A temporal window attention mechanism for short-term wind power forecasting

Accurate prediction of wind power is of great significance for the safe and stable operation of the power system. To address this issue, many data driven approaches are proposed but their prediction accuracy usually decline substantially in the case of severe wind speed fluctuations. This article proposes a temporal window attention mechanism for wind power prediction based on wind speed fluctuations identification. A wind speed fluctuation feature extraction method and fuzzy C-means (FCM) based clustering method is proposed to accurately identify different fluctuation period. Then, a temporal window attention mechanism is proposed to ameliorate the over-smoothing phenomenon that can occur in conventional temporal prediction models. Wind farm data from a site in China is used to validate the prediction performance, and the results show that the method effectively reduces the loss of prediction accuracy under wind speed fluctuations.