Abstract: This study presents an LSTM-based wind power forecasting model enhanced by multi-source meteorological data fusion. By integrating satellite remote sensing, ground observations, and numerical weather prediction, a spatiotemporal fusion framework is established. The model achieves an MAE of 8.2MW, 29.9% lower than the random forest baseline, and shows improved responsiveness to wind fluctuations via a gated adaptive mechanism. Experimental results indicate a 24.8% reduction in prediction error and a 28% gain in stability under convective weather, demonstrating its effectiveness in volatile renewable energy scenarios.