USE dmd predict wrong result

Start with the CAROM::DMD calls to takeSample, train, projectInitialCondition, and predict, then inspect getParametricDMD with the IMQ and LS options. Reproduce the reported case using the supplied snapshot values and compare predictions at 600 and 18000 seconds; done means explaining or correcting the unexpected 116 and -333 results.

generator dmd 

for(auto fileName:fileNames)
    {

        std::vector<Eigen::MatrixXd> temp =utility.creatMedSingleSnapshot(ON_NODES,fileName,meshName,filedName,true);
        values.push_back(temp);

        int SizeData = temp.front().size();
        double dt = 600.;//600 second
        CAROM::DMD *dmd_u=new CAROM::DMD(SizeData,dt);

        for(int i = 0;i<temp.size();++i)
        {
            dmd_u->takeSample(values[index][i].data(),i*dt);
        }

        dmd_u->train(16);
        dmds.push_back(dmd_u);



        double flux = p[index].front();
        // double flux = 400.;
        std::shared_ptr< CAROM::Vector> param_vector(new CAROM::Vector(1, false));
        (*param_vector)(0) = flux;
        param_vectors.push_back(param_vector);



        index++;
    }

predict code

CAROM::DMD *dmd_u_p;
    CAROM::Vector desired_param(1, false);
    desired_param(0) = 70.;//load flux
    CAROM::getParametricDMD(dmd_u_p, param_vectors_ptr, dmds, &desired_param,
                            "IMQ", "LS", closest_rbf_val);

    std::unique_ptr<CAROM::Vector> init(new CAROM::Vector(5548,true));//FEM point size is 5548
    (*init)=20.;//init value
    dmd_u_p->projectInitialCondition(init.get());
    double predictTime = 18000.;
    auto value2 =dmd_u_p->predict(predictTime);

one point value in snapshot as follow: step time 6000

20
48.2409
60.205
65.475
67.8718
68.9871
69.5142
69.7658
69.8868
69.9451
69.9734
69.9871
69.9937
69.997
69.9985
69.9993
69.9996
69.9998
69.9999
70
70
70
70
70
70
70
70
70
70
70
70

when i predict 600s the value is 116 adn when i predict 18000s the value is -333. i don't konw why ?