Explainable Deep Learning for Sustainable Solar and Wind Power Forecasting
Abstract
Solar PV and wind energy development in ongoing power systems is a means of achieving sustainability in power generation. Climatic uncertainty and renewables' variability, however, add to power forecasting volatility, which impacts energy management and reliability. Most existing deep learning models are hard to interpret and are trained to make the best prediction, limiting their ability in operational decisionmaking. For that, propose a novel Explainable Deep Learning (XDL) framework for Sustainable Solar and Wind Power Forecasting (SWPF) in this paper that addresses these drawbacks. The proposed framework integrates the multimodal extraction of meteorological features, the temporal attention-enhanced deep learning model, the explainable feature attribution, the uncertainty-aware prediction, and the reliability calibration to achieve accurate and explainable predictions of renewable energy resources. The extensive experiments demonstrate that XForecastNet is the most accurate, robust, and interpretable forecasting solution compared to traditional machine learning and state-of-the-art deep- learning approaches. The suggested framework would be beneficial for better prediction of the renewable energy sources, for a more stable grid, for more utilization of renewable energy sources, and for the possibility to operate the power system more sustainably.