Ensemble Deep Learning–Based Wind Power Forecasting With Self‐Adaptive Osprey Optimization Algorithm

Wind power forecasting plays a critical role in the reliable integration of renewable energy into modern power grids, as the inherent variability of wind generation poses challenges to grid stability, energy scheduling, and reserve management. To address the need for highly accurate and computationally efficient short‐term forecasting models, this study proposes a novel ensemble deep learning framework with optimal feature selection Wind Power Forecasting (WPF‐SAO). The methodology begins with data acquisition from a publicly available wind power data set, followed by Z ‐score normalization and data set balancing. A Self‐Adaptive Osprey (SAO) Optimization algorithm is introduced for optimal feature selection, enhancing exploration capability and convergence speed through adaptive control mechanisms. The selected features are then processed using an ensemble model integrating DenseNet, Backpropagation‐Based Stacked Autoencoder, and Gated Recurrent Unit, with majority voting used for final prediction. Experimental results demonstrate superior performance compared with existing models, achieving a minimum Mean Absolute Error of 0.0096 and a Root Mean Squared Error of 0.2302. The WPF‐SAO framework effectively improves forecasting accuracy while maintaining moderate computational complexity, providing a robust and efficient solution for short‐term wind power forecasting and supporting reliable renewable energy integration into power systems.