A Generalized Regression Neural Network Based on Fuzzy Means Clustering and Its Application in System Identification

Abstract

A method to simplify the generalized regression neural networks (GRNN) structure with a large numbers of training samples is proposed. The amount of pattern units is proportionate to the training samples. So in order to simplify the GRNN's structure, some of the representative samples should be selected to build the network. This paper takes the fuzzy means clustering algorithm. It combines with a similarity measurement, which is calculated between input elements, to find the best clustering centers. According to the simulation results, this strategy can largely simplify the GRNN's structure and significantly improve the network's efficiency with just a tiny of loss in accuracy. The network structure built in this strategy can learn quickly, and is suitable to deal with the problems of nonlinear system identification.