A multi-timescale wind power forecasting method based on selection of similar days

In the progress of grid planning schemes selection, annual wind power data in different timescale is needed in the calculation of characters. In this paper, a multi-timescale wind power forecasting method based on selection of similar days is proposed. The annual wind power will be clustered by scale of days, and then the LS-SVM forecasting model of each cluster will be trained separately. In the forecasting progress, a similar day of the forecasted day is selected in its cluster as training data. The forecasting model of this cluster will be used to get forecasting results, the time-resolution of which is in level of hour. The forecasting wind power of each day will be combined into annual wind power. A case studied shows that the proposed forecasting method is correct and effective.