<p>Accurate wind power forecasting is essential for enhancing the integration of renewable energy sources, thereby supporting global decarbonization initiatives. However, the inherent stochastic nature of wind resources significantly complicates short-to-medium-term forecasting, introducing operational uncertainties within power systems. Despite substantial improvements in existing forecasting techniques, conventional models often fail to achieve consistently high accuracy, necessitating methodological advancements. To address this limitation, we introduce a novel multi-scale forecasting framework integrating fuzzy information granulation and a multi-objective optimization strategy. The fuzzy information granulation technique effectively captures intrinsic features from highly volatile wind speed data, significantly reducing the data complexity and mitigating noise interference for deep learning models. Moreover, our combined model leverages multiple neural networks employing diverse predictive principles, adaptively integrating their outputs via heuristic optimization algorithms. This approach simultaneously enhances prediction accuracy and robustness. Experimental validation using the Penglai wind farm dataset highlights the outstanding performance of our proposed framework. Importantly, the fuzzy information granulation-based collaborative optimization algorithm effectively resolves the critical trade-off between prediction accuracy and computational efficiency in wind speed forecasting systems.</p>