Optimal Control Theory SF 2852

Size: px

Start display at page:

Download "Optimal Control Theory SF 2852"

Myles Jennings
5 years ago
Views:

1 Optimal Control Theory SF 2852 Johan Karlsson Optimization and Systems Theory, Department of Mathematics Royal Institute of Technology (KTH) Spring 2017

2 Optimal Control Theory SF Course information and course logistics 2 Course overview and application examples

3 Course Information and Course Logistics This is a course on optimal control Optimization problems involving difference or differential equations Many engineering problems are naturally posed as optimal control problems Economics and logistics Aerospace systems Automotive industry Autonomous systems and robotics Bio-engineering Process control Power systems

4 Course Contents Optimal control is an important branch of mathematics Roots in the calculus of variations Dynamic programming Pontryagin minimum principle (PMP) Linear quadratic control Model predictive control (MPC)

5 Teachers Johan Karlsson ( Silun Zheng, ( (Yuecheng Yang)

6 Course Material Optimal Control: Lecture notes Exercise notes on optimal control theory Some complementary material will be handed out. It will also be posted on the course homepage.

7 Four optional homework sets Homework HW0 Some basics on linear systems and get started on convex optimization solvers. (No bonus points). Due April 3, at HW1 Discrete dynamic programming and model predictive control. Due on April 20, at HW2 Computational methods (Project): Final version (as pdf) May 9, at Discussion/Feedback May 11, at 8.15 (mandatory to get bonus points). HW3 PMP and related topics. Due on May 16, at Each homework has 3-5 problems. Grading: Each homework may give maximum 2 bonus credits for the exam.

8 Final Written Exam The final exam takes place on Wednesday May 31, 2017 at You must register for the exam during 18 Apr - 15 May. Use "My Pages". Grade A B C D E FX Total credit (points) Total credit = exam score + homework score. The maximum exam score = 50. Maximum bonus from the homework sets = 6. You may use Mathematics Handbook and a formula sheet found on the course homepage.

9 Course Overview and Application Examples 1 Discrete time optimal control theory 2 Continuous time optimal control theory

10 Discrete Time Optimal Control N 1 min φ(x N ) + f 0 (k, x k, u k ) k=0 subj. to x k+1 = f (k, x k, u k ) x 0 given, x k X k u k U(k, x k ) The variables x 1,, x N (states) and u 0,, u N 1 (controls) are the ones we optimize over. The functions φ and f 0 determine the terminal and running costs. The function f and sets X k and U determine feasible states and controls. Sequential decision problem which contain as special cases Graph search problems Combinatorial optimization Discrete time optimal control

11 Shortest Path problem Initial node Terminal node Stage 0 Stage 1 Stage 2 Stage 3 Stage 4 Stage 5 Find the shortest path from the initial node at stage 0 to the terminal node at stage 5

12 The Knapsack Problem n max p j x j j=1 subject to n w j x j c, x j = 0 or 1, j = 1,..., n. j=1

13 Discrete Optimal Control Theory Discrete optimal control problem Dynamic programming Optimization theory Global theory Feedback solutions Local theory Open loop control Model predictive control Feedback solution Online optimization

14 Continuous Time Optimal Control t1 min Φ(t 1, x(t 1 ))+ f 0 (t, x(t), u(t))dt subj. to t 0 ẋ(t) = f (t, x(t), u(t)) x(t 0 ) S 0, x(t 1 ) S 1 u U(x) The variables are the functions x(t) (state trajectory) and u(t) (control signal) that we optimize over. Sometimes we let also t 1 be a variable. The functions φ and f 0 determine the terminal and running costs. The function f and sets S 0, S 1 and U(x) determine feasible states and controls.

15 Continuous Time Optimal Control The feasible solutions should satisfy ẋ(t) = f (t, x(t), u(t)) x(t 0 ) S 0, x(t 1 ) S 1 u U(x) The trajectories x(t) should start in S 0 and end in S 1, i.e., S 0 and S 1 are manifolds (intersection of surfaces).

16 Optimal Control of Car Problem: Shortest time. (= shortest path when constant speed v) ẋ(t) = v cos(θ(t)), x(0) = 0, x(t ) = x min T ẏ(t) = v sin(θ(t)), y(0), y(t ) = ȳ subj. to ω,t θ(t) = ω(t), θ(0) = 0, θ(t ) = θ ω(t) v/r, T 0

17 Orbit Transfer of Satellite Problem: Orbit transfer thrust speed ξ3 u ξ1 ξ2 gravitational center

18 Orbit Transfer of Satellite Problem: max x 1 (T ) u s.t. ẋ 1 = c3 2x 2 ẋ 2 = x 3 2 x sin u x1 2 1/c 1 c 2 c 3 t ẋ 3 = c2 3 x 2x 3 x 1 + c 3 cos u 1/c 1 c 2 c 3 t x(0) = (1 0 1) T S f = {x R 3 x 2 = 0, x 3 1 x = 0} 1

19 Resource Allocation max u( ) T 0 subj. to (1 u(t))x(t)dt + x(t ) ẋ(t) = αu(t)x(t), (0 < α < 1) x(0) = x 0 > 0 u(t) [0, 1], t A portion u(t) of the production rate x(t) is invested in the factory (and increases the production capacity) The rest (1 u(t))x(t) is stored in a warehouse. Maximimize the sum of the total amount of goods stored in the warehouse and the final capacity of the factory.

20 Optimal Control Theory Optimal control problem Dynamic programming Pontryagin minimum principle Global theory Feedback solutions Local theory Open loop control Feedback control u(t) = ψ(t, x(t)) (depends on state) Open loop control u(t) = ψ(t)

Optimal Control Theory SF 2852

university-logo Optimal Control Theory SF 2852 Ulf Jönsson Optimization and Systems Theory, Department of Mathematics Royal Institute of Technology (KTH) Spring 2011 university-logo Optimal Control Theory