A modified forecasting algorithm for wind power based on SVM

Aiming at the saturation characteristic of SVM in large sample environment, a modified SVM forecasting algorithm for wind power forecasting is proposed in this paper. The key point of the modified SVM forecasting algorithm is converting large sample set to small sample set by making classification. In this method, the optimal regression size for SVR is firstly sought out for the actual sample, and then the training samples are divided into several categories according to wind power output with different class labels. Based on SVC, train out classification model; based on SVR, regression model of each class can be built. Forecast data of wind power can be obtained by taking the text data into above classification model and corresponding regression model. At last, the proposed algorithm is applied to a wind farm of Shandong Province; and the results verify its validity and effectiveness.