Short-Term Wind Power Forecasting Based on Dual-Optimized VMD-CNN-BiLSTM

To tackle issues such as high data volatility, temporal dependencies, complex feature extraction, and low parameter tuning efficiency in wind power forecasting, this paper proposes a dual-optimization model for short-term wind power forecasting based on RIME-VMD and MSSA-CNN-BiLSTM. First, the Rime Optimization Algorithm (RIME) is employed to adaptively refine the key parameters of Variational Mode Decomposition (VMD), decomposing wind power into intrinsic modal functions (IMFs) of different frequencies to reduce signal complexity. Second, by integrating the local feature extraction capabilities of Convolutional Neural Network (CNN) with the bidirectional temporal dependency capture capabilities of Bidirectional Long Short-Term Memory Network (BiLSTM), a hybrid deep learning architecture is constructed. Additionally, the Multi-strategy Sparrow Search Algorithm (MSSA) is introduced to perform global hyperparameter optimization, thereby addressing the shortcomings of manual parameter tuning. The final power forecast is obtained through the prediction of each IMF component and the reconstruction of the results. Experiments demonstrate that the presented prediction model attains a root mean square error (RMSE) of 0.0333, a mean absolute error (MAE) of 0.0265, and a coefficient of determination (R2) of 0.9901. Seasonal validation shows that the model’s R2 exceeds 0.983 in all four seasons—spring, summer, autumn, and winter—demonstrating good generalization capability. Relative to the BiLSTM model, its RMSE and MAE are reduced by 50.52% and 46.57%, respectively, while R2 increases by 3.36%, effectively addressing the issue of insufficient accuracy in short-term wind power forecasting.