Abstract: Neural network is a widely used method in the prediction of wind power, and its training algorithm is one of important factors that affect the prediction accuracy. The authors explore two training methods: clustering method and orthogonal least square algorithm. Based on the actual data that coming from a wind farm in North China and the numerical weather prediction data, the clustering method and orthogonal least square algorithm are verified, and the difference between RBF predict models and BP predict model is finally studied. The results show that the prediction of wind power before 24 hours by using of RBF neural network model is better than that by BP neural network model. Especially RBF model based on orthogonal least squares algorithm has higher accuracy, and its prediction curve can well fit the actual power curve.