Abstract
Accurately predicting the wind power output of a wind farm across varioustime scales utilizing Wind Power Forecasting (WPF) is a critical issue in windpower trading and utilization. The WPF problem remains unresolved due tonumerous influencing variables, such as wind speed, temperature, latitude, andlongitude. Furthermore, achieving high prediction accuracy is crucial formaintaining electric grid stability and ensuring supply security. In thispaper, we model all wind turbines within a wind farm as graph nodes in a graphbuilt by their geographical locations. Accordingly, we propose an ensemblemodel based on graph neural networks and reinforcement learning (EMGRL) forWPF. Our approach includes: (1) applying graph neural networks to capture thetime-series data from neighboring wind farms relevant to the target wind farm;(2) establishing a general state embedding that integrates the target windfarm's data with the historical performance of base models on the target windfarm; (3) ensembling and leveraging the advantages of all base models throughan actor-critic reinforcement learning framework for WPF.