Comparative study of kriging and deep neural networks as surrogate models for parameter prediction of PMSM