A new framework of neural network for nonlinear system modeling

In this paper, a new modeling framework of neural network for nonlinear system is proposed. We point out problems in modeling systems with traditional neural networks, that is, difficulty for analyzing internal representation, no reproducibility in system modeling (approximation), and no assumption about system property. Based on these considerations, we suggest three main improvements. The first is design of a nonlinear output function. The second is a deterministic scheme for weight initialization. The third is an updating rule for weight parameter. Simulation results show beneficial characteristics of our proposed method.