Abstract: An algorithm based on parameter optimization(KV-RBF)is proposed for optimizing RBF(radial basis function) neural network structure. Firstly， the K-means++ algorithm， which makes the clustering algorithm more accurate， is improved to find an appropriate initial center for the hidden layer node of RBF neural network. Then， considering the influence of data distribution and scaling factor selection， the width of neuron basis function in the hidden layer is calculated by variance measurement. Finally， the network parameters are modified to improve the nonlinear approximation ability of the network. The experimental results show that the proposed RBF neural network based on parameter optimization has good approximation effect and generalization ability.

Key words: radial basis function(RBF), neural network, K-means++ algorithm, kernel function, data density estimation