Research on applicability of ultra-short term wind power forecasting models

The intermittent and variation of wind power pose a threat to the dispatch of electric grid, while wind power prediction is an effective way to solve this problem. The time scale of ultra-short term wind power prediction usually stays between the levels of minutes to hours, thus it can provide real-time technique supports to power grid dispatching and make sure the secure, stable and economic operation of electric power system. Three algorithms are adopted to establish the ultrashort term wind power prediction model, which including BP, SVM and RBF. The predicted models are applied to carry on wind power prediction in three real wind farms which locates in different regions of China, and the applicability of these three models was analysed via the comparison of various predicted results. Besides, a wind farm which has the highest predicted efficiency was taken as an example, so that we can make an analysis of the predicted error caused by different time span. The whole prediction results show that the annual variation of prediction error will change with wind farm and seasonal factors. With the increase of time span, the RMSE of forecasting power takes on ascending trend. The higher average forecasting error is, the higher error of every single predicted point, which is the same with low error situation. Due to the frequent variation of weather system in spring and winter, the three forecasting models have unstable predicted effectiveness in those two seasons. By contrast, the BP-ANN and RBF-ANN models have better forecasting results, while the SVM model acts a little bit worse.