Scribd
Upload
90 views

Optimal Control (Course Code: 191561620)

1. The only equilibrium point of the nonlinear system is (0,0,0). A Lyapunov function candidate of V(x) = x12 + x22 + x32 shows the equilibrium is stable but perhaps not asymptotically stable. 2. A Lyapunov function of V(x) = xTPx where P satisfies the Riccati equation shows the equilibrium (0,0) is asymptotically stable. 3. The optimal solution to the minimization problem is given by x(t) = (8-7t)2/3, satisfying the boundary conditions and making the second order condition satisfied. 4. The optimal control is -1

Uploaded by

Abdesselem Boulkroune
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
90 views

Optimal Control (Course Code: 191561620)

1. The only equilibrium point of the nonlinear system is (0,0,0). A Lyapunov function candidate of V(x) = x12 + x22 + x32 shows the equilibrium is stable but perhaps not asymptotically stable. 2. A Lyapunov function of V(x) = xTPx where P satisfies the Riccati equation shows the equilibrium (0,0) is asymptotically stable. 3. The optimal solution to the minimization problem is given by x(t) = (8-7t)2/3, satisfying the boundary conditions and making the second order condition satisfied. 4. The optimal control is -1

Uploaded by

Abdesselem Boulkroune
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 4

Optimal Control

(course code: 191561620)


Date: 05-04-2015
Place: CR-3H
Time: 08:45-11:45

1. Consider the nonlinear system

ẋ 1 −x 1 + x 2
   
ẋ 2  = x 1 − x 2 − x 1 x 3  . (1)
ẋ 3 −x 3 + x 1 x 2
(a) Determine all points of equilibrium.
(b) Consider equilibrium x̄ = (0, 0, 0). What does candidate Lyapunov func-
tion V (x) = x 12 + x 22 + x 32 allow us to conclude about the stability prop-
erties of this equilibrium?

2. Consider

ẋ 1 = −x 1 + x 2
ẋ 2 = −x 1 − 2x 2

with equilibrium x̄ = (0, 0). Determine a Lyapunov function V (x) such that
V̇ (x) = −2x 12 − x 22 and verify that this V (x) is a strong Lyapunov function for
this system.

3. Consider minimizing
Z 1
x(t )ẋ(t )2 dt ,
0

over all functions x(t ) subject to x(0) = 4, x(1) = 1.

(a) Argue that if the optimal solution x(t ) is nonnegative for all t ∈ [0, 1]
and x(0) = 4, x(1) = 1 that then ẋ(t ) ≤ 0 for all t ∈ [0, 1].
(b) Which function x(t ) ≥ 0, ẋ(t ) ≤ 0 solves the Beltrami identity and sat-
isfies the boundary conditions x(0) = 4, x(1) = 1?
x γ+1 (t )
[Hint: you may want to use that x γ (t )ẋ(t ) = a iff γ+1 = at + b when-
ever x(t ) > 0 and γ 6= −1.]
(c) Is Legendre’s second order condition satisfied?

4. Consider

ẋ(t ) = −x(t ) + u(t ), x(0) = e, x(2) = 1.

This is a system with both initial and final constraint. We want to minimize
Z 2
|u(t )| dt
0

with u(t ) ∈ [−1, 1] for all t ∈ [0, 2]. (Notice the absolute value in the cost
function.)

1
(a) Determine the Hamiltonian H
(b) Express the optimal u(t ) in terms of the costate p(t ) and argue that at
any moment in time we have either u(t ) = −1 or u(t ) = 0 or u(t ) = +1
(c) Determine the costate equations and its general solution p(t )
(d) Show that if u(0) 6= 0 then u(t ) is constant over t ∈ (0, 2].
(e) Determine the optimal input for the given initial and final constraint
x(0) = e, x(2) = 1

5. Consider the optimal control problem

ẋ(t ) = u(t ), x(0) = x 0

with u(t ) ∈ R and cost


Z 1 Z ∞
2
J [0,∞) (x 0 , u(·)) = u (t ) dt + 4x 2 (t ) + u 2 (t ) dt .
0 1

(a) Assume first that x(1) is given.


R ∞ Determine the optimal cost-to-go from
t = 1 on: V (x(1), 1) := minu 1 4x 2 (t ) + u 2 (t ) dt .
R1
(b) Express the optimal cost J [0,∞) (x 0 , u(·)) as J [0,∞) (x 0 , u(·)) = 0 u 2 (t ) dt +
Sx 2 (1). (That is: what is S?)
(c) Solve the optimal control problem: determine the optimal cost J [0,∞) (x 0 , u(·))
and express the optimal input u(t ) as a function of x(t ). [Hint: see the
hint of problem 3.(b)].

problem: 1 2 3 4 5

points: 2+3 4 2+4+2 1+3+2+2+2 3+2+4


Exam grade is 1 + 9p/p max .

Euler-Lagrange:
∂ d ∂
µ ¶
− F (t , x(t ), ẋ(t )) = 0
∂x dt ∂ẋ
Beltrami:
∂F
F −( )ẋ = C
∂ẋ
Standard Hamiltonian equations for initial conditioned state:
∂H (x, p, u),
ẋ = x(0) = x 0 ,
∂p
∂H (x, p, u), ∂S(x(T ))
ṗ = − , p(T ) =
∂x ∂x
LQ Riccati differential equation:

Ṗ (t ) = −P (t )A − A T P (t ) + P (t )B R −1 B T P (t ) −Q, P (T ) = S

Hamilton-Jacobi-Bellman:
∂V (x, t ) ∂V (x, t )
· ¸
+ min f (x, u) + L(x, u) = 0, V (x, T ) = S(x)
∂t u∈U ∂x T

2
1. (a) only (0, 0, 0)
(b) For “ease of exposition” denote (x 1 , x 2 , x 3 ) as (x, y, z). The V is contin-
uously differentiable and positive definite. V̇ (x) = 2x(−x + y) + 2y(x −
y −xz)+2z(−z +x y) = 2[−xx +x y +x y −y y −x y z +−zz +x y z] = −2(xx −
2x y + y y + zz) = −2(x − y)2 − 2z 2 so it is ≤ 0 but not < 0 (for z = 0, x =
y 6= 0). So stable but perhaps not asymptotically stable.
(You don’t have to invoke LaSalle but if you do then you’ll see that it
is in fact asymptotically stable.)
1 51
2. the linear equation P A + A 0 P = −2 0
0 −1 gives P = 6 1 2 . Since p 11 = 5/6 > 0
£ ¤ £ ¤

and det(P ) = 1/4 > 0 this matrix is positive definite, so V := x 0 P x > 0, V̇ < 0
and thus V is a strong Lyapunov function.

3. (a) Suppose x(t ) has a positive derivative, then connecting the local max-
ima/minima such as here in red

x(0) x(0)

x(1) x(1)

makes ẋ = 0 on these regions so makes x ẋ 2 smaller. The optimal x


R

hence has no local maxima/minima.


(b) Beltrami says C = x ẋ 2 − (x2ẋ)ẋ = −x ẋ 2 . So x 1/2 ẋ = a for some con-
stant a. By the hint this means x 3/2 (t ) = (3/2)(at +b). Initial condition
then becomes 43/2 = (3/2)b so 8 = (3/2)b so b = 16/3. Final condition:
13/2 = 3/2(a + b) so a = 2/3 − b = −14/3. That is x(t ) = (8 − 7t )2/3 .
∂2 F (t ,x(t ),ẋ(t ))
(c) Yes: ∂ẋ 2
= 2x(t ) ≥ 0 for all t ∈ [0, 1].

4. (a) H = p(−x + u) + |u|


(b) If p > 1 then p(−x + u) + |u| is minimal for u = −1. If p < −1 then
p(−x +u)+|u| is minimal for u = +1. If −1 < p < 1 then p(−x +u)+|u|
is minimal for u = 0:

−1 if p(t ) > 1


u(t ) = 0 if |p(t )| < 1

+1 if p(t ) < −1

(c) ṗ = p without final condition (because there is a final condition on x).


General solution is p(t ) = c et
(d) If u(0) 6= 0 then u(t ) = ±1 so ∓p(0) ≥ 1. Since p(t ) = p(0) et and et
increases we have that then ∓p(t ) > 1 for all t > 0, so u(t ) = ±1 for all
t >0
(e) A bit tricky to explain:

3
part (d) says that u(0) = ±1 implies u(t ) = ±1 for all t ∈ [0, 1]. These
are not feasible:
• If u(t ) = 1 for all time then ẋ = −x + u, x(0) = e gives x(t ) = 1 +
(x(0) − 1) e−t > 1 for all time so not x(2) = 1.
• If u(t ) = −1 for all time then ẋ = −x + u, x(0) = e gives x(t ) = −1 +
(x(0) + 1) e−t for all time so not x(2) = 1.
Hence u(t ) must be zero initially, so |p(0)| < 1. As time increases the
value |p(t )| might become 1 at some time t 0 . For t > t 0 the value of
u(t ) must then be +1 or −1 for the rest of time. Since we need to end
up at x(2) = 1 this mean that if t 0 < 2:

x u=1 (t ) = 1 + (x(2) − 1) e2−t , x u=−1 (t ) = −1 + (x(2) + 1) e2−t


| {z }
1

for all t ∈ [t 0 , 2]. See the plot:

x(0) = e

xu=0 (t )
x u =−
1(
t)

xu=1 (t )
x(2) = 1

e−1

Clearly the only possible solution is the red one: for t ∈ [0, 1] we have
u(t ) = 0 and for t ∈ [1, 2] we have u(t ) = +1.

5. (a) The Algebraic Riccati becomes 0 = P 2 − 4. So P = 2: V (x(1), 1) = 2x(1)2


(b) The principle of optimality says that Sx 2 (1) = V (x(1), 1) = P x 2 (1). So
S =P =2
(c) The Riccati differential equation becomes

Ṗ = P 2 , P (1) = 2.

This is of the form P γ Ṗ = a for γ = −2 and a = 1. so the hint of the hint


says that P −1 (t )/(−1) = t +b so P (t ) = 1/(−b − t ). Given that P (1) = 2 it
follows that b = −3/2, so
1
P (t ) = , t ∈ [0, 1].
3/2 − t

The optimal cost hence is x 2 (0)P (0) = 23 x 02 and u(t ) = −P (t )x(t ) for
t ∈ [0, 1] and u(t ) = −2x(t ) for t > 1.

You might also like