A spatiotemporal wind power forecasting method based on dual-view graph fusion and dual-granularity residual learning.

Accurate wind power forecasting for turbine clusters is critical for reliable power system operation. Existing spatiotemporal graph neural network (ST-GNN) methods typically rely on a single static graph representation, which limits their ability to capture diverse spatial relationships. To address these issues, we propose a Spatiotemporal Dual-View Graph Network (ST-DVGN) for wind power prediction, which models turbine interactions from two complementary views: physical spatial proximity and historical power statistical dependency. Instead of introducing complex dynamic or heterogeneous graphs, the proposed framework constructs dual-view graph representations over a fixed topology and integrates them through a unified fusion mechanism. In addition, a dual-granularity temporal module is designed to capture both local fluctuations and long-term trends in wind power time series. Extensive experiments demonstrate that ST-DVGN consistently outperforms baselines in prediction accuracy. Moreover, under cluster-level incomplete observation scenarios, the RMSE of ST-DVGN decreases by a maximum of only 6%, with the model still maintaining high prediction stability and robustness. These results indicate that dual-view spatial modeling provides an effective and robust solution for wind power forecasting in realistic operating environments.