Multi-source Spatio-Temporal Feature Fusion Network for Short-Term Wind Power Forecasting

Short-term wind power forecasting is essential for ensuring grid stability, optimizing energy dispatch, and improving the economic efficiency of renewable energy systems. Despite advancements in addressing the intermittency and non-stationary distribution of wind power data, existing methods face significant limitations. Many fail to effectively capture collaborative complementarity and dimension-adaptive fusion of global features, often neglecting the preservation of intrinsic characteristics and temporal dynamics during feature extraction, which limits forecasting accuracy. To address these issues, this study introduces a multi-source spatiotemporal feature scaling and fusion network that employs a feature synergy-dynamic fusion mechanism. By integrating Lasso feature selection with adaptive spatiotemporal extraction, the framework dynamically models complementary dependencies among multi-source heterogeneous features while suppressing noise and redundant coupling. Additionally, a novel feature scaling and fusion module combined with a gated recurrent unit facilitates adaptive fusion, enhancing cross-modal feature representation. Experimental results show that our method outperforms existing models on real datasets, our method reduces MAE by 23.6%, RMSE by 21.4% and MAPE by 34.09%, thereby validating its effectiveness and practical applicability.