Multiobjective Criteria for Nonlinear Model Selection and Identification with Neural Networks

This paper presents anew approach to model selection and identification of nonlinear systems via neural networks and genetic algorithms, based on multiobjective performance criteria. It considers three performance indices (or cost functions) in the objectives, which are the distance measurement and maximum difference measurement between the real nonlinear system and the nonlinear model, and the complexity measurement of the nonlinear model, instead of single performance index. The Volterra polynomial basis function network and the Gaussian radial basis function network are applied to approximate the nonlinear system. A numerical algorithm for multiobjective nonlinear model selection and identification using neural networks and genetic algorithms is developed.