root/pmsm/pmsm_unkQpf.cpp @ 78

/*
  \file
  \brief Models for synchronous electric drive using IT++ and BDM
  \author Vaclav Smidl.

  -----------------------------------
  BDM++ - C++ library for Bayesian Decision Making under Uncertainty

  Using IT++ for numerical operations
  -----------------------------------
*/

#include <itpp/itbase.h>
#include <estim/libKF.h>
#include <estim/libPF.h>
#include <stat/libFN.h>

#include "pmsm.h"

using namespace itpp;

//!Extended Kalman filter with unknown \c Q
class EKF_unQ : public EKFCh , public BMcond {
public:
        //! Default constructor
        EKF_unQ ( RV rx, RV ry,RV ru,RV rQ ) :EKFCh ( rx,ry,ru ),BMcond ( rQ ) {};
        void condition ( const vec &Q0 ) {
                Q.setD ( Q0,0 );
                //from EKF
                preA.set_submatrix ( dimy+dimx,dimy,Q._Ch() );
        }; 
};

int main() {
        // Kalman filter
        int Ndat = 10000;

//      cout << KF;
        // internal model
        IMpmsm fxu;
        //                  Rs    Ls        dt       Fmag(Ypm) kp   p    J     Bf(Mz)
        fxu.set_parameters ( 0.28, 0.003465, 20*1e-6, 0.1989,   1.5 ,4.0, 0.04, 0.0 );
        // observation model
        OMpmsm hxu;

        vec mu0= "100 100 100 1";
        vec Qdiag ( "0.1 0.1 0.01 0.00001" );
        vec Rdiag ( "0.02 0.02" );
        vec vQ = "0.01:0.01:100";

        chmat Q ( Qdiag );
        chmat R ( Rdiag );

        RV rQ ( "100","{Q}","2","0" );
        EKF_unQ KFE ( rx,ry,ru,rQ );
        KFE.set_parameters ( &fxu,&hxu,Q,R );
        KFE.set_est ( mu0, chmat ( 1000*ones ( 4 ) ) );

        mgamma evolQ ( rQ,rQ );
        //evolQ.set_parameters ( 10000.0 ); //sigma = 1/10 mu

        MPF<EKF_unQ> M ( rx,rQ,evolQ,evolQ,100,KFE );

        epdf& KFEep = KFE._epdf();
        epdf& Mep = M._epdf();
        // initialize
        evolQ.set_parameters ( 1.0 ); //sigma = 1/10 mu
        evolQ.condition ( "0.5 0.5" );
        epdf& pfinit=evolQ._epdf();
        M.set_est ( pfinit );
        evolQ.set_parameters ( 1000.0 ); //sigma = 1/10 mu

        //simulator values
        vec dt ( 2 ); // output (isa isb)
        vec wt ( 2 ); // noise on dt
        vec ut ( 2 ); //
        vec xt=mu0;   // initial state
        vec et ( 4 ); // noise on xt

        mat Xt=zeros ( 4,Ndat ); // True trajetory of xt
        mat XtE=zeros ( 4,Ndat ); // Estimate of xt using EKF (known Q)
        mat XtM=zeros ( 6,Ndat ); // Estimate of xt using EKF-MPF
        Xt.set_col ( 0,mu0 );
        XtM.set_col ( 0,Mep.mean() );

        for ( int t=1;t<Ndat;t++ ) {
                //simulator
                // UniRNG.sample_vector ( 2,wt );

                if ( rem ( t,500 ) <200 ) ut = rem ( t,500 ) *ones ( 2 );
                else
                        ut=zeros ( 2 );

                NorRNG.sample_vector ( 4,et );
                NorRNG.sample_vector ( 2,wt );
                xt = fxu.eval ( xt,ut ) + Q.sqrt_mult ( et );
                dt = hxu.eval ( xt,ut ) + R.sqrt_mult ( wt );

                //estimator
                KFE.bayes ( concat ( dt,ut ) );
                M.bayes ( concat ( dt,ut ) );

                Xt.set_col ( t,xt );
                XtE.set_col ( t,KFEep.mean() );
                XtM.set_col ( t,Mep.mean() );
        }

        it_file fou ( "pmsm.it" );

        fou << Name ( "xth" ) << Xt;
        fou << Name ( "xthE" ) << XtE;
        fou << Name ( "xthM" ) << XtM;
        //Exit program:
        return 0;

}