Hybrid signal decomposition with sparrow optimized spatiotemporal network for wind power forecasting and grid resilience

Accurate wind power forecasting is critical for addressing grid instability caused by renewable energy intermittency. This study proposes a hybrid decomposition-optimization framework integrating signal processing and deep learning techniques. The methodology combines Complete Ensemble Empirical Mode Decomposition with Adaptive Noise and Variational Mode Decomposition for multiscale feature extraction, augmented by k-means clustering based on sample entropy for frequency-domain component identification. A sparrow search algorithm-optimized convolutional bidirectional long short-term memory network with attention mechanisms enables simultaneous spatial-temporal pattern recognition. Experimental validation using operational data from Inner Mongolia wind farms demonstrates significant performance improvements: R2 increased by 15.894%, with corresponding reductions in MSE (91.98%), RMSE (71.68%), MAE (71.15%), and MAPE (10.639%) compared to conventional models. The framework effectively resolves non-stationary signal processing challenges while maintaining computational efficiency, providing grid operators with actionable forecasting intervals for dispatch optimization. These advancements contribute to enhanced renewable energy integration and power system resilience.