An on-line learning radial basis function network and its application

Abstract

To improve the on-line predictive capability of radial basis function (RBF) networks, a novel sequential learning algorithm is developed referred to as sequential orthogonal model selection (SOMS) algorithm. The RBF network is adapted on-line for both network structure and connecting parameters. Based on SOMS algorithm, a multi-step predictive control strategy is introduced and applied to ship control. Simulation results of ship course control experiment demonstrate the applicability and effectiveness of the SOMS algorithm. © 2008 Springer-Verlag Berlin Heidelberg.