Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
Preface
This lecture reviews the optimal control problem. The concepts of an admissible control, an admissible
trajectory, and the general form of the performance measure are presented. We also review the basic
forms of performance measures corresponding to certain classes (types) of optimal control problem.
During the course we will repeatedly use this material as we learn methods of solving the optimal control
problem.
Find an admissible control u which causes the system
x (t ) a( x(t ), u(t ), t )
to follow an admissible trajectory x that minimizes the performance measure
tf
u is called an optimal control and x an optimal trajectory.
An admissible control is a control history (i.e. control input values during the interval [ t 0 , t f ] ) which
satisfies the control constraints during the entire interval [ t 0 , t f ] . Similarly, an admissible trajectory is
a state trajectory which satisfies the state variable constraints during the entire interval [ t 0 , t f ] .
Let us examine a simple optimal control problem, i.e. a minimum time problem, involving an automobile.
The engine of the automobile limits its acceleration to some maximum value M 1 , while the automobiles
brakes limit the deceleration to some maximum value M 2 . Thus, the controls must satisfy the following
control input constraints:
0 u1 (t ) M 1
(2.1)
M 2 u 2 (t ) 0
1
ECE 551 LECTURE 1
Assume that the automobile starts from rest and accelerates in a straight-line path stopping a distance R
later, so that the state boundary conditions are as follows:
0
x(t 0 )
0
(2.2)
R
x(t f )
0
If the automobile never backs up, then we also have the following additional state constraints:
0 x1 (t ) R
(2.3)
0 x 2 (t )
tf
k u (t ) k u
t0
1 1 2 2 (t )dt G (2.4)
Where, G represents the quantity of fuel in the gas tank. Taken collectively, the constraints determine
admissibility. We can solve this problem using the techniques of this course to obtain the following
solution:
u1*( t )
M1
t0 (t0+tf) tf t
u2*( t )
(t0+tf) tf
t0
M2 t
2
ECE 551 LECTURE 1
x1*( t )
t0 (t0+tf) tf t
x2*( t )
(t0+tf) tf t
t0
Performance Measures
The performance measure defines optimality for a particular situation or problem. The most general form
of the performance measure is given by the following expression:
tf
The first term on the right of equation (2.5) represents the cost of reaching the final state. It is often
referred to as the terminal cost. The integral term in (2.5) represents costs associated with the control
input and the resulting state trajectory. The integrand is referred to as the Lagrangian after Lagrange who
studied optimization problems involving integral cost functions. Performance measures that only involve
terminal costs are referred to as Mayer cost functions after Christian Mayer. Performance measures that
only involve an integral cost function are referred to as Lagrange cost functions after Lagrange. The
general form which includes both terminal and integral cost functions is referred to as Bolza cost functions
after Oscar Bolza. Let us now consider some specific examples of performance measure types.
3
ECE 551 LECTURE 1
Minimum-time Problem
Transfer a system from an arbitrary initial state x(t 0 ) x 0 to a specified target set S in minimum time.
Hence, we have that
tf
J dt t f t 0 (2.6)
t0
Minimize the deviation of the final state of a system from its desired value r (t f ) . As one possible
measure, we have that
N 2
J xi (t f ) ri (t f ) (2.7)
i 1
J x(t f ) r (t f ) x(t
T
f ) r (t f )
2
(2.8)
x(t f ) r (t f )
Where, a is the norm of the vector a . For greater generality, we can insert a real symmetric positive
semi-definite n X n weighting matrix H , i.e. we have the following quadratic form
J x(t f ) r (t f ) H x(t
T
f ) r (t f )
2 (2.9)
x(t f ) r (t f )
H
4
ECE 551 LECTURE 1
Note that if H I (identity matrix), then equations (2.8) and (2.9) are equal. Also, if H is a diagonal
matrix, then each of its diagonal elements must be non-negative, since H is positive semi-definite. By
adjusting or choosing the values of the elements of H , we can quantify the importance of deviations of
each final state of the system xi (t f ) from its desired value ri (t f ) . Thus, relatively larger values of select
elements of H assign greater significance to those deviations, while if an element is set equal to zero,
then no significance is attributed to that deviation.
Transfer a system from an arbitrary initial state x(t 0 ) x 0 to a specified target set S with minimum
expenditure of control effort. One possible measure is as follows:
tf
J
t0
u (t ) dt (2.10)
Equation (2.9) could correspond to a spacecraft whose rocket engines thrust is proportional to fuel
consumption. If there are multiple controls, then we have that
tf
M
J i u i (t ) dt (2.11)
t0 i 1
Where, i is a non-negative weighting factor. Another possible form of the performance measure might
be as follows:
tf
J u 2 (t ) dt (2.12)
t0
In this case, the situation might be a source driving a passive resistor network. If there are multiple
controls, then we have that
5
ECE 551 LECTURE 1
tf
J u
T
(t ) R u(t ) dt
t0
(2.13)
tf
2
u(t ) R
dt
t0
Where, R is a real symmetric positive definite weighting matrix. In general, the elements of R can be
functions of time.
Tracking Problem
Maintain the system state x(t ) as close as possible to the desired state r (t ) in the interval [ t 0 , t f ] . Thus,
we have that
tf
J x(t ) r (t )
2
Q (t )
dt (2.14)
t0
Where, Q (t ) is a real symmetric n X n matrix that is positive semi-definite for all t t 0 , t f . Note that
equation (2.14) assumes that the controls (admissible set) are bounded. If that is not the case, then we
have that
dt
tf
J x(t ) r (t ) u(t )
2 2
Q (t ) R (t )
(2.15)
t0
Where, R (t ) is a real symmetric m X m matrix that is positive definite for all t t 0 , t f . If it is also
important that the states be close to their desired values at t t f , then we have
dt
tf
2
J x(t f ) r (t f ) x(t ) r (t ) u(t )
2 2
Q (t ) R (t )
(2.16)
H
t0
6
ECE 551 LECTURE 1
Regulator Problem
This is a special case of the tracking problem where the desired state r (t ) is equal to zero for all
t t0 , t f .
Let us make the following state variable and control input assignments:
x1 (t ) h Aircraft altitude
x 2 (t ) h Altitude rate
0 1 0 0 0
0 a a 23 0
x (t ) 22 x(t ) 0 u (t ) (2.17)
0 0 0 1 0
0 a 42 a 43 a 44 b4
7
ECE 551 LECTURE 1
hd ( t )
450
= 0.064 rad = 3.650
0 30 Time (sec)
d/dt ( hd ( t ) )
30 Time (sec)
0
7050
0 Range (ft)
15
d ( t ) )
0.0873 radian = 50
30 Time (sec)
0
7050
0 Range (ft)
d/dt ( d ( t ) )
30 Time (sec)
0
7050
0 Range (ft)
8
ECE 551 LECTURE 1
2
2
J k h h(30) hd (30) k h h(30) hd (30) k (30) d (30)
2
30
q h ( ) h( ) hd ( ) q h ( ) h( ) hd ( ) d
2 2
(2.18)
30
q ( ) ( ) d ( ) q ( ) ( ) d ( ) d
2 2
30
r e ( ) e ( ) ed ( ) d
2
Where, is a dummy variable of integration. Recalling the desired state trajectory, we have that
hd (30) 0 Feet
hd (t ) 15 Feet/sec
hd (30) 15 Feet/sec
d (t ) 0.0873 Radians
d (t ) 0 Radians
d (30) 0 Radians
9
ECE 551 LECTURE 1
2
2
J k h h(30) k h h(30) 15 k (30) 0.0873
2
30
q h ( ) h( ) 450 15 q h ( ) h( ) 15 d
2 2
(2.19)
30
q ( ) ( ) 0.0873 q ( ) ( ) d
2 2
30
r e ( ) e ( ) d
2
x( ) r( )
30 (2.20)
Q( ) x( ) r ( ) r e ( ) u 2 ( ) d
T
Where,
k h 0 0 0
0 k h 0 0
H
0 0 k 0
0 0 0 0
And,
q h ( ) 0 0 0
0 q h ( ) 0 0
Q( ) (2.21)
0 0 q ( ) 0
0 0 0 q ( )
10