Now after using common optimization to improve the BP algorithm,the improved BP neural network almost become very complex and consume more human resources during the prediction process.To solve these shortcomings,this paper presents a transfer function self-optimization algorithm to improve the neural network,then apply the improved network to wind power prediction.Take a period time of operating data in a northeast wind farm as the experimental samples to analyze prddiction outcomes by using both traditional and improved BP neural network.Prediction results show that the improved BP neural network not only enhances the convergence reat,but also prediction accuracy.