Multiple RBF networks for on-line identification of nonlinear systems

In this paper, algorithms for the use of multiple neural networks for on-line identification are developed. The radial basis function (RBF) networks with only linear weights requiring estimation combined with the Kalman filter algorithm forms the essence of the identification algorithm. The algorithms for uni-modal and multi-modal system identification with multiple networks makes use of the multiple model algorithm which performs on-line model selection. Two algorithms for identification of multimodal systems are introduced based on `hard´ and `soft´ switching methodologies. In the former, only the most probable network undergoes adaptation by the Kalman filter and in the latter all networks are adapted by appropriate weighting of the observation