A method based on RBF neural network to directly forecast wind power for the future 24 h is proposed. To overcome the defects of traditional clustering algorithm, a fuzzy clustering algorithm combining with genetic algorithm, simulated annealing algorithm and pattern recognition is proposed based on fuzzy C-means algorithm. The fuzzy clustering algorithm and some common methods are used to select the center of radial basis function, and the learning of weight is solved by Orthogonal Least Square (OLS), based on measured data. The results indicate that the method to forecast wind power can improve the prediction accuracy, and prove the superiority of integrated clustering algorithm.