Wind power prediction based on sequential time clustering using SVM

Wind power prediction is important to synchronization of wind power. Conventional statistical methods were improved by considering the daily similarity of wind speed, but the sequential information was still ignored. This article puts forward a new method based on similarity and sequential time clustering, in which a year is divided into several continuous time series by clustering twice, with one catching the daily similarity and the other capturing the continuous wind statistics. Furthermore, different from the usual ANN model, SVM modeling is employed in the article to avoid trapping into local optimal. Experiment on a wind farm shows that it gains the error (RMSE/Installed Capacity) of 16.04%, consistently outperforming the method considering only daily similarity by relatively 7.2%.