Workaround sought for required accuracy error

[stuart]

I would appreciate some guidance on working around the following errors, which occur whether or not the optimization is constrained:

[ACADO] Error: The integration routine stopped as the required accuracy can not be obtained
Code: (RET_UNSUCCESSFUL_RETURN_FROM_INTEGRATOR_BDF)
File: /home/stuartg/ACADOtoolkit/src/integrator/integrator_bdf.cpp
Line: 2021

[ACADO] Error: Initialization of NLP solver failed
Code: (RET_UNABLE_TO_INTEGRATE_SYSTEM)
File: /home/stuartg/ACADOtoolkit/src/dynamic_discretization/shooting_method.cpp
Line: 279

[ACADO] Error: Initialization of NLP solver failed
Code: (RET_NLP_INIT_FAILED)
File: /home/stuartg/ACADOtoolkit/src/nlp_solver/scp_method.cpp
Line: 214

[ACADO] Error: Initialization of optimization algorithm failed
Code: (RET_OPTALG_INIT_FAILED)
File: /home/stuartg/ACADOtoolkit/src/optimization_algorithm/optimization_algorithm_base.cpp
Line: 591

Attached is a Pontryagin Maximum output for unconstrained optimization.

The toy problem is shown below:

#include <acado_toolkit.hpp>
#include <include/acado_gnuplot/gnuplot_window.hpp>

int main( ){

    USING_NAMESPACE_ACADO

// INTRODUCE THE VARIABLES
//  -------------------------------------
    DifferentialState        uu,k,temp   ;     // the differential states
    AlgebraicState           inv,prod,dam;     // the algebraic states
    Control                  mu,c        ;     // the control inputs
//  Parameter                T           ;     // the time horizon T
    DifferentialEquation     f           ;     // the differential equation
    TIME                     t           ;     // time

    const double t_start = 0.0;
    const double t_end   = 100.0;

// DEFINE A DIFFERENTIAL EQUATION
//  -------------------------------------
    f << dot(uu)   == -exp(-0.015*t)*(-1 + c)/c;  // an implementation
    f << dot(k)    == inv-0.04*k;                 // of the model equations
    f << dot(temp) == pow(k,0.85)*(1-mu)/1000;   
    f<<          0 == prod-pow(k,0.85)*dam;
    f<<          0 == inv-prod+c;
    f<<          0 == dam-1/(1+0.01*pow(temp,2));

// DEFINE AN OPTIMAL CONTROL PROBLEM
//  -------------------------------------   
    OCP ocp( t_start, t_end, 10);            // time horizon & steps
    ocp.minimizeMayerTerm( uu );              // minimization obj

    ocp.subjectTo( f                       );  // minimize obj s.t. the model,
    ocp.subjectTo( AT_START, uu   == 0.0   );  // initial values:
    ocp.subjectTo( AT_START, k    == 8.0   );
    ocp.subjectTo( AT_START, c    == 1.7   );
    ocp.subjectTo( AT_START, temp == 0.5   );
    ocp.subjectTo( AT_START, mu   == 0.005 );

//    ocp.subjectTo( AT_END  , s == 10.0 );   // the terminal constraints for s
//    ocp.subjectTo( AT_END  , v ==  0.0 );   // and v,

    ocp.subjectTo( 0.0 <= mu  <=  1.0   );
    ocp.subjectTo( 0.0 <= k   <=  5.0e6 );
    ocp.subjectTo( 0.0 <= inv <=  0.3e6 );
//  -------------------------------------

    GnuplotWindow window;
        window.addSubplot( uu, "THE UTILITY uu"         );
        window.addSubplot( k, "THE ASSETS k"            );
        window.addSubplot( c, "THE CONSUMPTION c"       );
    window.addSubplot( inv, "THE INVESTMENT inv"    );
    window.addSubplot( mu, "THE ABATEMENT mu"       );
        window.addSubplot( temp, "THE TEMPERATURE temp" );

// DEFINE AN OPTIMIZATION AND SOLVE THE OCP:
//  -------------------------------------   
    OptimizationAlgorithm algorithm(ocp);     // the optimization algorithm

//    algorithm.set( ABSOLUTE_TOLERANCE    , 1e-7          );
//    algorithm.set( INTEGRATOR_TOLERANCE  , 1e-7          );
//    algorithm.set( HESSIAN_APPROXIMATION , EXACT_HESSIAN );

    algorithm << window;
    algorithm.solve();                        // solves the problem.

    return 0;
}

[stuart]

Hi, thought I would bump this post to see if anyone might suggest why this problem cannot solve? Thanks, Stuart

[hcl734]

Any updates? Running in the same problem