Ultra-short-term wind power forecasting based on deep Bayesian model with uncertainty

Wind energy is an important renewable clean energy resource. However, the stochastic and volatile nature of wind power brings significant challenges to the power system's reliable and secure operation. Accurate and reliable wind power prediction is critical for the integration of wind power into the grid. The existing wind power forecasting (WPF) methods lack an assessment of the reliability of the predicted results, which may result in a financial penalty for the wind energy producers. An accurate prediction with reliability measurement is urgently needed to encounter the intricate nature of the problem. In this paper, a Bayesian framework-based bidirectional gated logic unit (BiGRU) method was proposed for ultra-short-term wind power forecasting. First, an encoder-decoder (ED) architecture was combined with a BiGRU time series modeling and feature-temporal attention (FT-Attention) to improve the accuracy of wind power prediction. Then, two uncertainty losses were applied to improve the model's performance further. The proposed method obtains the uncertainty of forecast results, which effectively eliminates the untrusted results. Our proposed method demonstrated promising results for ultra-short-term wind power forecasting due to its competitive performance compared with traditional forecasting methods.