Evolutionary training of a q-Gaussian radial basis functional-link nets for function approximation

Abstract

In this paper, radial basis functional-link nets (RBFLNs) based on a q-Gaussian function is proposed. In order to enhance the generalization performance of a modified radial basis function neural network and enhance the performance of the new network, the evolutionary algorithm named real-coded chemical reaction optimization (RCCRO), is presented for training the new network. A developed RCCRO, has been shown to perform well in many optimization problems. A RCCRO is employed to select the non-extensive entropic index q and the other parameters of the network. The experimental results of the function approximation show that the proposed approach can improve the performance of RBFLNs.