An Introduction To Optimal Control Applied To Disease Models

An Introduction to Optimal
Control Applied to Disease

Models
Suzanne Lenhart
University of Tennessee, Knoxville

Departments of Mathematics
Lecture1 – p.1/37
Example
Number of cancer cells at time

(exponential growth) State
Drug concentration Control

known initial data

minimize

where the first term represents number of cancer

cells and the second term represents harmful
effects of drug on body. Lecture1 – p.2/37
Optimal Control
Adjust controls in a system to achieve a goal

System:
Ordinary differential equations
Partial differential equations
Discrete equations
Stochastic differential equations
Integro-difference equations
Lecture1 – p.3/37
Deterministic Optimal Control
Control of Ordinary Differential Equations (DE)

control

state

State function satisfies DE

Control affects DE

Goal (objective functional)

Lecture1 – p.4/37
Basic Idea
System of ODEs modeling situation

Decide on format and bounds on the controls
Design an appropriate objective functional
Derive necessary conditions for the optimal control
Compute the optimal control numerically
Lecture1 – p.5/37
Design an appropriate objective functional
–balancing opposing factors in functional
–include (or not) terms at the final time
Lecture1 – p.6/37
Big Idea
In optimal control theory, after formulating a

problem appropriate to the scenario, there are
several basic problems :
(a) to prove the existence of an optimal control,
(b) to characterize the optimal control,
(c) to prove the uniqueness of the control,
(d) to compute the optimal control numerically,
(e) to investigate how the optimal control
depends on various parameters in the model.
Lecture1 – p.7/37
Deterministic Optimal Control- ODEs
Find piecewise continuous control and

associated state variable to maximize

!

$

#
#
"
subject to
%

'

&
#
and
(

)*
*
&
"
Lecture1 – p.8/37
Contd.
Optimal Control achieves the maximum
. -

/
,
+
Put into state DE and obtain
. -

. -

/
,
,
+
0
corresponding optimal state
. -

/
,
0
, optimal pair
. -

. -

/
,
,
+
Lecture1 – p.9/37
Necessary, Sufficient Conditions
Necessary Conditions
If , are optimal, then the following
4 3

4 3
5
2
2
1
conditions hold
..
.
Sufficient Conditions
If , and (adjoint) satisfy the conditions
4 3

4 3

7
2
2
1
..
.
then , are optimal.
4 3

4 3

5
2
2
1
Lecture1 – p.10/37
Adjoint
like Lagrange multipliers to attach DE to objective func-

tional.
Deterministic Optimal Control- ODEs
Find piecewise continuous control and
: 9

;
8
associated state variable to maximize
: 9
;
<
@
: 9

: 9

: 9

;
C;
:
?>
=
<
8
B
B
A
subject to
D
: 9

: 9

: 9
: 9

;
;
<
<
8
E
B
and
G9
;
<
<
<
HI
I
E
Quick Derivation of Necessary Condition
Suppose is an optimal control and
K
J
L
corresponding state. variation function,
ON
P
M
.
R
Q
u*(t)+ ah(t)
u*(t)
another control.
O N

ON
P
M
K
J
state corresponding to ,
O N
M
K
S
Q
T
O N
P
U
S
Q
T
O N

O N
P
O N

P
P
P
K
M
W
Q
V
T
OU
Contd.
At ,
_
Z
Z
X
a `

Y
Y
]
y(t,a)
x * (t)
x0
all trajectories start at same position

when control
\
_
b
b
Z
Z
X
X
[
c
Y
Y
]
]
e
\
_
g_
b
f
d
X
^
^
Y
]
a
Maximum of w.r.t. occurs at .

d
Z
^
Y
ih
t

w k h
k j
l
Adding
t xh
h m
v w sr
q p
n n o n o n
jy
xh
h
Contd.
x j t
jy
z m
l
u
x j
to our
{ | x j
i v
z m
k
k j
{ | x j
m
m k m
m xh
m
xh u
y
gives
}j
y
z m
l
j
v
z m }j
{
j k
{
v
k ~ m
~ m
y
y j
v
}j z m
l
j
z m
l {
v
k
}j
{
m
k
v

l

Contd.

|

l

|
|

here we used product rule and

¡

l

.

|

¢ ¢
¢ ¢ ¤ £
¥
¥
¤ £
° ®
¾ µ § ¦
´ ·°
Contd.
´
¥ © ª ¨
Arguments of
Arguments of
» »
¨ ¹ ¨ « ¬ «

© ¸
§ ¦ ¤
¸ µ ¸
º° « ¬ «
¤ ®
°
´
© ª ¨ ¯ ª ¨
®
« ¬ «

¤ «
¯ ¸
» ¼ º° ® ±l°
µ « ²
± °
© ¸ «
terms are
terms are
¾
» ½ ´ » ¤ ®
²
·° ·°
® « ¤
¬ ¤ ® ³
µ ´
Ã ¢ ¶ ¢ »
² ´ ·° ¤
» ·° ³
´ ®
¬ »
´
¤ ¤
± ´ µ
ÂÁ
À o
¿ ¿ o ¿ o ¿ ¶
¢
± ·°
²
· ° ´
² ·
´ § « ¬ «

´ ® ® ¤
.
¹
° ¤
¯ ª ¨ µ
³
º°
·°
´
® ´
´
.
µ «
«
¯ ¸ ¤ ¬
´ » ¼ º°
³
¢
¤
· ½ ´
ß
ËÊ
Í Û ÚÈ Ê È
Ë ÉÊ
Ì
á Í Ì Í Ì

Þ ª Ð
Ò Ë Ê
Ý Ü
Ê
Choose
Þ ª Ð Û Î
Ò Ñ ª ÏÐ
Ó Ö Ä
Contd.
Ô ÆÅ
Ò | ËÊ
Õ Ç
Ô Þ Ø È Ò
Ö Ì
Ó
Ô
Ìß
s.t.
Õ
ËÈ Ê ËÊ Ô
Optimality condition.
Ö×Ì
Þ Ø Ì
arbitrary variation Ë
Ìà
Ò ËÊ
ËÈÊ
Ò Ø Ì
Ñ l
Ó
Ô
Ò ËÊ
Õ
Ô
Ò
Í Ì
Ó
transversality condition
Õ
Û Ô
ÌÙ
for all
âÛ
ã
adjoint equation
Using Hamiltonian
Generate these Necessary conditions from
Hamiltonian
å ä

å ä
å ä

ê
é
ì é
í
ç
è
ë
æ
æ
integrand (adjoint) (RHS of DE)
maximize w.r.t. at
î
è
è
ï
optimality eq.
ì é
ð
ð
ë
ë
ñ
ñ
ï
adjoint eq.
ò
ê
é
ì é
í
ë
ë
ó
ô
ï
ä
ê
é
ð
ë
Converted problem of finding control to maximize
objective functional subject to DE, IC to using
Hamiltonian pointwise.
For maximization
ö
at as a function of
ü
ú
ù
÷
÷ õö
ø
For minimization
ö
at as a function of
û
ü
ú
ù
÷
÷ õö
ø
Two unknowns and
þ
ý
ÿ
introduce adjoint (like a Lagrange multiplier)
Three unknowns , and
þ
ý
ÿ
nonlinear w.r.t.
ý
Eliminate by setting
þ

ý

and solve for in terms of and
þ
þ
ý
ÿ
Two unknowns and
þ
ÿ
with 2 ODEs (2 point BVP)

+ 2 boundary conditions.
Pontryagin Maximum Principle
If and

are optimal for above problem, then there

exists adjoint variable s.t.

at each time, where Hamiltonian is defined by

and


Hamiltonian

$

&

%
"
"
#
!
!
maximizes w.r.t. , is linear w.r.t.

'
#
#

$
(
&
&

%
"
"
!
!
bounded controls, .

,
+
+
*
Bang-bang control or singular control #
Example:
0
&
&

%
#
. "
/
"
1
cannot solve for

&
-
3
%
#

1
is nonlinear w.r.t. , set and solve for

5
'
3
#
#

4
optimality equation. Lecture1 – p.23/37

Example 1
C B
D
? >
@
A
8 7
9
?
6
=
:
<
subject to
F
? >
? >
@
? >
@
JI>
K
E
E
G
G
H
What optimal control is expected?
m c
n [ c c Y
[ Z d d [\Z
d d [ b
S L
N M
O
` R R
[ b
S
[ b
^ c
Y
h
R
Q i Q P
e
d ]
m g j d e
k R R
Y
[ b ^ R
_ VUTS
`
]
`
l XW
^
[ c V
[
Y `T
integrand
Y h `
[ g
c
[ W
]
c
i cT c
Q i f
k
j a
a R
P
[ W [ g
`
f
c
Example 1 worked
h
[
Q e c
`
RHS of DE
[
at
R
g R
S
cT
R
W
Example 2
| s
D
w v
@
w v
x
q p
9
w
o
{
z
r
y
subject to
s
}
w v
@
w v
w v
@
Jv
y
u

~

z
What optimal control is expected?
Hamiltonian

The adjoint equation and transversality condition
give

@

@

Example 2 continued
and the optimality condition leads to

¤ £

¥

@

¡

¢

¦

The associated state is
¤ £

¥

@

¡
§
¢

¦
¨
Graphs, Example 2
Example 2.2
3
State 2
−1
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
Time
8
6
Adjoint
0
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
Time
0
−2
Control
−4
−6
−8
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
Time
® ®
» »
µ Á ´ µ ´ µ ´ ¬ ¼ µ \ ´
¼ µ º
® µ º ® ª ©
¬® «

¬®
®
¯ ¯
®
® ®
¶ ¼ ¶
¼ · ¬
_
¯
¯ ¬
º
Example 3
µ ¯ ¬
¹ ¬ ¹ ® °
¹ ¶ ¶ °
®
» ® ¯
»
¯
¯
¶ ¸«
¬ «
¯ ± ² U±°
® µ » ° µ
¹
¯ ¶
® ³
¹ ®
½ » ¹
¶
«
»
¶
at
· ¬ ¶
»± ¬
±
¾
¿
» ¬
À«
± µ ¾
µ
¹ ¶
»
Contd.
É
Æ
ÄÅÃ
È
Â
Â
Ç
Ç
Æ
Æ
ÄÅÃ
É
É
Ê Â
/
Ç
Ç
ÌË
ÐË
Í
É
Ï Ç
Ñ
Â
Ä
Î
Solve for and then get .
É
Ò
Ò
Â
Ó
Î
Do numerically with Matlab or by hand

Ì
Ì
Â
ÙÚ Ø
ÙÚ Ø
Ö Ô
Ö Ô
Ü
È
×
×
Ä
Û
É
Æ
Ê
Ê
ä
æ ó ó ò ê
ô ê ò
ã
ì æ
à ßÝÞ
ç ë ë
ë
ë ì
ë ð
í â á
Exercise
ä ã
å
æ
ï î
ä æ
èç
Jðã
é
ç
ä
ó
ë
ô
ë ï ì
æ
control
ë
ü ü ü
þ

õ
û û
ø ÷ö

ü

þ

þ þ ú ù

ü
ü û

þ
þ ý
ÿ

ü
ü
û

û

þ ý
û

û ý

ý
þ þ
û

þ
Exercise completed
þ

control
û

ø
ø

û

Contd.

!

"
#

There is not an “Optimal Control" in this case.
Want finite maximum.
Here unbounded optimal state
unbounded OC
Opening Example
&% Number of cancer cells at time
'
&
$
(exponential growth) State

Drug concentration Control
&%
'
(
)
$
&%
'
&%
'
+
$
(
*
,
&)
known initial data

.-%
'
$
0/$
*
1
minimize
2
%
'
&%
'
&)
$
(
/
See need for bounds on the control. See salvage
term.
Further topics to be covered
Interpretation of the adjoint

Salvage term
Numerical algorithms
Systems case
Linear in the control case
Discrete models
more info
See my homepage www.math.utk.edu lenhart
3
4
Optimal Control Theory in Application to Biology
short course lectures and lab notes
Book: Optimal Control applied to Biological Models
CRC Press, 2007, Lenhart and J. Workman

An Introduction To Optimal Control Applied To Disease Models

Transféré par

Informations du document

Description originale:

Titre original

Copyright

Formats disponibles

Partager ce document

Partager ou intégrer le document

Options de partage

Avez-vous trouvé ce document utile ?

Ce contenu est-il inapproprié ?

Droits d'auteur :

Formats disponibles

An Introduction To Optimal Control Applied To Disease Models

Transféré par

Droits d'auteur :

Formats disponibles

An Introduction to Optimal

Control Applied to Disease

University of Tennessee, Knoxville

known initial data

where the first term represents number of cancer

Adjust controls in a system to achieve a goal

Control of Ordinary Differential Equations (DE)

State function satisfies DE

Goal (objective functional)

System of ODEs modeling situation

In optimal control theory, after formulating a

Find piecewise continuous control and

Optimal Control achieves the maximum

like Lagrange multipliers to attach DE to objective func-

Find piecewise continuous control and

Suppose is an optimal control and

all trajectories start at same position

Maximum of w.r.t. occurs at .

with 2 ODEs (2 point BVP)

exists adjoint variable s.t.

Bang-bang control or singular control #

cannot solve for

is nonlinear w.r.t. , set and solve for

optimality equation. Lecture1 – p.23/37

and the optimality condition leads to

The associated state is

Do numerically with Matlab or by hand

(exponential growth) State

known initial data

Interpretation of the adjoint

See my homepage www.math.utk.edu lenhart

Vous aimerez peut-être aussi