A short-term wind power hybrid forecasting Research based on two-step decomposition and BBO optimized FNETformer-OSRELM deep Learning

Accurate short-term wind power forecasting is essential for the reliable operation of modern power systems. We propose a hybrid forecasting framework that combines multi-scale decomposition, frequency-domain enhancement, and intelligent optimization. First, a two-step decomposition method combining ICEEMDAN and VMD is applied to extract multi-scale frequency features and alleviate mode mixing. Then, the Pearson correlation coefficient is employed to select features strongly correlated with wind power. Subsequently, a frequency-domain enhanced mechanism (FNet module) replaces the attention mechanism in iTransformer, forming the improved FNETformer model that better captures periodic characteristics. The lightweight OS-RELM model is applied to predict high-frequency components, while FNETformer handles medium- and low-frequency components. Furthermore, BBO algorithm is used to adaptively optimize the hyperparameters of the hybrid model. Taking May as a representative example, comparative experiments from five perspectives show that the proposed model achieves a reduction in RMSE by 4.61%–35.27%, a decrease in MAE by 6.26%–35.07%, and an improvement in R 2 by 0.43%–6.01% compared with other benchmark models. Overall, the results demonstrate that the proposed method effectively enhances the model’s adaptability, robustness, and forecasting accuracy.