Upload
others
View
4
Download
0
Embed Size (px)
Citation preview
1
6-Aug-09 Vira ChankongEECS, CWRU
1
Numerical Optimization
Instructor: Vira ChankongElectrical Engineering and Computer ScienceCase Western Reserve University
Phone: 216 368 4054, Fax: 216 368 3123E-mail: [email protected]
A WorkshopAt
Department of Mathematics Chiang Mai University
August 4-15, 2009
6-Aug-09 Vira ChankongEECS, CWRU
2Vira ChankongEECS. CWRU
2
Session:Optimization of Dynamic Systems
Vira ChankongVira ChankongCase Western Reserve UniversityCase Western Reserve University
Electrical Engineering and Computer ScienceElectrical Engineering and Computer Science
2
6-Aug-09 Vira ChankongEECS, CWRU
3
Module Objective Module Objective
To develop functional knowledge and skills in
• modeling dynamic systems and/or optimal control problems
• determining a state trajectory and/or a control of dynamic systems that optimizes a performance functional subject to constraints on control and/or states.
6-Aug-09 Vira ChankongEECS, CWRU
4
Module HighlightsModule HighlightsTwo pronged approach:
1. Focus on the two traditional approaches for dealing with dynamicoptimization/optimal control problems
• the variational approach based on calculus of variations leading to the maximum principle
• the dynamic programming approach and its corresponding Hamilton-Jacobi-Bellman (HJB) equation. When applied to linear optimal control problem derivation of similar results based on the concept of Lyapunov stability will also be demonstrated.
2. Focus on numerical methods for solving large real-world dynamic optimization and optimal control problems with complex constraints. The two numerical approaches are the indirect approach and the direct approach. MATLAB will be used to implement methods discussed.
Applications to engineering and economic problems will be illustrated throughout
3
6-Aug-09 Vira ChankongEECS, CWRU
5
ReferencesReferencesText:
Optimal Control Theory, D.E. Kirk, Dover Publications, ISBN: 0486434842, 2004
References:
1. Practical Methods for Optimal Control Using Nonlinear Programming, J.T. Betts, SIAM, ISBN: 0898714885, 2001
2. Applied Dynamic Programming for Optimization of Dynamical Systems, R.D. Robinett III, D.G. Wilson, G. Richard Eisler and J. E. Hurtado, SIAM, ISBN: 089715865, 2006
6-Aug-09 Vira ChankongEECS, CWRU
6
Dynamic OptimizationDynamic Optimization00
( )
: [ , ]
( ) ( )0 0 0 0 0 0 0
( ) ( )
min ( , ,.., , )
subject to , ( ) , ( ) ,.., ( )
and possibly , ( ) , ( ) ,.., ( )
f
f
t n
tx t t R
n n
n nf f f f f f f
g x x x t dt
t x t x x t x x t x
t x t x x t x x t x
J→
=
= = =
= = =
∫
Note:1) The final time may or may not be specified.
If is specified problem
(or
fixed-end-time
fixed-terminal-time)If is not specified variable- en
f
f
f
t
t
t
⇒
⇒ problem
d-time
open-terminal-time, open ( horor izon)2) If is specified, and if
( ) is also fixed problem
( ) is constraine
fixed-end-point
constrained-terminald (i.e. ( ) ) problem
( ) is not fixed or has no restric
-point
frtion
f
f
f f
f
t
x t
x t x t S
x t
⇒
∈ ⇒
⇒ ee-end-poin prot blem
4
6-Aug-09 Vira ChankongEECS, CWRU
7
Optimal Control Problems
00: [ , ]
0
0 0
System Dynamics
min ( ( ), )+ ( ( ), ( ), )
subject to ( ) ( ( ), ( ), ); where state : [ , ]
( )
:
Initial conditions:Terminal or final
f
mf
t
f f tt t R
nf
h t t g t t t dt
t t t t
J
t t R
t
→=
= →
=
∫ux x u
x a x u x
x x
0 0
Find a control ( ) to take the system from the initial state( ) at the initial time to the final
(
state ( ) at the final tim
cond
e
i )tions:
.
f
f f
f
tt t
t
t t
=
x x
x x
u
6-Aug-09 Vira ChankongEECS, CWRU
8
Optimal Control Problems
0
min max 0
0
Possible additional constraints:
( ) for all [ , ]
e.g. ( ) for all [ , ]
( ) for all [ , ]
Contraints on control:
Contraints o
e.g.
n state:
t f
f
t f
t U t t t
t t t t
t X t t t
∈ ∈
≤ ≤ ∈
∈ ∈
u
u u u
x
min max 0 ( ) for all [ , ]ft t t t≤ ≤ ∈x x x
5
6-Aug-09 Vira ChankongEECS, CWRU
9
Optimal Control ProblemsNote:1) As before, may be specified (
or may not specififixed-end-time)
variable-end-time or open term
ei l
d
(na
ft
time)
2) If is specified,
( ) may be fixed ( or
fixed-end-point or hard terminal constraint)softconstrained ( ) (
or not fixed or nterminal con
o restrictiostraint)
fren ( e-end-point)
f
f
f
t
x tx t S∈
6-Aug-09 Vira ChankongEECS, CWRU
10
Optimal Control Problems
g = 0,
Special Mayer Problems: (
: i.e. ( ( ), ) ------------------
( )
Various types of Mayer Problem
(Li
nea) ar
(
r
)
)
b
Maye
f f
f
JJ h t t
J c t
• =
=
x
x
( ) ( )( )- ( ) ( )- ( ) (Terminal control proble m)T
f f f fJ t t t t= x r H x r
0
00
= 0,
Special Lagrange Problems:
= 1, i.e.
Lagrange Problem
i.e. ( ( ), ( ), ) --------
1 -
= ( ) , i.e.
(a) (Mini
mum time)
(b ) (
f
f
t
t
t
ft
J g t t t dt
J dt t t
h
g
t tJg
• =
= =
=
∫
∫u u
x u
0
0
2
)
=
(Minimum fuel)
(c) (Mi ( ) , i.e. nimum energy( ) ( )
)
f
f
t
t
t
t
Tg t t
t
dtt
d
J =
∫
∫u u u
6
6-Aug-09 Vira ChankongEECS, CWRU
11
Optimal Control Problems
( ) ( )( ) ( )
0
Tr
ac
(
k
(
iSpecial Bolza Problems:
(a)
), )+ ( ( ), ( ), ) --------
( )- ( ) ( )- ( )
=
ng probl
( )- ( )
B
( )- ( )
olza Proble
=
m:
ms
e
ft
f f t
T
f f f f
T
J h t t g t t t dt
t t t t
t t t
h
g t
• = ∫x x u
x r H x r
x r Q x r
( ) ( ) ( ) ( )0
( ) ( )
( )- ( ) ( )- ( ) ( )- ( ) ( )- ( ) ( ) ( )
Note: , and are w eighting matri es
c
f
T
tT T Tf f f f t
t t
J t t t t t t t t t t dt
+
= + +⇒ ∫
u Ru
x r H x r x r Q
Q
x r u Ru
H R
0
(b) =
=
Regulator problem ( ) ( )
( ) ( ) ( ) ( )
( ) ( ) ( ) ( ) ( ) ( )
:
f
Tf f
T T
tT T Tf f t
t t
t t t
h
t
J t t t t t t dt
g +
⇒ = + +∫
x Hx
x Qx u Ru
x Hx x Qx u Ru
6-Aug-09 Vira ChankongEECS, CWRU
12
Example 1: (Brachistochrome) A bead of mass 1 unit descends along a wire joining two fixed points (x0,y0) and (xf,yf). We wish to find the shape of the wire so that the bead completes its slide in minimum time.
(x0,y0)
(xf,yf)
t = t0
7
6-Aug-09 Vira ChankongEECS, CWRU
13
Example 1: (Brachistochrome) Model:
(x0,y0)
(xf,yf)
t = t0
y = y(x), y: [x0,xf]→R, y ∈ C1 ( )0 0
0 0
1 ( )min 2 ( )
subject to ( )
Minimum timeFixed terminal point problem(Hard terminal constr
Dynamic Optimization Mod
a
e
i
)
l:
n )
(
t
fx
f x
f f
y xt dxg y y x
y x yy x y
′+=
−
=
=
∫
6-Aug-09 Vira ChankongEECS, CWRU
14
Example 2: (River-Crossing 1) A boat travels with constant velocity Vwith respect to the water. In the region the velocity of the current is parallel to x-axis but varies with y. Given the destination (xf,yf) on the other side of the river, find the path to be taken by the boat to minimize the travel time,
(x0,y0)=
(0,0)
(xf,yf)
t = t0
t = tf
vc(y)
x
y
θ
V0
0 0 0 0
0min 1
subject to(Equation of M otion):
cos ( ) sin
Optimal Control M odel:
Dynamics
Initial conditions: Final c
( ) ; ( ) ( )
M inim
onditions: ; )
(
um
ft
f t
c
f f f f
t dt
x V v yy V
x t x y t yx t x y t y
t
θθ
=
= +
+
== =
= =
∫
timeFixed terminal point problem
8
6-Aug-09 Vira ChankongEECS, CWRU
15
Example 3: (River-Crossing 2) A boat travels with constant velocity Vwith respect to the water. In the region the velocity of the current is parallel to x-axis but varies with y. Given the final time tf, find the path to be taken by the boat to maximize the landing distance on the other side of the river,
(x0,y0)=
(0,0)
(xf,yf)
t = t0
t = tf
vc(y)
x
y
θ
V
0 0 0 0
max = ( )
subject to(Equation of M otion):
cos ( )
Optimal Control M odel:
Dynamics
Initial conditions: Final con
di
sin ( ) ; ( )
Linear M ayer
ti
problem
(: )
w
ons
f
c
f f
J x t
x V v yy V
x t x y t yy t y
θθ
= +=
= ==
ith semi-hard terminal constraint
6-Aug-09 Vira ChankongEECS, CWRU
16
Example 4: (Braking and Acceleration): Want to move a car of mass m from 0 to xf in minimum time.
x(0) = 0
0
0
1 2
2
1 2
1
min = 1
subject to(Equation of Motion):
( ) ( ) ( )
Optimal Control:
Dynamics
Initial conditio (ns: Terminal co
0)nditions:
0; (0 0( )
)
ft
f
f f
J t dt
x x tx t t
x xx t x
α β
=
== +
= ==
∫
2
1 1 2 2
1 2
; ( ) 0
For each [0, ] :
On control: 0 ( ) ; 0 ( )
Contraints:
On state:
0 ( ) ;
0 ( )
f
f
f
x t
t t
u t M u t Mx t x x t
=
∈
≤ ≤ ≤ ≤≤ ≤ ≤
tf
x(tf) = xfx(t)
α (t) +β (t) t
1
2 1
( ) ( ) ( ) ( ) ( )
( ) ( )
EOM:Stat
(e :
)s
x t t tx t x tx t x t x t
α β= +== =
1 1
2 2
: ( ) ( )
0 1 0 0,
0 0
Compact Dynamic
1 1
( ) ( ) ( )( ) , ( )
( )
s
) ( ) (
t t
x t u t tt t
x t u t tαβ
= +
⎛ ⎞ ⎛ ⎞= =⎜ ⎟ ⎜ ⎟⎝ ⎠ ⎝ ⎠⎛ ⎞ ⎛ ⎞ ⎛ ⎞
= = =⎜ ⎟ ⎜ ⎟ ⎜ ⎟⎝ ⎠⎝ ⎠ ⎝ ⎠
x Ax Bu
A B
x u
Minimum timeHard terminal constraint with bound constraints
9
6-Aug-09 Vira ChankongEECS, CWRU
17
Example 5: (Taking-off) An aircraft of mass m (assumed point mass) is to be lifted by a constant trust T to reach the cruising altitude H at time tf, at maximum speed (along x-direction).
H
t0= 0
t = tf
v
x
y
β(t)
T2
1 2
2
3 4
4
1
max = ( )
(Equation of Motion):
cos
Opt
imal Control Mod
el:
Dynamics
sin
(
Initial conditions:
fJ x t
x xTxm
x xTx gm
x
β
β
=
=
=
= −
2 3 4
3 4
3
0) (0) (0) (0) 0 ( ) ; ( ) 0
On control 0 ( ) /2, [0, ]
On state: ( ) 0, 1, .
Fin
.4; (
al conditions:
Constraints:
)
f f
f
i
x x xx t H x t
u t t t
x t i x t H
π
= = = == =
≤ ≤ ∈
≥ = ≤1 2 1
3 4 3
( ) cos ( ) ( ) sin ( )
( ) ( ); ( ) ( ) ( )
EOM:
States:
Contr ( ) ( ); ( ) ( ) ( )
)ol ( ) (
mx t T tmy t T t mgx t x t x t x t x t
x t y t x t x t y tu t t
ββ
β
== −= = == = =
= Linear Mayer problem with semi-hard terminal constraint
6-Aug-09 Vira ChankongEECS, CWRU
18
Example 6: (Rover Control) Want to control the speed of a Mariner Mars rover at about 5 mph using as little energy as possible. The controller is the output voltage of a battery-operated voltage regulating system.
1 2
1 2
( ) ( ) ( )
( ) ( )
( ) ( ) ( ) ( )
( ) (
Dynamics:
States:
Controls:
); ( ) ( )
( ) ( ); ( ) ( )
ff f f
t f
L
f
L
di tR i t L e t
dtt K i t
t I t B t t
x t i t x t t
u t e t u t
torque
t
λ
λ θ θ λ
θ
λ
+ =
= − − −
= + +
= =
= =
Voltage Regulating System
Rf
Battery
Lf
if
e(t)Ra
La
Ia (constant)
λ(t)
θ(t)
λL(t)
Viscous friction coefficient B
10
6-Aug-09 Vira ChankongEECS, CWRU
19
Example 6: (Rover Control)( )2
2 1 10
1 1 1
2 1 2 2
1 2
Min ( ( ) 5) ( ) ( )
:
1 ( ) ( )
1
Dynamics
Initial condi
( ) - ( ) ( )
(0) (0) 0
tions: Final con None On
ditions: Constrai t :
n s
ft
f
f f
t
J k x t wx t u t dt
Rx x t u t
L L
K Bx x t x t u tI I I
x x
= − +
= +
= −
= =
∫
1 max
2 max
1 max
2 max
control: ( ) , [0, ]
( ) , [0, ]
On state: ( ) , [0, ]
( ) , [0, ]
f
f
f
f
u t e t t
u t t t
x t I t t
x t t t
λ
≤ ∈
≤ ∈
≤ ∈
≤Ω ∈Minimum energy problem with free terminal conditions but with bound constraints
Voltage Regulating System
Rf
Battery
Lf
if
e(t)
Ra
La
Ia (constant)
λ(t)
θ(t)
λL(t)
Viscous friction coefficient B
6-Aug-09 Vira ChankongEECS, CWRU
20
Example 7: (Mixture) Want to find inflow rate of fresh water (control) to the two-tank system so that the salt concentration in tanks A and B are equal using minimum amount of fresh water.
u(t)
50gal tank A
u(t) u(t)
50gal tank B
80lb salt at
t=0
Fresh water at t=0
1
2
( ) lbs of salt in A at time
Assume: 1) Well-mixed at all time
( ) lbs of salt in B at
2) I
time
ncompressible fluid States:
Co ( ) gal/m of fresh watentrols: r
x t tx t t
u t
===
passing thru at time t
0
1 1
2 1 2
1 2
1 2
min = ( )
:( ) ( )
50( ) ( ) ( ) ( )
50 5
Optimal Control Model:
Dynamics
Initial conditions: Terminal conditions:
Contrai
0(0) 80; (0) 0
( )- ( ) = 0
For cnts: ea
ft
f f
J u t dt
u tx x t
u t u tx x t x t
x xx t x t
= −
= −
= =
∫
max
1 2
h :On control: 0 ( ) , [0, ]
On state: 0 ( ) 80; 0 ( )
fu t u t t
x t x t
≤ ≤ ∈
≤ ≤ ≤
Minimum control effort withSoft terminal constraint and bound constraints
11
6-Aug-09 Vira ChankongEECS, CWRU
21
Optimizing Yeast or Ethanol Production in a Bioreactor
Bioreactors are large vessels that serve as an environment for biochemical reactions to occur. Typical uses include the growth of microorganisms and the breakdown of products.
The environment within the vessel is controlled to optimize performance. Typical control variables include nutrient feed rate, oxygen air flow rate, and temperature. There is large economic incentive to develop control strategies to maximize the production of baker’s yeast and ethanol, two important commercial products produced in bioreactors. Yeast is typically grown off a solution containing glucose and other nutrients essential for cellular growth. When glucose concentration in the medium is high or when there is a limited supply of oxygen, the yeast microorganism excretes ethanol.
Source: D. Moore, MS Thesis, EECS, CWRU, 2007
6-Aug-09 Vira ChankongEECS, CWRU
22
Optimizing Yeast or Ethanol Production in a Bioreactor
X = yeast concentration (g/l)S = substrate (glucose) concentration (g/l)E = ethanol concentration (g/l)O = dissolved oxygen (O2) concentration (g/l)C = dissolved carbon dioxide (CO2) concentration (g/l)V = liquid volume (l)Fin = Substrate feed rate (l/h)Sin = Influent Substrate Concentration (g/l)D = Fin/V = Dilution rate (1/h)OTR = kLa(OS – O) = O2 transfer rate (g/L h-1)CER = kVkLaC = CO2 evolution rate (g/L h-1)m = Maintenance term (g of S /g of X h-1)
Model of Growth Dynamics
7-Aug-09
We wish to maximize production of yeast, or ethanol or both by controlling the substrate feed rate, airflow (O2) and temperature
12
6-Aug-09 Vira ChankongEECS, CWRU
23
Two Ways to Solve Dynamic
Optimization/Optimal Control Problems
1. Indirect Method:
• Solve the necessary conditions for optimality derived through variationalprinciples rooted in Calculus of Variations
• That is: Solve two-point boundary value problems (TPBVP)
6-Aug-09 Vira ChankongEECS, CWRU
24
Two Ways to Solve DynamicOptimization/Optimal Control Problems
2. Direct Method:• Optimize the functional directly as
constrained optimization• Require conversion to nonlinear programs
through transcription of the ODEs(dynamics of system)
• Often possesses high sparsity and special structure
13
6-Aug-09 Vira ChankongEECS, CWRU
25
Direct Method for Dynamic Optimization
• Convert to nonlinear programs through direct transcription of the ODEs (dynamics of system) or the corresponding DAEs
• Use nonlinear optimizer such SQP to solve the resulting nonlinear programs (Software such as SNOPT by Boeing, etc.
• Fine-tune the result through mesh-refinement techniques
6-Aug-09 Vira ChankongEECS, CWRU
26
Direct Method: Transcription methodsDirect Method: Transcription methods
( )
1
1
1 2 3 4
1
2 1 1
( , , )
1 wh
Euler:
ere
2 26
( , , )1
Classical Runge-Katta:
( , ,2
k k k k k k
k k k k
k
k k k k
kk k k k
h th
h th
h t
+
+
+
= += +
= + + +
=
= + +
x x a x ux x s
s k k k k
k a x u
k a x k u
( )
3 2 1
4 3 1 1
1 1 1
1
Trapezoidal: Hermit-Simp
)2
1 ( , , )2 2
( , , )( , , ) ( ( , , ), , )
( , ,
son:
) 6
kk k k k
k k k k
k k k k k k k k k k k k k
kk k k k k
hh t
h th t h t t
ht
+
+ +
+ + +
+
= + +
= +
= + + +
= + +
k a x k u
k a x k ux x a x u a x a x u u
x x a x u a( )
( ) ( )
1 1 1
1 1 1
1 1 1
( ( , , ), , )
1 where ( , , ) ( , , ) ( ( , , ), , )2 8
and ( , , )2
k k k k k k k k
kk k k k k k k k k k k k k k k k k
kk k k k
h t t
hh t t h t t
ht
+ + +
+ + +
+ + +
+ +
= + + + − +
= +
x a x u u a
x x x a x u a x u a x a x u u
a a x u
14
6-Aug-09 Vira ChankongEECS, CWRU
27
Direct Method: Transcription methods
Band, Stair-case Structure and Sparsity of resulting matrix:
Employ special numerical tricks to take advantage of the special structure and sparsity of the resulting problem
6-Aug-09 Vira ChankongEECS, CWRU
28
Indirect Method: Derivation of Optimality Conditions
•• EulerEuler--Lagrange Equations and all Boundary Lagrange Equations and all Boundary ConditionsConditions
•• HamiltonHamilton--JacobiJacobi--Bellman ConditionsBellman Conditions•• PontryaginPontryagin’’ss Minimum PrincipleMinimum Principle
15
6-Aug-09 Vira ChankongEECS, CWRU
29
Indirect Method: Fundamentals
1 2 1 22
var
2
( ( )) : where [ , ] and :[ , ]
( *, ) ( *, ) ( )
where ( ) 0 as
Path optimization: Functional
0Op
Variations of a Functional:
n
increment iation error term
J x x x x x x R
J Jδ δ δ ο δ
ο δ δ
∈ →
Δ = +
→ →
y y
y y y y y
y y
0 0
2
var
Function ( ( )) or ( ( ), ) : where [ , ] and :[ , ]
( *, ) ( *, ) ( )
timal control:
( *, , ) ( *, ,
nf f f f f
increment iation error term
f f
increment
J t J t t t t t t t R
J J
J t J t
δ δ δ ο δ
δ δ δ δ δ
∈ →
Δ = +
Δ =
x x x
x x x x x
x x x x22
var
) ( , )f
iation error term
tο δ δ+ x
6-Aug-09 Vira ChankongEECS, CWRU
30
0
( )
( )(
1 21 2
)
1) Key variation formular: ( , )
( ( )) ( , ,.., , )
Then ( , ) ..
For example:
..
f
T
n
t nf t
nn
n
JJ
J J Jx
J t g x x x t dt
J J JJ x x
x x
x
x x x
xx x
δ δ δ
δ δ δ δ δ
δ δ δ
∂=∂
∂ ∂ ∂= + + +∂ ∂
=
∂ ∂ ∂= + + +∂ ∂ ∂
∂
∫x
x x
x x xx
0
0
00
( )( )
2) Term like are dealt with thru integrati
on by par
.
t
i.e.
.
f
ff
f
t
t
tt
tt
t nnt
g g g
g x dtx
g g gx dt
x x x dtx x
xx x t x
x
δ
δ δ δ
δ δ
∂ ∂ ∂⎛ ⎞= + + +⎜ ⎟∂ ∂ ∂⎝∂⎛ ⎞
⎜ ⎟∂⎝ ⎠
⎛∂ ∂ ∂ ∂⎛ ⎞ ⎛ ⎞ ⎛ ⎞= −⎜ ⎟ ⎜ ⎟ ⎜ ⎟⎜∂ ∂ ∂ ∂⎝ ⎠ ⎝ ⎠ ⎝⎝
⎠
⎠
∫
∫
∫0
ft
txdtδ⎞⎟
⎠∫
Indirect Method: Fundamentals
16
6-Aug-09 Vira ChankongEECS, CWRU
31
* is an extremal of ( ) only if ( *, ) 0 for all admissible3) Fundamental Theorem of Calculus of Variations:
4) Fundamental Lemmas of Calculus of Variation
a) Let ( ) [ , ].
s
.
:
J J
x C a bα
δ δ δ=
∈
x x x x x
a
1
b 1
a
b
b)
If ( ) ( )
Let (
0 for
) [ , ].
If ( ) ( ) 0 for all ( ) [ , ] with ( ) ( ) 0
then ( ) fo
all ( ) [ , ] with ( ) ( ) 0
then ( ) 0 for all ]
[ ,x C a b
x h x d
x h x dx h x C a b h a h b
x
x h x C a b h a h b
b
x c
x aα
α
α
α
α
= ∈ = =
=
∈
′ = ∈ = =
=
∈
∫
∫
( )
1
b 1
a
c) Let ( ) and (x) [ , ].
If ( ) ( ) ( ) ( ) 0 for all ( ) [ , ] with ( ) ( ) 0
then ( ) ( )
r all [
for all
, ]
[ , ]
x C a b
x h x x h x dx h x C a b h a h b
x x
x b
a b
a
x
α β
α β
β α
∈
′+ = ∈ = =
= ∈
∈
′∫
Indirect Method: Fundamentals
6-Aug-09 Vira ChankongEECS, CWRU
32
00( ) ( , , ) , :[ , ]
(fixed end point)
* is an extre
Now consider
a) Case: and ( ) are fixed
0 ( *,
mal of ( ) only if ( *, ) 0 for all admissibl
)
e .
ft
ft
f f
J x g x x t d
t x t
J JJ x x x
J x J x x
t R
x
x t t
xδ
δ
δ
δ
δ
δ∂ ∂
⇒
=
∂
→
= =∂
=
+
∫
x
0
00
for
=
all admissible
for all admissib
(integration by part)
le
ff
f
tt
tt
t
t
g
x xx
g gx x dt xx x
g x gx x dtx tx x
δ δ
δ δ δ
δδ δ⎛ ⎞∂ ∂ ∂⎛ ⎞ ⎛ ⎞+⎜ ⎟ ⎜ ⎟⎜ ⎟∂ ∂ ∂⎝
∂ ∂⎛ ⎞= +⎜ ⎟∂ ∂⎝ ⎠
∂⎝∂⎠ ⎠⎝ ⎠
−∫
∫
0
0
for all admissible
(since ( ) 0 and ( ) 0)
Fundamental Lemm
=
as of Calcu
0 (b s y lu o
ft
t
f
g x xx
x t x t
g
g dtt x
gt xx
δ δ
δ δ
⎛ ⎞∂ ∂⎛ ⎞⎜ ⎟⎜ ⎟∂ ∂⎝ ⎠⎝ ⎠
∂ ∂⎛
∂−
∂
= =
⇒∂ ⎞ =⎜ ⎟−
∂ ⎝∂ ∂ ⎠
∫
f Variations 4a)
This is Euler-Lagrange Equation
Indirect Method: Necessary Conditions
17
6-Aug-09 Vira ChankongEECS, CWRU
33
00For
with and ( ) fixed (fixed en
* is an ext
( ) ( , , ) , :
remal
d point)
0
[ , ]
Euler-L
of ( ) only if
agrange Equati
(secon
on
d-
Necessary Conditions:
f
f f
t
ft
t x t
gJ x
J x g x x t
gx t
t t t
x
d x R
∂ ∂ ∂⎛ ⎞− =⎜ ⎟∂ ∂ ∂⎝
= →
⎠
∫
x
0 0
order ODE)
with 2-boundary point ( ) and ( )
Solved by shooting method (for example)
f f
x t xx t x
=
=
Indirect Method: Necessary Conditions
6-Aug-09 Vira ChankongEECS, CWRU
34
( )
( )
( , )
2 2 2
0
1 2
0
with (0) 0, ( 2) 1 (fixed end point)
( ) ( ) ( ) , : [ , ]
Euler-Lagrange Equatio
Exam
2 2 2 2 0
( ) cos sinWi
ple:
(
n
th 0)
g x x
f
x x
g g x x x xx t x t
x t c t c tx
J x x t x t dt x t t Rπ
π= =
∂ ∂ ∂ ∂⎛ ⎞− = − − = + =⎜ ⎟∂ ∂ ∂
= − →
∂⎝ ⎠= +⇒
=
∫
0 and ( 2) 1 ( ) 0cos 1sin = sin
Solved numerically by the shooting method
xx t t t t
π⇒
== +
Indirect Method: Necessary Conditions
18
6-Aug-09 Vira ChankongEECS, CWRU
35
00( ) ( , , ) , :[ , ]
(free end point)
* is an extremal o
For
b) Case: is fixed and ( ) is free
f
0 ( *, )
( ) only if ( *, , ) 0 for all admissible ( . , )
ft
ft
f f
f f
J
t x t
JJ x x
J x g x x t dt x t t
x
x J x x
R
x xxδ δ δδ
δ δ
δ=
⇒ =∂
= →
∂=
∫
x
0
00
for all admissible
for all admissible
= (integration by part)
f
f
ft
t
tt
t
t
Jx x xx
g gx x dtx t x
g gx x dt xx x
g xx
δ δ δ
δ δ
δ δ
δ
δ ⎛ ⎞∂ ∂ ∂⎛ ⎞ ⎛ ⎞+⎜ ⎟ ⎜ ⎟⎜ ⎟∂ ∂ ∂⎝ ⎠
∂+∂
∂ ∂⎛ ⎞=
⎝ ⎠
+⎜ ⎟∂ ∂
⎠−
∂⎝
⎝ ⎠
∂
∫
∫
0
0
*,
for all admissible , and ( )
(since
0 (
( ) 0 and ( )
=
E
0)
( ) f
f
t
f tx
f
f
t
g gx t dtx t
g g
g x x x tx
x t x
x x
t
t
xδ δ δ
δ δ
δ ⎛ ⎞∂
∂ ∂ ∂⎛ ⎞− =⎜ ⎟∂
∂ ∂⎛ ⎞+ ⎜ ⎟⎜ ⎟∂ ∂ ∂⎝ ⎠
∂ ∂⎝⇒
∂
⎠−
⎝
≠
⎠
∂
=
∫
*,
0
uler-Lagrange)
0
( ) 0fx t
gx
x t
∂=
∂
=
Indirect Method: Necessary Conditions
6-Aug-09 Vira ChankongEECS, CWRU
36
00Now consider
and is free (variable end time, free terminal time or horizon)
a) Case: (
( , ) ( , , ) , : [ , ]
(free termina) is free and is indepe l condindent of
0 (
tions)
f
f
f
t
f ft
f
J x t g x x t dt
t
x t t
x t t R
J xδ⇒ =
= →∫
0
*,
*, , , ) for all admissible ( , , )
for all admissible , ( , , ) ( *, *, )
=
)
an
(
df f f
f
f
f
t t t
t t
T
fx t
f f f
f fg x x
x x t x x t
x x t
g
t dt g x x t dt
g gxx
xx
tt x
δδ
δ δ δ δ δ δ
δ δ δ
δ δ
++
⎛ ⎞∂ ∂ ∂⎛ ⎞+ ⎜ ⎟⎜ ⎟∂ ∂ ∂⎝ ⎠⎝
=
∂−
∂ ⎠
∫ ∫
( )
0
0*,
*,
(since ( ) - * ( ) )
( *, *, )
= - * ( ) ( *, *, )
=
f
f
f
f
f f
t
f
f
ft
T t
f
f
f f f ftx t
T
fx t
dt g x x t t
g gx x t t dt g x x t tx t x
x t
g xx
gxx
x t
g x
x tδ δ
δ
δ
δ
δ
δδ δ
+
⎛ ⎞∂ ∂ ∂⎛ ⎞+ +⎜ ⎟⎜ ⎟∂ ∂ ∂⎝ ⎠⎝
=
⎠
∂
∂−
∂
∂∂
+∂
∫
∫
0*,
*,
for all admissibl
0 (Euler-Lagr
(
e , an
*, *, ) * ( )
( *, *, )
ange)
0,
d
f
f
f
Tt
f f ftx t
T
f
f
x
f
x t
x
x
g gdt g x x
g gx t
t x t tt x x
gg xgx
x t
x
x
t
x
δ
δ δ
δ
δ
⎛ ⎞⎛ ⎞∂ ∂ ∂⎛ ⎞ ⎜ ⎟
∂ ∂ ∂⎛ ⎞− =⎜ ⎟∂ ∂ ∂⎝ ⎠
∂=
+ −⎜ ⎟⎜ ⎟ ⎜ ⎟∂ ∂ ∂⎝ ⎠⎝ ⎠ ⎝ ⎠
∂−
∂∂
−
⇒
∫
*,0* ( ) = 0, and ( ) 0
f
ft
x tx t =
Indirect Method: Necessary Conditions
19
6-Aug-09 Vira ChankongEECS, CWRU
37
00Now consider
and is free (variable end time, free terminal time or horizo
( , ) ( , , ) ,
n)
b) Cas
0 (Eu
e:
: [ , ]
(Hard terminal co( ) =
l
ntraint)
er-Lagra
f
f
f
f
f
t
ft
g
J x t g x x t dt x t t R
t
x t x
gx t x∂ ∂ ∂⎛ ⎞− =⎜ ⎟∂ ∂ ∂
→
⎝ ⎠
=
⇒
∫
*,
*
0
,
( *, *, ) * ( ) = 0, and
(Soft terminal contraint
( ) =
c) Case: ( ) =
nge)
, ( ) 0
0 (Euler-Lagrange)
) -- ( )
( ) (
( )
*,
f
f
T
f fx t
f f f
T
fx
f f
f f
t
gg x x t x tx
x t t
g t g
x t
g gx t
x t x
x t t
xx
x
θ δ θ δ
θ
∂−
∂
=
=
∂ ∂ ∂⎛ ⎞− =⎜ ⎟∂ ∂⇒
∂
∂+
⎠
∂
⎝
*
0
,
*, ) * ( ) = 0,
and ( ) = ( )( ) 0,f
f
f fx t
f
T
x t x t t
gx t x tx
θ=
∂−
∂
Indirect Method: Necessary Conditions
6-Aug-09 Vira ChankongEECS, CWRU
38
00Now consider
and is free (variable end time, free terminal time or horizon)
c) Case: ( ) = ( )
0 ( *, , , )
( , ) ( , , ) , : [ , ]
(Soft terminal contraint
for a
)
ll a
f
f
f f
t
t
f f
f fJ x t g x
t
x
x
t t
J x x x t
t dt x t t R
θ
δ δ δ δ⇒ =
= →∫
0
0*,
( *, *, ) ( *, *, )
dmissible
for all admissible
= ( ) ( *, *, )
=
f f f
f
f
f
t t t
t t
T t
f f ftx t
T
g x x t dt g x x t dt
g gx t dt g x x t tx t x
g
x
x
g xx
x
δ
δ
δδ
δδ δ
++
⎛ ⎞∂ ∂ ∂⎛ ⎞+ +⎜ ⎟⎜ ⎟∂ ∂ ∂⎝ ⎠⎝
∂
=
∂−
∂ ⎠
∂
∫ ∫
∫
( )0
0
*,
*,
(since ( ) - * ( ) )
- * ( ) ( *, *, )
= ( *, *, )
f
f
f
f
t
f f f f ftx t
T t
f ftx t
f f f fx t
g xx
gx x t t dt g x x t tt x
g gx d
x x t t
t g x x tg xxx t x
δ δ δ
δ
δ
δ δ δ
δ
∂−
∂
∂
⎛ ⎞∂ ∂⎛ ⎞+ +⎜ ⎟⎜ ⎟∂ ∂⎝ ⎠⎝ ⎠
⎛ ⎞∂ ∂ ∂ ∂⎛ ⎞+ + −⎜ ⎟⎜ ⎟∂ ∂ ∂⎝ ⎠⎝ ⎠−
∂
=
∫
∫*,
*,0
*,
for all admissible , an
0 (Euler-Lagrange)
0
* ( )
( *, *, ) * ( ) =, ( 0, and
0
d
)
f
ff
T
f fx t
T
fx t
fx t
f f
g gx t x
g x
g x t tx
gg
x x
x
t
xt x t
xtx
δ δ
δ
δ
∂ ∂ ∂⎛ ⎞− =⎜ ⎟∂ ∂
⎛ ⎞⎜ ⎟⎜ ⎟∂⎝ ⎠
∂−
∂
∂⎝ ⎠
∂
⇒
∂= =
Example Derivation of Case (c)
20
6-Aug-09 Vira ChankongEECS, CWRU
39
Weistrass Erdman Corner Co1) Piecewise Continuous Solution: Require additional or 2) Con
Further
straint
nditions
s onTranve
( ) rrsatility
equire the
extensions
use of mulConditions
tipliers M
:
tx
ost important cases are in optimal control problems:
Indirect Method: Necessary Conditions
6-Aug-09 Vira ChankongEECS, CWRU
40
0
0 0
( , , ) ( ( ), ) ( , , )
( , , )Boundary condi
Optimal Control Problems:
Optimize
I) is fixed and (
tions: ( )
) is fixeCas :
ds
II
e
ft
f f t
f f
fJ t h t t g t dt
t
t t
t
= +
==
∫x u x x u
s.t. x a x ux x
x
) is fixed and ( ) is free
III) is fixed, and m( ( )) 0
IV) is free and ( ) is free
V) is free and ( ) is fixed
VI) is free and ( ) = ( )
VII) is free,
f f
f f
f f
f f
f f f
f
t t
t t
t t
t t
t t t
t
=
x
x
x
x
x θ
and m( ( )) 0
VIII) is free, and m( ( ), ) 0f
f f f
t
t t t
=
=
x
x
Indirect Method: Necessary Conditions
21
6-Aug-09 Vira ChankongEECS, CWRU
41
Indirect Method: Necessary Conditions
0
0
0 0
0
( , , ) ( ( ), ) ( , , )
( , , )Boundary conditions: ( )
( ( ), ) ( ( ),Case IV:
Optimal Co
)
ntrol Problems:
Optimize
( ( ),
f
f
t
f f f t
t
f f t
J t h t t g t dt
tt
h t t h t t dt h t
= +
==
= +
∫
∫
x u x x u
s.t. x a x ux x
x x x
( )
0
0
a
( , , , )
) ignored (constant)
( ( ), ) ( ( ), ) =
( ( g ( , , , , ) ( , , ) ( , , )
fT
t
t
T
H t Hamiltonian
t
h t t h t t dtt
ht g t t−−
− −
⎛ ⎞∂ ∂⎛ ⎞ +⎜ ⎟⎜ ⎟⎜ ⎟∂ ∂⎝ ⎠⎝ ⎠
∂= + + − +
∫
x u p
x xxx
xx x u p x u p a x u x
f
0
0a
t
( *, , *, , *, , , )
* *
), ) ( ( ), )
( , , , , ) g ( , , , , )
=
( *, ) + ( ) * ( *, *,
f
a f f
t
Tf T
f f
T
t
a f t
f
J t
H H dt
h t
t t h t tt
J t t d
g t
t
t
δ δ δ δ δ δ
δ δ
δ
⎛ ⎞∂ ∂⎛ ⎞+ +⎜ ⎟⎜ ⎟∂ ∂⎝ ⎠⎝ ⎠
∂⎛ ⎞− +
∂⎛ ⎞⎜ ⎟
+⎜ ⎟∂⎝
∂ ∂⎝ ⎠
⇒ =
⇒
⎠
∫
∫x x u u p p x
u p xu x
xp x p a x u
x
xxx
x x u p x x u p
( * ( ),)
= 0 for all admissible , , ,
f ff
f f
h t tt
t
δ
δ δ δ δ
∂⎛ ⎞+⎜ ⎟∂⎝ ⎠
xx
x u x
6-Aug-09 Vira ChankongEECS, CWRU
42
free and ( ) free
Hamilto
Necessary Conditions fo
nian-Jacobi Conditions:
r Optimal Control Problems:Case IV: Free-end point problems:
*, *)is extremal, then there exists co-states *such that:
(
f ft t
⇒
x
x u p* = 0 (1)
* (2)
( , , )
H
t
H
∂∂
∂=
∂=
−
u
px
x a x u
0 0
( *, ) ( *( ),
OTHER CASES CAN BE SIMILARILY
(3)
+ (4)
and ( )DERIVED
( ) * = 0
f f ff f f
h t h t t
t
t H tδ δ∂ ∂⎛ ⎞ ⎛ ⎞
− +⎜ ⎟ ⎜ ⎟∂ ∂⎝ ⎝=
⎠ ⎠
x xp x
xx x
x
Indirect Method: Necessary Conditions