Nonlinear System Identification Using RBF Networks With Linear Input Connections

Abstract

This paper presents a modified RBF network with additional linear input connections together with a hybrid training algorithm. The training algorithm is based on kmeans clustering with square root updating method and Givens least squares algorithm with additional linear input connections features. Two real data sets have been used to demonstrate the capability of the proposed RBF network architecture and the new hybrid algorithm. The results indicated that the network models adequately represented the systems dynamic.