A novel framework for medium-term wind power prediction based on temporal attention mechanisms

Wind energy is a widely distributed, recyclable and environmentally friendly energy source that plays an important role in mitigating global warming and energy shortages. Wind energy's uncertainty and fluctuating nature makes grid integration of large-scale wind energy systems challenging. Medium-term wind power forecasts can provide an essential basis for energy dispatch, so accurate wind power forecasts are essential. Much research has yielded excellent results in recent years. However, many of them require additional experimentation and analysis when applied to other data. In this paper, we propose a novel short-term forecasting framework by tree-structured parzen estimator (TPE) and decomposition algorithms. This framework defines the TPE-VMD-TFT method for 24-h and 48-h ahead wind power forecasting based on variational mode decomposition (VMD) and time fusion transformer (TFT). In the Engie wind dataset from the electricity company in France, the results show that the proposed method significantly improves the prediction accuracy. In addition, the proposed framework can be used to other decomposition algorithms and require little manual work in model training.