f
p r o b l e m ( 1 . 1 ) - - ( 1 . 4 ) at the point (~', u) is to d e t e r m i n e a correction d,, f o r the control f u n c t i o n u ,
a n d a correction d. for the state trajectory ~ to minimize
" ½ho.[O]d.(O) + jo[fO.[t]d.(t) + fo,,[t-]d,,(t)]dt -p go.[T]d.(T) + d.(O)rM~d.(O)
1 "T r ?MzEt] MaEt~ Fd.~(t)] l+ T , ) :]1 . d, + (T )L ' I I~d , (T ) (3. 1)
L M 3 E t ] M 4 F t ] J La'o(t)J y d ,
s u b j e c t to
w h e r e
D[O]+D.[O]d,(O) = O.
d x ( t ) = Y . E t ] d . ( t ) + y o E t ] d , , ( t ) - - ~ ( t ) ~ f[t],
SET]+~,[T]d,(T) = o,
O ~ t ~ T ,
M , = ho,,EOq + ~,, ~ ,D . , , 703 E n "~ ' ,3"1
M ~ E t q = f o , , E t ~ + ~-~ ;.,f.,,,Et~ E R °×"j I
j I
M , E t ] = fo,,,,Et] + ;'.;f.,,,oEt~ q n . . . . .j ~ l
( 3 . 2 )
( 3 . 3 )
( 3 . 4 )
( 3 . 5 )
( 3 . 6 )
( 3 . 7 )
( 3 . 8 )
N o . 2 O p t i m a l C o n t r o l Problems w i t h S t a t e C o n s t r a i n t s ] 67
Ms = go,,ET] + ~ .mE~,,ET] E R'~" ( 3 . 9 )j . 1
a r e s e c o n d derivative matrices.
Let (2 , ,d=) be a solution o f the problem ( 3 . 1 ) - - ( 3 . 4 ) , then there e x i s t a real n u m b e r } ~ 0 ,
a vector function ). E R~ , v e c t o r s 3 E R P a n d ~ E / F , not a l l z e r o s , s u c h that
~ . ( t ) r = - - ) . ( t ) r f , [ t ] - - p f 0 , [ t ] - - ~ , ( t ) r M 2 [ t ] - - Pd, ,( t)rM~[t], O ~ t ~ T , ( 3 . 10)
f~(O) r - = - - pho,[-O] - - ~ro,[-O] - - ~ , ( O ) r M , , (3 . 1 1 )
f~ (T) T = pgo,[T] + ~ r F ~ ( T ) + ~ l , ( T ) r ] l t 3 , (3 . 12)
f ~ ( T ) r f , [ t ] Jr- pf0~[t] + pd~(t)rM3[t] -+- ~9~l, ,( t)rM4[4] = O, O ~ t ~ T . ( 3 . 13)
It f o l l o w s from the f i r s t o r d e r n e s e s s a r y condition ( 3 . I O ) - - ( 3 . 13) a n d the constraints (3 . 2 ) -
( 3 . 4 ) that if problem ( 3 . 1 / - ( 3 . 41 h a s a solution for w h i c h the constrained qualifications are
satisfied then the solution of the p r o b l e m ( 3 . 1 ) - ( 3 . 4 ) c a n be o b t a i n e d a s the solution o f the
f o l l o w i n g set of e q u a t i o n s .L
,L,(t)|,~. j L-- M2Et] -- f , E t ] r L~( t ) L - - M~Et] r ]
r f r t ] - ~ ( t ) ]+ L - So,-J] ' J ' o < , ~< :~, (3. t,~1
_ r ~ , ( t ) ]
L ] + M,E~]~.(~/--/o.1-~], o ~ t < 7, (3.1,~)E,~IZj ] ' L L t ] ' ] ;.(t/
+ o : = - - , ( 3 . 16 )L M, /J L}.(0) J D,[01" Lh~,E01U
o o .
L Ms - - I L, ; . (T)J L£',['/']rJ Lg0,E:r]'J
This system of equations becomes a standard linear multipoint boundary value problem, when the
equations in ( 3 . |5 ) are used to eliminate d,(t l , while the expressions in (3. 16) and (3. ]7)
constitute b o u n d a r y equations.
For the solution of the l i n e a r multipoint b o u n d a r y v a l u e p r o b l e m , approximation m e t h o d can
be e m p l o y e d , that i s , the t ime functions are approximated by a finite-dimensional base ( g e n e r a l l y
p i e c e w i s e polynomials o n [ 0 , T ] ) , a n d a la rge a n d s p a r s e system of i i n e a r e q u a t i o n s is g e n e r a t e d .
w h i c h usua l ly has the f o l l o w i n g form
I i:1(" 0J
168 Applied Mathematics--A Journal of Chinese Universities Vol. 8 Ser. B
w h e r e M a n d C are s p a r s e a n d b a n d e d m a t r i c e s , a n d d a n d ~ a r e coefficients to be determined for the
t ime functions. The s y s t e m (3. 18) m a y be s o l v e d with a n y s p a r s e t e c h n i q u e s , for e x a m p l e r a n g e
s p a c e or null s p a c e m e t h o d s .
The derivation of the q u a d r a t i c s u b p r o g r a m m i n g p r o b l e m is b a s e d on the fact that the
approximation h o l d s o n l y in a neighbourhood o f the s o l u t i o n ( x , u , o', 5.,/~). H e n c e it i s a s s u m e d that
the c u r r e n t iterate i s c o l s e to ( x , u , a , k , / ~ ) . In o r d e r to f o r c e a g l o b a l c o n v e r g e n c e , the solution o f
the q u a d r a t i c p r o g r a m m i n g problem is u s e d a s a s e a r c h direction of a m e r i t f u n c t i o n , w h i c h a s s i g n s
a real v a l u e to each p o i n t ( x , u , o ' , Z , a ) , a n d w h i c h approaches its m i n i m u m a t the point ( z , u , ~ , ) . ,
/~). A s u i t a b l e choice for the m e r i t function i s the a u g m e n t e d L a g r a n g i a n f u n c t i o n , introduced by
P o w e l l a n d Hestence ( 1 9 7 6 ) on finite-dimensional nonlinear p r o g r a m m i n g .
M(x,u,cx,).,g,p) = L ( x , u , c + , ; , . , / ~ ) -F- -~EIID(~(o))ll ~ -+- , u ( t ) , t )
dz r dx; i ] [ f ( x ( t ) , u ( t ) , t ) - - ~ ] d t + i l g c z < V ) ) I I ~, <3. 19)
With sufficiently l a rge v a l u e o f p , the ( ~ r , ~ r ) o b t a i n e d in the solution of the p r o b l e m ( 3 . 1 ) - - ( 3 .
4 ) is a d e s c e n t direction o f the m e r i t function M ( x , u , o ' , ) , , g , p ) at the point ( x , u ) , then a step
length 3 ~ 0 can be t a k e n a l o n g the direction s u c h that
w h e r e
M ( h ) < M ( O ) ,
M ( a ) = M ( x --~ a d , , u -q- ctd,,o" + a(o" - - a ) , ) , q- a ( ~ - - A) , p q- a ( ~ - - g ) , p ) ( 3 . 2 0 )
is a function of s i n g l e v a r i a b l e a , a n d g , L a n d ~ a r e o b t a i n e d in the solution of p r o b l e m ( 3 . 1 ) - -
( 3 . 4 ) . Then next approximation to the solution a n d multiplers a r e c a r r i e d out by setting
xi+: = xi Jr- ad~,, ui+l = u, q- a d : , ,
cr~+~ = a, + a(a, - - c~,), ;'.~+~ = L ÷ a ( L - - L ) ,
It i s p r o v e d t h a t w h e n
/l~+, = u, + a(~+ - - /~i).
i s s a t i s f i e d , ( d , , d , ) i s a d e s c e n t direction o f the m e r i t f u n c t i o n , w h e r e
= Z(d:,d=)/ll(d:,,d:,)[l~, '/=- II(d=,d,)ll~/ll(~-- -,,~-- z , ~ - ~)11~,
f r r M2 d~M ( d , , d ~ ) = d . ( O ) r M , d g O ) + o(d, , d ,r ) dt + d . ( T ) r M s d = ( T ) .M r M4 [ d ~
H e n c e the penalty f a c t o r p c a n be a d j u s t e d a t e v e r y iteration to s a t i s f y the i n e q u a l i t y ( 3 . 2 1 ) .
( 3 . 2 1 )
N o . 2 O 0 t i m a l C o n t r o l P r o b l e m s w i t h S t a t e C o n s t r a i n t s 169
4. Discrete sequential quadratic programming
The state-constrained optimal control p r o b l e m c a n be converted into a finite-dimensional
nonlinear p r o g r a m m i n g by discretizing the t ime f u n c t i o n s , that i s , the interval [-0, T] i s d i v i d e d into
s u b i n t e r v a l s , the i n t e g r a l is r e p l a c e d by n u m e r i c a l i n t e g r a l , a n d the differential is s u b s t i t u t e d by
difference. The r e s u l t i n g finite-dimensional nonlinear p r o g r a m m i n g i s s o l v e d by g e n e r a l s e q u e n t i a l
q u a d r a t i c p r o g r a m m i n g m e t h o d s , a n d by m a k i n g use of the characteristic o f the problem.
Let the i n t e r v a l [ - 0 , T ] be d i v i d e d into _Vs u b i n t e r v a l s w i t h e q u a l d i s t a n c e h = T ~ N , a n d h~ =
i h , i : 0 , 1 , . . . , N . Then the discretized state-constrained o p t i m a l control p r o b l e m h a s the f o l l o w i n g
form
m i n i m i z e
s u b j e c t to
ho(xo) -~- S hf°(xi'ui'hi) -~- gO(Xv)' ( 4 . l )i~O
x,+l - - x i : h f ( x i , u . , h i ) , i = 0 , I , " ' , N - - 1 , ( 4 . 2 )
D ( x o ) = 0 , ( 4 . 3 )
E ( z ~ ) = 0 , ( 4 . 4 )
w h e r e x~ ---- z ( h , ) E R", u, = u ( h , ) E R m, a n d there a r e total N ( n + m ) + n decision v a r i a b l e s a n d
Nn q- p t q e q u a l i t y constraints.
Let ]',, i : 0 , 1 , . " , N , u , , i---- 0 , 1 , - " , N - - 1 be a solution o f p r o b l e m ( 4 . 1 ) - - ( 4 . 4 ) , then
t h e r e e x i s t vectors 5.0,~1,'",,~.x E R " , ~ E R' a n d / ~ E / ~ , not all z e r o , s u c h that
^ A ^)r ).r 3 H ' ( z , , u , , ) ~ _ ~ )^ - - : - - , i----- 0 , 1 , . . . , N - - 1 ,
Oxi
~T go,(~,) +/.TE,(_;-,.),A3 =
w h e r e
H ' ( x i , u , , ) . i + j ) = A ( x , , u , ) "q- f i (x , ,u~)rzi+J
is a H a l m i t o n i a n function with
f o ( x , , u , ) = & o ( x i , u ~ , h i ) ,
f ' ( x i , u D = hf(z~,u~,h~).
The L a g r a n g i a n function for p r o b l e m (4. 1 ) - - ( 4 . 5 ) i s
L ( x , u , a , ) . , # ) = h 0 ( x 0 ) + trrD(xo)
+ ~ Efo(~,,,,,) + z+,(:'(~,,,,,) - x,+, + ~,)3i = O
( 4 . 5 )
( 4 . 6 )
( 4 . 7 )
( 4 . 8 )
( 4 . 9 )
( 4 . 1 0 )
1"70 Applied Mathematies--A Journal of Chinese Universities Voh 8 Ser. B
-k- g 0 ( z , ) q- ~ r E ( z x ) . ( 4 . 1 1 )
The s e c o n d derivative m a t r i x of the L a g r a n g i a n function L with respect to z a n d u is a b l o c k d i a g o n a l
m a t r i x
V 2 L ( z , u , a , ) . , u ) ---- d i a g ( M [ 0 ] " , M [ I ] , . . . , M [ . V - - 1 ] , M s ) ( 4 . 12)
w i t h
LM~D- I T M,Ez]..I ,,,~
M:0j.[M, + ,E0 lL M ~ E 0 Y M, EOL! ,×,
Ml = ho,~(xo) ~ ~ ajD:.~(xo),1 - i
MZ,3 = H~£i3,
M3[i] = I l l , [ i - ] , i-.~ 0 , 1 , . . . , _ x - 1 ,
M, D3 = H,.,D],
Ms = go:,(z,,-) + 2 u ~ E ~ , ( z , ) .j = l
i = 1 , 2 , . - - , N - - 1 , ( 4 . 1 3 )
(r = n - ~ - m ) , ( 4 . 1 4 )
( 4 . 1 5 )
( 4 . 1 6 )
( 4 . 1 7 )
( 4 . 1 8 )
( 4 . 1 9 )
d = [dxo,duo ,dZl , d u , , . . . ,dz:~_, , d u ~ - ~ , d x ~ ] r ,
~") - - [-h0,(x0) + ~,(zo,uo),fS~(z~,uo),f~,(x~ , ~ ) , . . . ,
u ~ * " - ~ z . . . , . _ , ) , ~ . ( z , ) J ~fL-~(z~-,, . , - ~ , j o . ~ .,--~
c t*) ~- [ D ( x o ) , f f ( x o , u o ) - - xl "t- z o , f I ( x ~ , u : ) - - Xz + x i , " ' ,
y " - ~ ( z , - _ , , u , _ , ) - r , + z~_, , E ( z , ) ] ~
( 4 . 2 0 )
w h e r e
u*) c a n be expressed a s
m i n i m i z e
s u b j e c t to A " / 6 -~- c (~) = 0 ,
With these expressions the q u a d r a t i c s u b p r o g r a m m i n g of p r o b l e m ( 4 . 1 ) - ( 4 . 4 ) at the point ( x (~) ,
No. 2 Optimal Control Problems with State Constraints 171
A(')r :
- D , ( x o )
r ~ [ o [ - I
r , D ] r_ .D] - I
. . , , . .
r,E_v - t]
( 4 . 2 2 )
that is,
du, = - - [ M , Ei] ]-- 'E(A,,) T 4- f~TZ,--, 4- M3D]rdx,] ,
- - A ( ~ ) ' 6 = c <'). ( 4 . 29 )
i = 0 , 1 , . . . , X - - 1 . ( 4 . 3 0 )
r , [ _ \ ~ - 1 ] _ j lj"
E,(z,)
with I 1 [ i ] = I q - f ' , ( x , , u , ) , I ' 2 [ i ] = f ' o ( x , , u , ) , a n d 1t(k) is either ~ 2 L ( ' ) or its approximation.
W e s o l v e the p r o b l e m ( 4 . 2 0 ) to get a solution 6c,) a n d n e w estimates of L a g r a n g i a n multipliers
, # a n d ). , then 6 (~ is u s e d a s a s e a r c h direction to minimize a m e r i t f u n c t i o n w h i c h is o b t a i n e d by
a u g m e n t i n g the L a g r a n g i a n function w i t h a penalty t e r m ,
M ( x , u , q , ) . , ~ t , p ) = L ( x , u , q , Z , / . z )
3 - - 1
+ [l lD(xo) IIN + -- + z, + ( 4 . 2 3 )i . I)
Once a step l e n g t h , say ak , i s d e t e r m i n e d s u c h that M(ctk) < M ( O ) , a n e w approximation to the
solution a n d n e w estimates of L a g r a n g e multipliers a r e o b t a i n e d a s
x~ '~ ' ; = z~ ') + akdxp~, i = 0 , 1 , ' " , X - - 1 , ( 4 . 2 4 )
Z/~ ' q - l ) = H!' ) - ] - ( l kd l l~k) ,
a ~+') = a (') ÷ a ~ ( ? , - - a<~>), ( 4 . 2 5 )
Z~'-1) = ).~) 4 - a,(~., - - Z}~)), i = 1 , 2 , - " , N , ( 4 . 2 6 )
/ ~ ( ' - ' ) = /1Ck) 4 - a ~ ( ~ - - /1(~)). ( 4 . 2 7 )
It f o l l o w s from the f i r s t o r d e r n e c e s s a r y conditions o f p r o b l e m ( 4 . 2 0 ) that the solution of
problem ( 4 . 2 0 ) can be o b t a i n e d by s o l v i n g the f o l l o w i n g s y s t e m of e q u a t i o n s .-!= . ( 4 . 2 8 )
- A <')~ 0 i L c ") J
W h e n 1t-(~) t a k e s the s e c o n d derivative m a t r i x o f the L a g r a n g i a n f u n c t i o n , d,,, c a n be eliminated
from the s y s t e m
1 7 2 Appl ied Mathemat ics - -A J o u r n a l o f C h i n e s e Unive r s i t i e s V o l . 8 Ser. B
S u b s t i t u t i n g them into the f i r s t system of ( 4 . 2 8 ) , we g e t , a f t e r some m a n i p u l a t i o n , a s y s t e m o f
e q u a t i o n s
y,+~ - - y, = A,y,+~ + B,y, + f , , i = 0 , 1 , ' " , . V - - 1 , ( 4 . 3 1 )
Myo + Xy~ = b
w i t h
~-dx,
' b , [ i ]
, f , = i0
iLo
, b ~ I-- D(zc) 1- - E(z~ ) / '
- hL(xo)!
~oS (x,) j
b , [ i ] = f - - x,+, + x~ - - f : , (M4r - i ] ) - ' ( f~oo) T,
b~[ i ] ' '= - (fo,) + M ~ E i ] ( M , [ i ] ) -'(foo)" ' ,
( 4 . 3 2 )
( 4 . 3 3 )
( 4 . 3 4 )
A i
/11=
-0 - - M T [ - i ] 0
o z - M,D] o
0 0 0
0 0 0
-IL(zo) 0 0
0 0 o
ML I D,(zo) r
0 0 0
:lO;
O_ ~Xr
O
0
0
0 r'/:,~
, B i =
- M 6 [ i ] - - I
- - M, Ei]
0
L 0
0 0
E,(x, ) 0
0 0
- - 1113 I
0 0
0 0
0 0
0 0
0
0
0
0
O.
0
0
O. r ¢ , r
0
0
0
E , ( x x )7 r × r
M~D] = J + f ; - - £ , ( M , [ Q ) ' M , [ i y .
/117[i] = f '~(M4[i])- ' ( f i , ) r
( 4 . 3 5 )
( 4 . 3 6 )
M s E i ] = I + (J~L) r - - M 3 E i l ( / 1 1 4 E i ] ) - ' ( f . ) r ,
MgEi] = M ~ [ i ] - - M 3 r - i l ( M , r i ] ) - ' M 3 [ i ] r ,
( 4 . 3 7 )
( 4 . 3 8 )
The solution o f the s y s t e m ( 4 . 3 1 ) can be g e n e r a t e d by the f o l l o w i n g f o r m u l a ei--I
u, = ~o(i,O)vo + ~ o ( i , j + 1)3"~, i = 1 , 2 , . . . , _ v ,j = 9
( 4 . 3 9 )
w i t h
3. -- 1
Yo = [-M + N q g ( N , O ) J - ' ( b - - ~ X c p ( . \ , j + 1 ) f j ) ,]=6
(4.4O)
No, 2 Optimal C o n t r o l P r o b l e m s with S t a t e C o n s t r a i n t s 173
~ o ( i , j ) = I ] l-, = ( t r . . ~ . . . r , _ , )
I ' : = ( 1 - - A : ) - ' ( I J r - B j ) , j = O , I , ' " . . V - - 1
f , = ( I - - A j ) - l f ~
( 4 . 4 1 )
is factorized into the form
W h e n the s e c o n d derivative m a t r i x of the L a g r a n g i a n function L is not e v a l u a t e d , but just
approximated by a positive definite m a t r i x , a n d the approximation is u p d a t e d by u s i n g u p d a t i n g
f o r m u l a e for e x a m p l e BFGS f o r m u l a e . The positive definite approximation, s a y t l-'~k), is stored in
L D L r form
w h e r e
11"(') ~--- d i a g (II~~) , i,l-~k), . . . , , , tr v - 1e) ,--w¢*)).', = L D L r ' ( 4 . 4 4 )
L ---- d i a g ( L 0 , Z ~ , ' " , L , ) ( 4 . 4 5 )
is a b l o c k d i a g o n a l a n d l o w e r t r i a n g u l a r m a t r i x , a n d [,,, i : 0 , ] , . . . , N are unit l o w e r t r i a n g u l a r
m a t r i c e s , D is a d i a g o n a l m a t r i x with positive d i a g o n a l elements.
The s y s t e m ( 4 . 2 8 ) is s o l v e d by u s i n g L D L r factorization o f the coefficient m a t r i x , that i s , the
m a t r i x
_ _ A(k) r
This can be done by just f o r m i n g the m a t r i x
= - D - I L - 1 A (*) ( 4 . 4 7 )
a n d u s i n g Q R factorization m e t h o d to the m a t r i x B to get L D L r factorization o f the m a t r i x
A ( ~ : [ , - r ] ) - I L - 1 A (~). In the processes o f f o r m i n g B a n d Q R factorization of B , the sparsity of the
m a t r i x A (k) is preserved. The solution o f ( 4 . 2 8 ) is then o b t a i n e d by b a c k w a r d a n d f o r w a r d
substitutions.
A f t e r a n e w iterate i s o b t a i n e d , e v e r y b l o c k W~~) , i ---- 0 , 1 , " ' , N in the m a t r i x lJ=(~) i s u p d a t e d
( i f n e c e s s a r y ) by u s i n g BFGS f o r m u l a e
r / ? ) ~ ) T W?)6?)5¢~:W(? )i = 0 , 1 , . . - , N . ( 4 . 4 8 )W~+I) _-- B'[~) + 6(~)T~(,) 6(kTW(k)6(~) '
w i t h
( 4 . 4 6 )
( 4 . 4 2 )
( 4 . 4 3 )
174 Applied Mathematics--A Journal of Chinese Universities Vol. 8 Ser. B
[z~,+~ _ xP'1= . i = O , ] , . . . . X - - 1 ,
a!~~ = ~ , + ~ __ ~,!~,
/
' ' " ~ \ - 1(t~) T
and d <~r = (a<]' K . d ~ ~ ,d~ ). The p a r a m e t e r ~9 is determined in such a way that
T . T , •
d{~> q t) ~ 0 .2d ~' II ~',5C'}
so that the posi t ive defini te property of the m a t r i x 11-¢~} is guaranteed.
( 4 . 1 9 )
( , t . 50)
( 4 . 5 1 )
( 4 . 5 2 )
References
1] Bertsekas, D. P. , Constrained Optimization and Lagrange Multiplier Methods, Academic Press . New York,
1982.
2] Bry'sson, A. E. , Denham, W. F . . and Dreyfus. S. E. , Optimal programming problems with inequality
constraints I: Necessary conditions for extremal solutions. AIAA .lo~rnal , 1 (1963).
3] Dickmans, E. D . . and Wel l , K. H. , Approximate solution of optimal control problems using third order
Hermite polynomial functions, Proc. IF IP - -TC 7 Vl Techn. Conf. on Optimization Techniques, held at
Novosibirsk, 1974 . Proceedings Springer New York, 1975.
E4! Fletcher, R. , Practial Methods of Optimization Vol. 2. Constrained Optimization. John Wiley ~,. tons. New
York, 1981.
E5] Geerts , A. H. W. , Optimality conditions for solution of singular and s ta te constrained optimal control
problems in economics, M. Sc. Thesis. Eindhoven UniversiW of Technology, The Netherlands, 1985.
[ 6 ] Jacobson. D. H. , and Lele, M. M . . A transformation technique for optimal control problems with a s ta te
variable constraint, IEEE Trans. on Automatic Control. 14 : 5 (1969).
[_7] Jong de J. L. and Machielsen, K. C. P. , On the application of sequential quadrat ic programming to state-
constrained optimal control problems. Paper presented at the Fifth IFAC Workshop on Control Applications of
Nonlinear Programming and Optimization, Capr i , I ta ly , 1985 .
[8] Lorentz, J . , Numerical solution of the minimum-time flight of a glider through a thermal by use of multiple
shooting methods. Optimal Control Applications ~. Methods, 6 (1965).
Eg] Maurer, H. , On the minimum principle for optimal control problem with s ta te constraints, Schriftenreihe des
Rechenzentrums der Universitat Munster. FRG. Repor t , 41 (1979).
El0] Norr is , D. O. , Nonlinear programming applied to s ta te constrained optimization problems, Journal o f
Mathematical Analysis and Applications, 43 (1973).
[11] Machielsen, K. C. P. , Numerical solution of optimal control problems with s ta te constraints by sequential
quadratic programming in function space. CW1 Tract. 53 , 1988.