Adaptive hybrid deep learning framework for medium-term wind speed forecasting

Accurate wind speed forecasting is essential for stable wind farm operation but is challenged by nonlinear, stochastic, and multi-scale spatiotemporal dynamics. An adaptive hybrid forecasting framework is proposed that integrates complementary feature extraction with sequence modeling. Evaluations across the Pacific Grove weather station and the VALHALL_A dataset show that the model achieves coefficients of determination above 0.97 and reduces root-mean-square error by over 20% relative to benchmark models while maintaining low maximum error. These results indicate that the integration of these components facilitates the effective modeling of both short-term variations and long-term dependencies in complex wind dynamics. The framework incorporates transparency mechanisms that enable interpretability and, together with its predictive performance, provide accurate, reliable, and scalable wind forecasting for renewable energy integration.