Improving short-term wind power forecasting accuracy is an urgent requirement for the development of large-scale wind power, and it is also the key to ensuring the integrated operation of wind power.In this paper, a method to improve the forecasting accuracy based on clustering is proposed without increasing the complexity of the model.First, training samples are processed by the fuzzy C-means clustering method optimized by subtractive clustering.Then the forecasting model base corresponding to different data set is established.Finally, different forecasting data are matched to the data in the clustered data set so that the optimized model is selected for wind power forecasting.A lot of actual data of wind farm in Shanxi province are used for simulation study.The results show that it can reduce the number of large prediction error so as to effectively improve the forecasting accuracy of wind power.