Abstract: For the problem of RBF parameter optimization neural network weights, and presents a method of RBF neural network optimization algorithm based on the improvement gray wolf. A nonlinear algorithm for convergence of the proposed algorithm convergence precision of gray wolf low.The hidden layer to the output layer weights matrix mapping algorithm in artificial to wolf, wolf using optimization algorithm has fast convergence speed and global search ability of the hidden layer of RBF network to the output layer weights are optimized to improve the RBF neural network. This paper uses the KDD CUP 99 data set for experiment, the experimental results show that the effect of classification results the proposed algorithm has better better detection and classification, and enhance the processing ability of the RBF neural network for nonlinear problems in a certain extent.