Static and Dynamic Optimization (42111)
|
|
- Arthur Higgins
- 5 years ago
- Views:
Transcription
1 Static and Dynamic Optimization (42111) Niels Kjølstad Poulsen Build. 303b, room 016 Section for Dynamical Systems Dept. of Applied Mathematics and Computer Science The Technical University of Denmark phone: mobile: :21 Lecture 8: Free dynamic optimization (D+C) 1/53
2 Outline of lecture Recap L7 Analytical solutions and Numerical methods Continuous time system Exercise DO.2 Reading guidance (DO: 11-14, 27-34) 2/53
3 Dynamic Optimization (D) Minimize J (ie. determine the sequence u i R m, i = 0,..., N 1) where: N 1 J = φ(x N )+ i=0 L i (x i,u i ) subject to (for i = 0, 1,... N 1): x i+1 = f i (x i,u i ) x 0 = x 0 Given: N horizon (number of intervals) x 0 R n initial state is known f i dynamics (R n+m+1 R n ) L i (x i,u i ) stage cost φ(x N ) terminal cost 3/53
4 Euler-Lagrange equations Defining the Hamiltonian function H i = L i (x i,u i )+λ T i+1 f i(x i,u i ) The KKT conditions on this problem (with equality constraints) results in: x i+1 = f i = λ T i = 0 T = H i x i = H i u i = ( ) T H i λ i L i +λ T i+1 f i x i x i L i +λ T i+1 f i u i u i with boundary conditions: x 0 = x 0 λ T N = x N φ N where for short: f i = f i (x i,u i ) L i = L i (x i,u i ) H i = H i (x i,u i,λ i+1 ) 4/53
5 This is a two-point boundary value problem (TPBVP) with N(2n +m) unknowns and equations. The quantity u i H i is the gradient of J wrt. u i. Equal to zero on optimal trajectories. λ 0 is the gradient of J wrt. x 0. 5/53
6 Type of solutions: Analytical solutions (for very simple problems) Semi analytical solutions (eg. the LQ problem) Numerical solutions 6/53
7 LQ problem I (Simple version). LQ Let us now focus on the problem of bringing the linear, first order system given by: x i+1 = ax i +bu i x 0 = x 0 along a trajectory from the initial state, such the cost function (for chosen p 0, q 0 and r > 0): J = 1 N 1 2 px2 N qx2 i ru2 i i=0 is minimized. The Hamiltonian for this problem is H i = 1 2 qx2 i ru2 i +λ [ ] i+1 axi +bu i and the Euler-Lagrange equations are: x i+1 = ax i +bu i λ i = qx i +aλ i+1 0 = ru i +bλ i+1 (1) (2) (3) which has the two boundary conditions x 0 = x 0 λ N = px N 7/53
8 The stationarity conditions (3), 0 = ru i +bλ i+1 give us a sequence of decisions u i = b r λ i+1 (4) if the costate is known. 8/53
9 Inspired from the boundary condition we will postulate a relationship λ i = s i x i (5) Actually separation of the variables (i and x). If we insert the control law, (4), and the costate candidate, (5), in the state equation, (1), we find or x i+1 = ax i b b r s i+1x i+1 x i+1 = [ 1+ b2 r s i+1] 1axi 9/53
10 From the costate equation, (2), we have s i x i = qx i +as i+1 x i+1 = [ q +as i+1 (1+ b2 r s i+1) 1 a ] x i which has to be fulfilled for any x i. This is the case if s i is given by the backwards recursion or s i = as i+1 (1 + b2 r s i+1) 1 a+q }{{} x 1 1+x = 1 x 1+x s i = q +s i+1 a 2 (abs i+1) 2 r +b 2 s i+1 s N = p (6) This can be solved (recursively and backwards). 10/53
11 With this solution (the sequence of s i ) we can determine the (sequence of) control actions or u i = b r λ i+1 = b r s i+1x i+1 = b r s i+1(ax i +bu i ) where u i = abs i+1 r +b 2 s i+1 x i K i = s i+1ab r +s i+1 b 2 = K i x i For the costate we have: λ i = s i x i which can be compared with (which can be proven) J = 1 2 s 0x /53
12 LQ problem - II Linear Dynamics: x i+1 = Ax+Bu x 0 = x 0 x i R n u i R m and a Quadratic objective function: J = 1 2 xt N Px N + 1 N 1 ( ) x T i 2 Qx i +u T i Ru i R > 0 i=0 The matrices, Q, R and P, are symmetric and positive semidefinite. The problem has the Hamiltonian: H i = 1 ( ) x T i 2 Qx i +u T i Ru i +λ T i+1 (Ax i +Bu i ) and the Euler-Lagrange equations are (necessary conditions): ( ) T λ H i = x i+1 = Ax+Bu x 0 = x 0 x H i = λ T i = x T i Q+λT i+1 A λt N = xt N P u H i = 0 T = u T i R+λT i+1 B 12/53
13 The solution to the LQ problem is: u i = K i x i where the gain is given by K i = ( B T S i+1 B +R ) 1 B T S i+1 A and S i is a solution to the Riccati equation S i = Q+A T S i+1 A A T S i+1 B [ B T S i+1 B +R ] 1 B T S i+1 A S N = P The matrix, S i is a symmetric, positive semidefinite matrix. Notice the Costate λ i = S i x i S i 0 which might be compared to: J = 1 2 xt 0 S 0x 0 13/53
14 Riccati Count Jacopo Francesco Riccati Born: 28 May 1676, Venice Dead: 15 April 1754, Treviso University of Padua Source: Wikipedia 14/53
15 Pause 15/53
16 Numerical methods Shooting methods (forward or backward) Gradient methods Brute force 16/53
17 Numerical methods - illustrated on a simple LQ problem. (LQ problem in the simple version). Let us now focus on the problem of bringing the linear, first order system given by: x i+1 = ax i +bu i x 0 = x 0 along a trajectory from the initial state, such the cost function: J = 1 N 1 2 px2 N + i=0 1 2 qx2 i ru2 i is minimized. The Euler-Lagrange equations are: x i+1 = ax i +bu i λ i = qx i +aλ i+1 0 = ru i +bλ i+1 (7) (8) (9) which has the two boundary conditions x 0 = x 0 λ N = px N 17/53
18 Forward shooting The stationarity condition (9) 0 = ru i +bλ i+1 gives us simply: u i = b r λ i+1 We can (for a 0) reverse the costate equation into λ i = qx i +aλ i+1 λ i+1 = λ i qx i a 18/53
19 Starting with x 0 and λ 0 ( guessed value ) we can for i = 0, 1,... N 1 iterate: λ i+1 = λ i qx i a u i = b r λ i+1 x i+1 = ax i +bu i We end up with x N and λ N which ( for correct λ 0 ) should fulfill λ N = px N ε N = λ N px N = 0 Notice: a problem if a << 1 or a >> 1. 19/53
20 Contents of a file (parms.m) setting the parameters. % Constants etc. alf=0.05; a=1+alf; b=-1; x0=50000; N=10; q=alf^2; r=q; p=q; 20/53
21 The following code (fejlf.m) solves these recursions. function err=fejlf(la0) parms % set parameters a,b,p,q,r,x0 la=la0; x=x0; for i=0:n-1, la=(la-q*x)/a; u=-b*la/r; x=a*x+b*u; end err=la-p*x; λ i+1 = λ i qx i a u i = b r λ i+1 x i+1 = ax i +bu i 21/53
22 Extented version of fejlf.m (for plotting). function [err,xt,ut,lat]=fejlf(la0) parms % set parameters a,b,p,q,r,x0 la=la0; x=x0; ut=[]; lat=la; xt=x; for i=0:n-1, la=(la-q*x)/a; u=-b*la/r; x=a*x+b*u; xt=[xt;x]; lat=[lat;la]; ut=[ut;u]; end err=la-p*x; 22/53
23 Master program (script). % The search for la0 la0g=10; % a wild guess% la0=fsolve( fejlf,la0g) % The simulation with the correct la0 [err,xt,ut,lat]=fejlf(la0); subplot(211); bar(ut); grid; title( Input sequence ); subplot(212); bar(xt); grid; title( Saldo ); 23/53
24 5 x 104 Input sequence x 104 Balance /53
25 Forward shooting method - Simplified If separation possible: reverse the costate equation and find u i from the stationarity condition. The Euler-Lagrange equations x i+1 = f i (x i,u i ) λ T i = L i (x i,u i )+λ T i+1 f i (x i,u i ) x i x i λ i+1 = h i (x i,λ i ) 0 T = L i (x i,u i )+λ T i+1 f i (x i,u i ) u i u i u i = g i (x i,λ i+1 ) Guess λ 0 (or another parameterization) and use x 0. Iterate for i = 0,1... N 1: 1 Knowing x i and λ i, determine u i and λ i+1 from the stationarity and the costate equation. 2 Update the state equation i.e. find x i+1 from x i and u i. At the end (i = N) check if λ T N = x N φ(x N ) ε = λ T N x N φ(x N ) = 0 T 25/53
26 Forward shooting method - General The Euler-Lagrange equations x i+1 = f i (x i,u i ) λ T i = L i (x i,u i )+λ T i+1 f i (x i,u i ) x i x i 0 T = L i (x i,u i )+λ T i+1 f i (x i,u i ) u i u i ( ) ( ) λi+1 xi = g u i i λ i Guess λ 0 (or another parameterization) and use x 0. Iterate for i = 0,1... N 1: 1 Knowing x i and λ i, determine u i and λ i+1 from the stationarity and the costate equation. 2 Update the state equation i.e. find x i+1 from x i and u i. At the end (i = N) check if λ T N = x N φ(x N ) ε = λ T N x N φ(x N ) = 0 T 26/53
27 Gradient methods The Euler-Lagrange equations are: x i+1 = f i (x i,u i ) λ T i = L i (x i,u i )+λ T i+1 f i (x i,u i ) x i x i 0 T = L i (x i,u i )+λ T i+1 f i (x i,u i ) u i u i = x i H i = u i H i which has the two boundary conditions x 0 = x 0 λ T N = x N φ(x N ) Guess a sequence of decisions, u i i = 0, 1,... N 1. Search for an optimal sequence of decisions, u i i = 0, 1,... N 1 using e.g. a Newton Raphson iteration: [ u j+1 i = u j 2 ] 1 i u 2 H i H i i u i 27/53
28 Brute force Guess a sequence of decisions, u i i = 0, 1,... N 1. 1 Start in x 0 and iterate the state equation forwards i.e. determine the state sequence, x i. 2 Determine the performance index. Search (e.g. using an amoeba method) for an optimal sequence of decisions, u i i = 0, 1,... N 1. 28/53
29 Pause 29/53
30 Continuous time Optimization Discrete time 0 N i Continuous time 0 T t The Schaefer model (Fish in the Baltics) x i+1 = x i +rhx i (1 αx i ) x 0 = u 0 h is the length of the intervals. The model can in continuous time be given as: ẋ t = dxt dt = rxt(1 αxt) x 0 = x 0 30/53
31 The fox(f) and rabbit(r) example. [ ] [ ] [ ṙ α = 1 r β 1 rf r F α 2 F +β 2 rf F ] 0 [ ] r0 = F 0 In general Dynamic (continuous time) state space model: ẋ 1 x 1 u ẋ 2 x. = 2 1 ft,. u ẋ m n t. x n t x 1 x 2. x n 0 = x 1 x 2.. x n 0 or in short: ẋ t = f t(x t,u t) x 0 = x 0 f : R n+m+1 R n The function, f, should be sufficiently smooth (existence and uniqueness). 31/53
32 Solution to ODE Analytical methods Numerical methods ẋ = f t(x t) x 0 = x 0 Euler integration (the most simple method) x t+h = x t +hf t(x t) is the most simple method. Others and more efficient numerical methods do exists. 32/53
33 The fox(f) and rabbit(r) example. [ ṙ F ] [ = α 1 r β 1 rf α 2 F +β 2 rf ] [ ] ur u f [ r F ] 0 [ ] r0 = F Lotka Volterra rabbit # fox, rabbit fox period 33/53
34 % function dx=dfoxc(t,x,u,a1,a2,b1,b2) % % Dynamic function for the continuous Lotka-Volterra system. % It determine the state derivative as function of the time, state vector % and system parameters. r=x(1); F=x(2); dx=[ a1*r-b1*r*f; -a2*f+b2*r*f ]-u; % # of Rabbits is the first state % # of Foxes is the second state % dx: derivative of x [ ṙ F ] [ = α 1 r β 1 rf α 2 F +β 2 rf ] [ ] ur u f [ r F ] 0 [ ] r0 = F 0 34/53
35 % function foxc % % This program simulates the trajectories for the Lotka-Volterra system. % a1=0.03; a2=0.03; b1=0.03/100; b2=b1; r=100; f=50; x0=[r;f]; Tstp=500; dt=0.1; Tspan=0:dT:Tstp; %Tspan=[0 Tstp]; u=[0.0;0.0]; % System parameters (enters in dfoxc function) % Initial # of rabbits % Initial # of foxes % Initial state value % Stop time % Step size in output data % Time span of which the solution is to be found % Alternative time span % Shootings of rabbits and foxes [time,xt]=ode45(@dfoxc,tspan,x0,[],u,a1,a2,b1,b2); % ODE solver % See dfox for dynamic % function 35/53
36 % % The rest of this program (until next function declaration) is just % plot and plot related commands plot(time,xt); grid; title( Lotka-Volterra ); ylabel( # fox, rabbit ); xlabel( period ); text(50,80, fox ); text(220,140, rabbit ); % print foxc.pps % Just a printing command % % end of main program % /53
37 Analytical solutions Discrete time Continuous time x i+1 = x i x i = C ẋ = 0 x = C x i+1 = x i +α x i = C +αi ẋ = α x t = C +αt x i+1 = ax i x i = Ca i ẋ = ax x t = Cexp(at) x i+1 = Ax i +Bu i x 0 = x 0 i x i = A i x 0 + A n j 1 Bu j j=0 ẋ = Ax+Bu x 0 = x 0 t x t = e At x 0 + e A(t s) Bu sds 0 Constant as C can be determined from boundary conditions. Examples are x 0 = x 0 or x N = x N 37/53
38 The Performance Index In discrete time we search for a sequence of decisions (u i performance index i = 0, 1,... N 1) such that the N 1 J = φ(x N )+ i=0 L i (x i,u i ) h is optimized s.t. the dynamics (and other possible constraints). In continuous time we search for a decision function (u t 0 t T) such that the performance index T J = φ T (x T )+ L t(x t,u t) dt 0 is optimized s.t. the dynamics (and other possible constraints). 38/53
39 Dynamic Optimization Free dynamic optimization: Minimize J (ie. determine the function u t, 0 t T) where: T J = φ T (x T )+ 0 L t(x t,u t) dt Objective subject to ẋ = f t(x t,u t) x 0 = x 0 Dynamics T is given x 0 is given f t(x t,u t) dynamics L t(x t,u t) kernel or running cost φ T (x T ) terminal loss and x T is free. 39/53
40 Euler-Lagrange Equations ẋ t = f t(x t,u t) λ T t = L t(x t,u t)+λ T t f t(x t,u t) x t x t 0 T = u t L t(x t,u t)+λ T t u t f t(x t,u t) with boundary conditions: x 0 = x 0 λ T T = x φ T(x T ) 40/53
41 Hamilton function Define the Hamilton function as: H(x,u,λ,t) = L t(x t,u t)+λ T t ft(xt,ut) Then the Euler-Lagrange equations (KKT conditions) for this problem can be written as: ẋ T = λ Ht λ T = x Ht 0T = u Ht H is the gradient of J wrt. u. u λ T 0 is the gradient of J wrt. x 0. The first equation is just the state equation ẋ = f t(x t,u t) 41/53
42 Properties of the Hamiltonian H t(x t,u t) = L t(x t,u t)+λ T t ft(xt,ut) Ḣ = t H + u H u+ x H ẋ+ λ H λ = t H + u H u+ = t H + u H u+ x H f +ft λ [ ] x H + λ T f = H = 0 for time invariant problems t along the optimal trajectories for x, u and λ. Motion control 42/53
43 Proof The Lagrange function for the problem is: T J L = φ T (x T )+ L t(x t,u t)dt 0 T + 0 λ T t [ft(xt,ut) ẋt]dt If partial integration T T λ T ẋdt+ λ T xdt = λ T T x T λ T 0 x is introduced the Lagrange function can be written as: J L = φ T (x T )+λ T 0 x 0 λ T T x T T + (L t(x t,u t)+λ T t f t(x t,u t)+ λ ) T t x t dt 0 and the Euler-Lagrange equations emerge from the stationarity of the Lagrange function. ( ) dj L = φ T λ T t dx T x T T ( + ) 0 x L+λT x f + λ T δxdt T ( + ) 0 u L+λT u f δudt 43/53
44 The Fundamental Lemma of the calculus of variation The Fundamental Lemma: Let f t be a continuous real-values function defined on a t b and suppose that: b f tδ t dt = 0 a for any δ t C 2 [a,b] satisfying δ a = δ b = 0. Then f t 0 t [a,b] 44/53
45 Motion control Optimal stepping (in continuous time) but in one dimension. Consider the problem of bringing the system ẋ = u x 0 = x 0 from the initial state along a trajectory such the cost J = 1 2 px2 T + T u2 t is minimized. The Hamiltonian function is H t = 1 2 u2 t +λtut and the Euler-Lagrange equations are: ẋ = u λ = 0 0 = u t +λ t with boundary conditions: x 0 = x 0 λ T = px T 45/53
46 The last two are easily solved: λ t = px T u t = λ t = px T The state equation (with the solution to u t) gives us x t = x 0 px T t from which we can find x T = and then x t = λ t = x T = x 0 px T T 1 1+pT x 0 0 for p ( ) p 1 1+pT t x 0 p 1+pT x 0 u t = p 1+pT x 0 and the Hamilton function is constant: H = 1 [ ] p 2 1+pT x 0 46/53
47 LQ problem Continuous time and Free Consider the linear dynamic system ẋ = Ax t +Bu t x 0 = x 0 and the cost function J = 1 2 xt T Px T + 1 T x T t Qx t +u T t Ru t dt 2 0 The problem has the Hamiltonian: H = 1 2 xt t Qxt ut t Rut +λt (Ax t +Bu t) and the Euler-Lagrange equations: ẋ = Ax t +Bu t x 0 = x 0 λ T t = x T t Q+λ T t A λ T T = xt T P 0 = u T t R+λT t B or ẋ = Ax t +Bu t x 0 = x 0 λ t = Qx t +A T λ t λ T = Px T u t = R 1 B T λ t 47/53
48 We will try the candidate function: Then λ t = S tx t λ t = Ṡtxt +Stẋt = Ṡtxt +St ( Ax t BR 1 B T S tx t ) If inserted in the costate equation λ t = Qx t +A T λ t Ṡtxt St ( Ax t BR 1 B T S tx t ) = Qx t +A T S tx t then for every x t: Ṡtxt = StAxt +AT S tx t +Qx t S tbr 1 B T S tx t which fulfilled if (the Riccati equation): Ṡt = StA+AT S t +Q S tbr 1 B T S t S T = P u t = R 1 B T S tx t 48/53
49 Minimum drag nose shape (Newton 1686) Find the shape i.e. r(x) of a axial symmetric nose, such that the drag is minimized. y θ a Flow x V l The decision u(x) is the slope of the profile: r = u = tan(θ) x r(0) = a 49/53
50 Minimum drag nose shape (Newton) Find the shape i.e. r(x) of a axial symmetric nose, such that the drag is minimized. a D = q C p(θ)2πrdr 0 q = 1 2 ρv 2 (Dynamic pressure) C p(θ) = 2sin(θ) 2 for θ 0 y θ a Flow x V l 50/53
51 Minimum drag nose shape (Newton) y θ a Flow x V Dynamic: r x = u r 0 = a tan(θ) = u Cost function (drag coefficient, including a blunt nose): C d = D l qπa 2 = 2r2 l +4 ru u 2dx 1 l 51/53
52 Minimum drag nose shape (Newton) r//a Cd = x/a 52/53
53 Reading guidance DO: 11-14, /53
Static and Dynamic Optimization (42111)
Static and Dynamic Optimization (421) Niels Kjølstad Poulsen Build. 0b, room 01 Section for Dynamical Systems Dept. of Applied Mathematics and Computer Science The Technical University of Denmark Email:
More informationStatic and Dynamic Optimization (42111)
Static and Dynamic Optimization (42111) Niels Kjølstad Poulsen Build. 303b, room 016 Section for Dynamical Systems Dept. of Applied Mathematics and Computer Science The Technical University of Denmark
More informationStatic and Dynamic Optimization
Static and Dynamic Optimization Course Introduction Niels Kjølstad Poulsen Informatics and Mathematical Modelling build. 321, room 16 The Technical University of Denmark email: nkp@imm.dtu.dk phone: +45
More informationDynamic Optimization
Dynamic Optimization Optimal Control Niels Kjølstad Poulsen Department of Informatics and Matematical Modelleling The Technical University of Denmark Version: 15 Januar 212 (B5) 216-11-14 11.36 2 Preface
More informationEN Applied Optimal Control Lecture 8: Dynamic Programming October 10, 2018
EN530.603 Applied Optimal Control Lecture 8: Dynamic Programming October 0, 08 Lecturer: Marin Kobilarov Dynamic Programming (DP) is conerned with the computation of an optimal policy, i.e. an optimal
More informationHomework Solution # 3
ECSE 644 Optimal Control Feb, 4 Due: Feb 17, 4 (Tuesday) Homework Solution # 3 1 (5%) Consider the discrete nonlinear control system in Homework # For the optimal control and trajectory that you have found
More informationDeterministic Dynamic Programming
Deterministic Dynamic Programming 1 Value Function Consider the following optimal control problem in Mayer s form: V (t 0, x 0 ) = inf u U J(t 1, x(t 1 )) (1) subject to ẋ(t) = f(t, x(t), u(t)), x(t 0
More informationPontryagin s maximum principle
Pontryagin s maximum principle Emo Todorov Applied Mathematics and Computer Science & Engineering University of Washington Winter 2012 Emo Todorov (UW) AMATH/CSE 579, Winter 2012 Lecture 5 1 / 9 Pontryagin
More informationFormula Sheet for Optimal Control
Formula Sheet for Optimal Control Division of Optimization and Systems Theory Royal Institute of Technology 144 Stockholm, Sweden 23 December 1, 29 1 Dynamic Programming 11 Discrete Dynamic Programming
More informationMathematical Economics. Lecture Notes (in extracts)
Prof. Dr. Frank Werner Faculty of Mathematics Institute of Mathematical Optimization (IMO) http://math.uni-magdeburg.de/ werner/math-ec-new.html Mathematical Economics Lecture Notes (in extracts) Winter
More informationLogistic Map, Euler & Runge-Kutta Method and Lotka-Volterra Equations
Logistic Map, Euler & Runge-Kutta Method and Lotka-Volterra Equations S. Y. Ha and J. Park Department of Mathematical Sciences Seoul National University Sep 23, 2013 Contents 1 Logistic Map 2 Euler and
More informationOptimal Control. Quadratic Functions. Single variable quadratic function: Multi-variable quadratic function:
Optimal Control Control design based on pole-placement has non unique solutions Best locations for eigenvalues are sometimes difficult to determine Linear Quadratic LQ) Optimal control minimizes a quadratic
More informationProblem 1 Cost of an Infinite Horizon LQR
THE UNIVERSITY OF TEXAS AT SAN ANTONIO EE 5243 INTRODUCTION TO CYBER-PHYSICAL SYSTEMS H O M E W O R K # 5 Ahmad F. Taha October 12, 215 Homework Instructions: 1. Type your solutions in the LATEX homework
More informationEE363 homework 2 solutions
EE363 Prof. S. Boyd EE363 homework 2 solutions. Derivative of matrix inverse. Suppose that X : R R n n, and that X(t is invertible. Show that ( d d dt X(t = X(t dt X(t X(t. Hint: differentiate X(tX(t =
More informationRobotics. Control Theory. Marc Toussaint U Stuttgart
Robotics Control Theory Topics in control theory, optimal control, HJB equation, infinite horizon case, Linear-Quadratic optimal control, Riccati equations (differential, algebraic, discrete-time), controllability,
More informationOptimal Control, TSRT08: Exercises. This version: October 2015
Optimal Control, TSRT8: Exercises This version: October 215 Preface Acknowledgment Abbreviations The exercises are modified problems from different sources: Exercises 2.1, 2.2,3.5,4.1,4.2,4.3,4.5, 5.1a)
More informationSuppose that we have a specific single stage dynamic system governed by the following equation:
Dynamic Optimisation Discrete Dynamic Systems A single stage example Suppose that we have a specific single stage dynamic system governed by the following equation: x 1 = ax 0 + bu 0, x 0 = x i (1) where
More informationLecture 4 Continuous time linear quadratic regulator
EE363 Winter 2008-09 Lecture 4 Continuous time linear quadratic regulator continuous-time LQR problem dynamic programming solution Hamiltonian system and two point boundary value problem infinite horizon
More informationStochastic and Adaptive Optimal Control
Stochastic and Adaptive Optimal Control Robert Stengel Optimal Control and Estimation, MAE 546 Princeton University, 2018! Nonlinear systems with random inputs and perfect measurements! Stochastic neighboring-optimal
More informationDynamical Systems & Lyapunov Stability
Dynamical Systems & Lyapunov Stability Harry G. Kwatny Department of Mechanical Engineering & Mechanics Drexel University Outline Ordinary Differential Equations Existence & uniqueness Continuous dependence
More informationHamilton-Jacobi-Bellman Equation Feb 25, 2008
Hamilton-Jacobi-Bellman Equation Feb 25, 2008 What is it? The Hamilton-Jacobi-Bellman (HJB) equation is the continuous-time analog to the discrete deterministic dynamic programming algorithm Discrete VS
More informationOptimal Control. Lecture 18. Hamilton-Jacobi-Bellman Equation, Cont. John T. Wen. March 29, Ref: Bryson & Ho Chapter 4.
Optimal Control Lecture 18 Hamilton-Jacobi-Bellman Equation, Cont. John T. Wen Ref: Bryson & Ho Chapter 4. March 29, 2004 Outline Hamilton-Jacobi-Bellman (HJB) Equation Iterative solution of HJB Equation
More informationMinimization with Equality Constraints
Path Constraints and Numerical Optimization Robert Stengel Optimal Control and Estimation, MAE 546, Princeton University, 2015 State and Control Equality Constraints Pseudoinverse State and Control Inequality
More informationA Tour of Reinforcement Learning The View from Continuous Control. Benjamin Recht University of California, Berkeley
A Tour of Reinforcement Learning The View from Continuous Control Benjamin Recht University of California, Berkeley trustable, scalable, predictable Control Theory! Reinforcement Learning is the study
More informationOptimal Control - Homework Exercise 3
Optimal Control - Homework Exercise 3 December 17, 2010 In this exercise two different problems will be considered, first the so called Zermelo problem the problem is to steer a boat in streaming water,
More informationThe Inverted Pendulum
Lab 1 The Inverted Pendulum Lab Objective: We will set up the LQR optimal control problem for the inverted pendulum and compute the solution numerically. Think back to your childhood days when, for entertainment
More informationEE C128 / ME C134 Final Exam Fall 2014
EE C128 / ME C134 Final Exam Fall 2014 December 19, 2014 Your PRINTED FULL NAME Your STUDENT ID NUMBER Number of additional sheets 1. No computers, no tablets, no connected device (phone etc.) 2. Pocket
More informationminimize x subject to (x 2)(x 4) u,
Math 6366/6367: Optimization and Variational Methods Sample Preliminary Exam Questions 1. Suppose that f : [, L] R is a C 2 -function with f () on (, L) and that you have explicit formulae for
More informationControlled Diffusions and Hamilton-Jacobi Bellman Equations
Controlled Diffusions and Hamilton-Jacobi Bellman Equations Emo Todorov Applied Mathematics and Computer Science & Engineering University of Washington Winter 2014 Emo Todorov (UW) AMATH/CSE 579, Winter
More informationSolution Methods. Richard Lusby. Department of Management Engineering Technical University of Denmark
Solution Methods Richard Lusby Department of Management Engineering Technical University of Denmark Lecture Overview (jg Unconstrained Several Variables Quadratic Programming Separable Programming SUMT
More informationOptimal Control. Lecture 3. Optimal Control of Discrete Time Dynamical Systems. John T. Wen. January 22, 2004
Optimal Control Lecture 3 Optimal Control of Discrete Time Dynamical Systems John T. Wen January, 004 Outline optimization of a general multi-stage discrete time dynamical systems special case: discrete
More informationLinear Quadratic Regulator (LQR) Design I
Lecture 7 Linear Quadratic Regulator LQR) Design I Dr. Radhakant Padhi Asst. Proessor Dept. o Aerospace Engineering Indian Institute o Science - Bangalore LQR Design: Problem Objective o drive the state
More informationOPTIMAL CONTROL. Sadegh Bolouki. Lecture slides for ECE 515. University of Illinois, Urbana-Champaign. Fall S. Bolouki (UIUC) 1 / 28
OPTIMAL CONTROL Sadegh Bolouki Lecture slides for ECE 515 University of Illinois, Urbana-Champaign Fall 2016 S. Bolouki (UIUC) 1 / 28 (Example from Optimal Control Theory, Kirk) Objective: To get from
More informationTopic # /31 Feedback Control Systems. Analysis of Nonlinear Systems Lyapunov Stability Analysis
Topic # 16.30/31 Feedback Control Systems Analysis of Nonlinear Systems Lyapunov Stability Analysis Fall 010 16.30/31 Lyapunov Stability Analysis Very general method to prove (or disprove) stability of
More informationCHAPTER 2 THE MAXIMUM PRINCIPLE: CONTINUOUS TIME. Chapter2 p. 1/67
CHAPTER 2 THE MAXIMUM PRINCIPLE: CONTINUOUS TIME Chapter2 p. 1/67 THE MAXIMUM PRINCIPLE: CONTINUOUS TIME Main Purpose: Introduce the maximum principle as a necessary condition to be satisfied by any optimal
More informationMath 337, Summer 2010 Assignment 5
Math 337, Summer Assignment 5 Dr. T Hillen, University of Alberta Exercise.. Consider Laplace s equation r r r u + u r r θ = in a semi-circular disk of radius a centered at the origin with boundary conditions
More informationThe Karush-Kuhn-Tucker conditions
Chapter 6 The Karush-Kuhn-Tucker conditions 6.1 Introduction In this chapter we derive the first order necessary condition known as Karush-Kuhn-Tucker (KKT) conditions. To this aim we introduce the alternative
More informationOptimal Control. McGill COMP 765 Oct 3 rd, 2017
Optimal Control McGill COMP 765 Oct 3 rd, 2017 Classical Control Quiz Question 1: Can a PID controller be used to balance an inverted pendulum: A) That starts upright? B) That must be swung-up (perhaps
More informationLecture 42 Determining Internal Node Values
Lecture 42 Determining Internal Node Values As seen in the previous section, a finite element solution of a boundary value problem boils down to finding the best values of the constants {C j } n, which
More information6.241 Dynamic Systems and Control
6.241 Dynamic Systems and Control Lecture 24: H2 Synthesis Emilio Frazzoli Aeronautics and Astronautics Massachusetts Institute of Technology May 4, 2011 E. Frazzoli (MIT) Lecture 24: H 2 Synthesis May
More informationCache Oblivious Stencil Computations
Cache Oblivious Stencil Computations S. HUNOLD J. L. TRÄFF F. VERSACI Lectures on High Performance Computing 13 April 2015 F. Versaci (TU Wien) Cache Oblivious Stencil Computations 13 April 2015 1 / 19
More information1. Type your solutions. This homework is mainly a programming assignment.
THE UNIVERSITY OF TEXAS AT SAN ANTONIO EE 5243 INTRODUCTION TO CYBER-PHYSICAL SYSTEMS H O M E W O R K S # 6 + 7 Ahmad F. Taha October 22, 2015 READ Homework Instructions: 1. Type your solutions. This homework
More informationLecture 9: Discrete-Time Linear Quadratic Regulator Finite-Horizon Case
Lecture 9: Discrete-Time Linear Quadratic Regulator Finite-Horizon Case Dr. Burak Demirel Faculty of Electrical Engineering and Information Technology, University of Paderborn December 15, 2015 2 Previous
More informationLecture 6. Foundations of LMIs in System and Control Theory
Lecture 6. Foundations of LMIs in System and Control Theory Ivan Papusha CDS270 2: Mathematical Methods in Control and System Engineering May 4, 2015 1 / 22 Logistics hw5 due this Wed, May 6 do an easy
More informationLinear-Quadratic-Gaussian (LQG) Controllers and Kalman Filters
Linear-Quadratic-Gaussian (LQG) Controllers and Kalman Filters Emo Todorov Applied Mathematics and Computer Science & Engineering University of Washington Winter 204 Emo Todorov (UW) AMATH/CSE 579, Winter
More informationLegendre Transforms, Calculus of Varations, and Mechanics Principles
page 437 Appendix C Legendre Transforms, Calculus of Varations, and Mechanics Principles C.1 Legendre Transforms Legendre transforms map functions in a vector space to functions in the dual space. From
More informationECSE.6440 MIDTERM EXAM Solution Optimal Control. Assigned: February 26, 2004 Due: 12:00 pm, March 4, 2004
ECSE.6440 MIDTERM EXAM Solution Optimal Control Assigned: February 26, 2004 Due: 12:00 pm, March 4, 2004 This is a take home exam. It is essential to SHOW ALL STEPS IN YOUR WORK. In NO circumstance is
More informationNumerical Optimal Control Overview. Moritz Diehl
Numerical Optimal Control Overview Moritz Diehl Simplified Optimal Control Problem in ODE path constraints h(x, u) 0 initial value x0 states x(t) terminal constraint r(x(t )) 0 controls u(t) 0 t T minimize
More informationTheoretical Justification for LQ problems: Sufficiency condition: LQ problem is the second order expansion of nonlinear optimal control problems.
ES22 Lecture Notes #11 Theoretical Justification for LQ problems: Sufficiency condition: LQ problem is the second order expansion of nonlinear optimal control problems. J = φ(x( ) + L(x,u,t)dt ; x= f(x,u,t)
More informationFeedback Optimal Control of Low-thrust Orbit Transfer in Central Gravity Field
Vol. 4, No. 4, 23 Feedback Optimal Control of Low-thrust Orbit Transfer in Central Gravity Field Ashraf H. Owis Department of Astronomy, Space and Meteorology, Faculty of Science, Cairo University Department
More informationMODERN CONTROL DESIGN
CHAPTER 8 MODERN CONTROL DESIGN The classical design techniques of Chapters 6 and 7 are based on the root-locus and frequency response that utilize only the plant output for feedback with a dynamic controller
More informationDirect Methods. Moritz Diehl. Optimization in Engineering Center (OPTEC) and Electrical Engineering Department (ESAT) K.U.
Direct Methods Moritz Diehl Optimization in Engineering Center (OPTEC) and Electrical Engineering Department (ESAT) K.U. Leuven Belgium Overview Direct Single Shooting Direct Collocation Direct Multiple
More informationAMS 27L LAB #8 Winter 2009
AMS 27L LAB #8 Winter 29 Solving ODE s in Matlab Objectives:. To use Matlab s ODE Solvers 2. To practice using functions and in-line functions Matlab s ODE Suite Matlab offers a suite of ODE solvers including:
More informationChapter 2 Optimal Control Problem
Chapter 2 Optimal Control Problem Optimal control of any process can be achieved either in open or closed loop. In the following two chapters we concentrate mainly on the first class. The first chapter
More informationLECTURE NOTES IN CALCULUS OF VARIATIONS AND OPTIMAL CONTROL MSc in Systems and Control. Dr George Halikias
Ver.1.2 LECTURE NOTES IN CALCULUS OF VARIATIONS AND OPTIMAL CONTROL MSc in Systems and Control Dr George Halikias EEIE, School of Engineering and Mathematical Sciences, City University 4 March 27 1. Calculus
More informationNon-Differentiable Embedding of Lagrangian structures
Non-Differentiable Embedding of Lagrangian structures Isabelle Greff Joint work with J. Cresson Université de Pau et des Pays de l Adour CNAM, Paris, April, 22nd 2010 Position of the problem 1. Example
More informationUNIVERSITY OF MANITOBA
Question Points Score INSTRUCTIONS TO STUDENTS: This is a 6 hour examination. No extra time will be given. No texts, notes, or other aids are permitted. There are no calculators, cellphones or electronic
More informationw T 1 w T 2. w T n 0 if i j 1 if i = j
Lyapunov Operator Let A F n n be given, and define a linear operator L A : C n n C n n as L A (X) := A X + XA Suppose A is diagonalizable (what follows can be generalized even if this is not possible -
More informationOptimal Control with Learned Forward Models
Optimal Control with Learned Forward Models Pieter Abbeel UC Berkeley Jan Peters TU Darmstadt 1 Where we are? Reinforcement Learning Data = {(x i, u i, x i+1, r i )}} x u xx r u xx V (x) π (u x) Now V
More informationPUTNAM PROBLEMS DIFFERENTIAL EQUATIONS. First Order Equations. p(x)dx)) = q(x) exp(
PUTNAM PROBLEMS DIFFERENTIAL EQUATIONS First Order Equations 1. Linear y + p(x)y = q(x) Muliply through by the integrating factor exp( p(x)) to obtain (y exp( p(x))) = q(x) exp( p(x)). 2. Separation of
More informationOptimization-Based Control. Richard M. Murray Control and Dynamical Systems California Institute of Technology
Optimization-Based Control Richard M. Murray Control and Dynamical Systems California Institute of Technology Version v2.1b (2 Oct 218) c California Institute of Technology All rights reserved. This manuscript
More informationNEAR-OPTIMAL FEEDBACK GUIDANCE FOR AN ACCURATE LUNAR LANDING JOSEPH PARSLEY RAJNISH SHARMA, COMMITTEE CHAIR MICHAEL FREEMAN KEITH WILLIAMS A THESIS
NEAR-OPTIMAL FEEDBACK GUIDANCE FOR AN ACCURATE LUNAR LANDING by JOSEPH PARSLEY RAJNISH SHARMA, COMMITTEE CHAIR MICHAEL FREEMAN KEITH WILLIAMS A THESIS Submitted in partial fulfillment of the requirements
More informationBayesian Decision Theory in Sensorimotor Control
Bayesian Decision Theory in Sensorimotor Control Matthias Freiberger, Martin Öttl Signal Processing and Speech Communication Laboratory Advanced Signal Processing Matthias Freiberger, Martin Öttl Advanced
More informationManifesto on Numerical Integration of Equations of Motion Using Matlab
Manifesto on Numerical Integration of Equations of Motion Using Matlab C. Hall April 11, 2002 This handout is intended to help you understand numerical integration and to put it into practice using Matlab
More informationSubject: Optimal Control Assignment-1 (Related to Lecture notes 1-10)
Subject: Optimal Control Assignment- (Related to Lecture notes -). Design a oil mug, shown in fig., to hold as much oil possible. The height and radius of the mug should not be more than 6cm. The mug must
More informationVectors, matrices, eigenvalues and eigenvectors
Vectors, matrices, eigenvalues and eigenvectors 1 ( ) ( ) ( ) Scaling a vector: 0.5V 2 0.5 2 1 = 0.5 = = 1 0.5 1 0.5 ( ) ( ) ( ) ( ) Adding two vectors: V + W 2 1 2 + 1 3 = + = = 1 3 1 + 3 4 ( ) ( ) a
More informationSteady State Kalman Filter
Steady State Kalman Filter Infinite Horizon LQ Control: ẋ = Ax + Bu R positive definite, Q = Q T 2Q 1 2. (A, B) stabilizable, (A, Q 1 2) detectable. Solve for the positive (semi-) definite P in the ARE:
More informationAMS 212A Applied Mathematical Methods I Lecture 14 Copyright by Hongyun Wang, UCSC. with initial condition
Lecture 14 Copyright by Hongyun Wang, UCSC Recap of Lecture 13 Semi-linear PDE a( x, t) u t + b( x, t)u = c ( x, t, u ) x Characteristics: dt d = a x, t dx d = b x, t with initial condition t ( 0)= t 0
More informationOrdinary Differential Equations
Ordinary Differential Equations In this lecture, we will look at different options for coding simple differential equations. Start by considering bicycle riding as an example. Why does a bicycle move forward?
More informationGraded Project #1. Part 1. Explicit Runge Kutta methods. Goals Differential Equations FMN130 Gustaf Söderlind and Carmen Arévalo
2008-11-07 Graded Project #1 Differential Equations FMN130 Gustaf Söderlind and Carmen Arévalo This homework is due to be handed in on Wednesday 12 November 2008 before 13:00 in the post box of the numerical
More informationInfinite-dimensional nonlinear predictive controller design for open-channel hydraulic systems
Infinite-dimensional nonlinear predictive controller design for open-channel hydraulic systems D. Georges, Control Systems Dept - Gipsa-lab, Grenoble INP Workshop on Irrigation Channels and Related Problems,
More informationESC794: Special Topics: Model Predictive Control
ESC794: Special Topics: Model Predictive Control Nonlinear MPC Analysis : Part 1 Reference: Nonlinear Model Predictive Control (Ch.3), Grüne and Pannek Hanz Richter, Professor Mechanical Engineering Department
More informationMethods for Computing Periodic Steady-State Jacob White
Introduction to Simulation - Lecture 15 Methods for Computing Periodic Steady-State Jacob White Thanks to Deepak Ramaswamy, Michal Rewienski, and Karen Veroy Outline Periodic Steady-state problems Application
More informationMinimization with Equality Constraints!
Path Constraints and Numerical Optimization Robert Stengel Optimal Control and Estimation, MAE 546, Princeton University, 2017 State and Control Equality Constraints Pseudoinverse State and Control Inequality
More informationLecture 5 Linear Quadratic Stochastic Control
EE363 Winter 2008-09 Lecture 5 Linear Quadratic Stochastic Control linear-quadratic stochastic control problem solution via dynamic programming 5 1 Linear stochastic system linear dynamical system, over
More informationNumerical Methods for ODEs. Lectures for PSU Summer Programs Xiantao Li
Numerical Methods for ODEs Lectures for PSU Summer Programs Xiantao Li Outline Introduction Some Challenges Numerical methods for ODEs Stiff ODEs Accuracy Constrained dynamics Stability Coarse-graining
More informationQuadratic Stability of Dynamical Systems. Raktim Bhattacharya Aerospace Engineering, Texas A&M University
.. Quadratic Stability of Dynamical Systems Raktim Bhattacharya Aerospace Engineering, Texas A&M University Quadratic Lyapunov Functions Quadratic Stability Dynamical system is quadratically stable if
More informationLinear Quadratic Regulator (LQR) Design II
Lecture 8 Linear Quadratic Regulator LQR) Design II Dr. Radhakant Padhi Asst. Professor Dept. of Aerospace Engineering Indian Institute of Science - Bangalore Outline Stability and Robustness properties
More informationEE C128 / ME C134 Feedback Control Systems
EE C128 / ME C134 Feedback Control Systems Lecture Additional Material Introduction to Model Predictive Control Maximilian Balandat Department of Electrical Engineering & Computer Science University of
More informationAlberto Bressan. Department of Mathematics, Penn State University
Non-cooperative Differential Games A Homotopy Approach Alberto Bressan Department of Mathematics, Penn State University 1 Differential Games d dt x(t) = G(x(t), u 1(t), u 2 (t)), x(0) = y, u i (t) U i
More informationLecture 9 Approximations of Laplace s Equation, Finite Element Method. Mathématiques appliquées (MATH0504-1) B. Dewals, C.
Lecture 9 Approximations of Laplace s Equation, Finite Element Method Mathématiques appliquées (MATH54-1) B. Dewals, C. Geuzaine V1.2 23/11/218 1 Learning objectives of this lecture Apply the finite difference
More informationDeterministic Optimal Control
page A1 Online Appendix A Deterministic Optimal Control As far as the laws of mathematics refer to reality, they are not certain; and as far as they are certain, they do not refer to reality. Albert Einstein
More informationAlgorithms for Constrained Optimization
1 / 42 Algorithms for Constrained Optimization ME598/494 Lecture Max Yi Ren Department of Mechanical Engineering, Arizona State University April 19, 2015 2 / 42 Outline 1. Convergence 2. Sequential quadratic
More informationSolution of Stochastic Optimal Control Problems and Financial Applications
Journal of Mathematical Extension Vol. 11, No. 4, (2017), 27-44 ISSN: 1735-8299 URL: http://www.ijmex.com Solution of Stochastic Optimal Control Problems and Financial Applications 2 Mat B. Kafash 1 Faculty
More informationCE 191: Civil and Environmental Engineering Systems Analysis. LEC 05 : Optimality Conditions
CE 191: Civil and Environmental Engineering Systems Analysis LEC : Optimality Conditions Professor Scott Moura Civil & Environmental Engineering University of California, Berkeley Fall 214 Prof. Moura
More informationState Regulator. Advanced Control. design of controllers using pole placement and LQ design rules
Advanced Control State Regulator Scope design of controllers using pole placement and LQ design rules Keywords pole placement, optimal control, LQ regulator, weighting matrixes Prerequisites Contact state
More informationChapter 3 Second Order Linear Equations
Partial Differential Equations (Math 3303) A Ë@ Õæ Aë áöß @. X. @ 2015-2014 ú GA JË@ É Ë@ Chapter 3 Second Order Linear Equations Second-order partial differential equations for an known function u(x,
More informationWeak Convergence of Numerical Methods for Dynamical Systems and Optimal Control, and a relation with Large Deviations for Stochastic Equations
Weak Convergence of Numerical Methods for Dynamical Systems and, and a relation with Large Deviations for Stochastic Equations Mattias Sandberg KTH CSC 2010-10-21 Outline The error representation for weak
More informationStabilization and Passivity-Based Control
DISC Systems and Control Theory of Nonlinear Systems, 2010 1 Stabilization and Passivity-Based Control Lecture 8 Nonlinear Dynamical Control Systems, Chapter 10, plus handout from R. Sepulchre, Constructive
More informationModern Optimal Control
Modern Optimal Control Matthew M. Peet Arizona State University Lecture 19: Stabilization via LMIs Optimization Optimization can be posed in functional form: min x F objective function : inequality constraints
More informationMathematics for Engineers II. lectures. Differential Equations
Differential Equations Examples for differential equations Newton s second law for a point mass Consider a particle of mass m subject to net force a F. Newton s second law states that the vector acceleration
More informationOptimal control of nonlinear systems with input constraints using linear time varying approximations
ISSN 1392-5113 Nonlinear Analysis: Modelling and Control, 216, Vol. 21, No. 3, 4 412 http://dx.doi.org/1.15388/na.216.3.7 Optimal control of nonlinear systems with input constraints using linear time varying
More informationLecture 4. Chapter 4: Lyapunov Stability. Eugenio Schuster. Mechanical Engineering and Mechanics Lehigh University.
Lecture 4 Chapter 4: Lyapunov Stability Eugenio Schuster schuster@lehigh.edu Mechanical Engineering and Mechanics Lehigh University Lecture 4 p. 1/86 Autonomous Systems Consider the autonomous system ẋ
More informationThe roots are found with the following two statements. We have denoted the polynomial as p1, and the roots as roots_ p1.
Part II Lesson 10 Numerical Analysis Finding roots of a polynomial In MATLAB, a polynomial is expressed as a row vector of the form [an an 1 a2 a1 a0]. The elements ai of this vector are the coefficients
More informationDeterministic Optimal Control
Online Appendix A Deterministic Optimal Control As far as the laws of mathematics refer to reality, they are not certain; and as far as they are certain, they do not refer to reality. Albert Einstein (1879
More informationTrajectory-based optimization
Trajectory-based optimization Emo Todorov Applied Mathematics and Computer Science & Engineering University of Washington Winter 2012 Emo Todorov (UW) AMATH/CSE 579, Winter 2012 Lecture 6 1 / 13 Using
More informationPrinciples of Optimal Control Spring 2008
MIT OpenCourseWare http://ocw.mit.edu 16.323 Principles of Optimal Control Spring 2008 For information about citing these materials or our Terms of Use, visit: http://ocw.mit.edu/terms. 16.323 Lecture
More informationOptimisation and optimal control 1: Tutorial solutions (calculus of variations) t 3
Optimisation and optimal control : Tutorial solutions calculus of variations) Question i) Here we get x d ) = d ) = 0 Hence ṫ = ct 3 where c is an arbitrary constant. Integrating, xt) = at 4 + b where
More informationLinear and Loglinear Approximations (Started: July 7, 2005; Revised: February 6, 2009)
Dave Backus / NYU Linear and Loglinear Approximations (Started: July 7, 2005; Revised: February 6, 2009) Campbell s Inspecting the mechanism (JME, 1994) added some useful intuition to the stochastic growth
More informationDefinite versus Indefinite Linear Algebra. Christian Mehl Institut für Mathematik TU Berlin Germany. 10th SIAM Conference on Applied Linear Algebra
Definite versus Indefinite Linear Algebra Christian Mehl Institut für Mathematik TU Berlin Germany 10th SIAM Conference on Applied Linear Algebra Monterey Bay Seaside, October 26-29, 2009 Indefinite Linear
More information