Lotka Volterra fishing problem (ACADO)
From mintOC
Revision as of 18:51, 31 January 2016 by FelixMueller (Talk | contribs)
This pages states the code to solve the (ralaxed) Lotka Volterra fishing problem with the code generation fo the ACADO Toolkit from Matlab (via Matlab interface). Hence one needs Matlab and ACADO. The m-file should be located in the matlab interface of ACADO and the path <ACADOROOT>/interfaces/matlab/examples/codegeneratio/nmpc/. While we use the NMPC more precisely the RTI capability of the ACADO codegeneration, this features are misused to implement a 20 step SQP solver without stopping criteria.
ACADO (Matlab)
The source code for a fixed grid discretisation with fixed stepsize integrator. This is an m-file that should be called from Matlab AFTER calling make from matlab in the matlab interface of ACADO.
% implements the lotka volterra fishing problem on mintoc.de clc; clear all; close all; Ts = 0.1; EXPORT = 1; %% Variables DifferentialState x0 x1; Control w; n_XD = length(diffStates); n_U = length(controls); % Constants c0 = 0.4; c1 = 0.2; %% Differential Equation f = dot([x0; x1]) == [ x0-x0*x1-c0*x0*w; ... -x1 + x0*x1-c1*x1*w]; h = [diffStates]; hN = [diffStates]; %% MPCexport acadoSet('problemname', 'mpc'); N = 200; ocp = acado.OCP( 0.0, 12, N ); W_mat = eye(n_XD,n_XD); WN_mat = eye(n_XD,n_XD); W = acado.BMatrix(W_mat); WN = acado.BMatrix(WN_mat); ocp.minimizeLSQ( W, h ); ocp.minimizeLSQEndTerm( WN, hN ); ocp.subjectTo( 0 <= w <= 1 ); ocp.setModel(f); mpc = acado.OCPexport( ocp ); mpc.set( 'HESSIAN_APPROXIMATION', 'GAUSS_NEWTON' ); mpc.set( 'DISCRETIZATION_TYPE', 'MULTIPLE_SHOOTING' ); mpc.set( 'SPARSE_QP_SOLUTION', 'FULL_CONDENSING_N2'); mpc.set( 'INTEGRATOR_TYPE', 'INT_IRK_GL4' ); mpc.set( 'NUM_INTEGRATOR_STEPS', N ); mpc.set( 'QP_SOLVER', 'QP_QPOASES' ); mpc.set( 'HOTSTART_QP', 'NO' ); mpc.set( 'LEVENBERG_MARQUARDT', 1e-10 ); if EXPORT mpc.exportCode( 'export_MPC' ); copyfile('../../../../../../external_packages/qpoases', 'export_MPC/qpoases') cd export_MPC make_acado_solver('../acado_MPCstep') cd .. end %% PARAMETERS OPTIMIZATION X0 = [0.5 0.7]; input.x0=X0'; Xref = [1 1]; input.x = repmat(Xref,N+1,1); Xref = repmat(Xref,N,1); input.od = []; Uref = zeros(N,n_U); input.u = Uref; input.y = [Xref(1:N,:)]; input.yN = Xref(N,:).'; input.W = diag([1 1]); input.WN = diag([1 1]); %% SOLVER LOOP display('------------------------------------------------------------------') display(' SOLVER Loop' ) display('------------------------------------------------------------------') for i=1:20 tic % Solve NMPC OCP output = acado_MPCstep(input); input.x=output.x; input.u=output.u; disp([' (RTI step: ' num2str(output.info.cpuTime*1e6) ' µs)']) end %% PLOT RESULTS t_end = 12; % States figure(1) plot([0:t_end/N:t_end],output.x) ylabel('States') xlabel('Time') legend('x1','x2') % Control figure(2) plot([0:t_end/N:t_end-t_end/N],output.u) ylabel('Control (u)') xlabel('Time')