Short-Term Wind Power Prediction via Spatial Temporal Analysis and Deep Residual Networks

Abstract

Wind power is a rapidly growing source of clean energy. Accurate short-term forecasting of wind power is essential for reliable energy generation. In this study, we propose a novel wind power forecasting approach using spatiotemporal analysis to enhance forecasting performance. First, the wind power time-series data from the target turbine and adjacent neighboring turbines were utilized to form a graph structure using graph neural networks (GNN). The graph structure was used to compute the spatiotemporal correlation between the target turbine and adjacent turbines. Then, the prediction models were trained using a deep residual network (DRN) for short-term wind power prediction. Considering the wind speed, the historic wind power, air density, and historic wind power in adjacent wind turbines within the supervisory control and data acquisition (SCADA) system were utilized. A comparative analysis was performed using conventional machine-learning approaches. Industrial data collected from Hami County, Xinjiang, China, were used for the case study. The computational results validate the superiority of the proposed approach for short-term wind-power forecasting.