Deep belief network based k-means cluster approach for short-term wind power forecasting

Wind energy is the intermittent energy and its output has great volatility. How to accurately predict wind power output is a problem that many researchers have been paying attention to and urgently need to solve. In this paper, a deep belief network (DBN) model is developed for wind power forecasting. The numerical weather prediction (NWP) data was selected as the input of the proposed model and the data directly affects the prediction precision. The NWP data and wind data in the wind farm have the similar characteristics. Therefore, in this paper the k-means clustering algorithm was joined to deal with the NWP data. Through clustering analysis, a large number of NWP samples, which has the great influence in forecasting accuracy, are chosen as the input of the DBN model to improve the efficiency of the model. The DBN model was validated by the Sotavento wind farm in Spain. The results of DBN forecasting were compared with those of Back-propagation neural network (BP) and Morlet wavelet neural network (MWNN). The results show that the forecasting error of DBN model was mostly at a small level, and the forecasting accuracy of the proposed method outperforms BP and MWNN by more than 44%.