Study of User-Specified Parameters in Nonlinear Dynamics Modeling with Neural Networks

This work presents a study of the NN-NLSS method, an algorithm to model nonlinear dynamic systems that brings together the advantages of classic system identification on one hand, and some attractive properties of machine learning techniques on the other hand. This paper aims at gaining more insight on the approach, focusing in particular on the impact that different user-specified settings have on the algorithm performance. The analysis is performed by means of a simulation problem inspired by the challenging Wiener-Hammerstein benchmark example.