Short-term Wind Power Cluster Prediction Method Based on Multi-Spatial-Scale Features

To address the high modeling complexity in wind power cluster prediction caused by wide-area high-dimensional meteorological features, this study builds feature extraction modules at global and wind farm scales. Specifically, it uses Global Block-wise Attention Transformer for global feature processing and Dynamic Hybrid Sparse Graph Attention for wind farm-level feature extraction to reduce model complexity and improve computational efficiency. Applied to a Jilin wind power cluster, results show the method effectively enhances model computational efficiency and wind power output prediction accuracy.