An alternate radial basis function neural network model

A new robust RBF neural network model is presented that when compared with a conventional RBF neural network has mathematically sound learning properties and better function approximation capabilities. The proposed RBF function uses log-sigmoid functions as the basis function that eliminates any risk of mathematical instabilities, as can be the case during the learning phase of the Gaussian basis radial function networks. The performance of the proposed scheme is illustrated by simulation results of a nonlinear system identification problem. The results indicate that the proposed model performs well for nonlinear system identification problems.