New learning algorithm for RBF neural networks with applications to nonlinear system identification

Abstract

An identification technique is presented for nonlinear discrete-time multivariable dynamical systems based on RBF neural nets. The ways to fix the neural net structure and the weights are addressed as two different problems with separately developed on-line algorithms for their determination. Currently, the determination of the RBF net structure is still heuristics-based and this may lead to modelling error, and possible breakdown of the weight updating algorithm. Thus, there is a need to develop theory that can help aid the generation of RBF neural net structures.