Short-term wind power prediction with a new PCC-GWO-VMD and BiGRU hybrid model enhanced by attention mechanism

The integration of wind power into the grid significantly depends on the accuracy and reliability of wind power prediction. However, the task of wind power prediction faces significant challenges due to the stochastic nature of wind speed. This study proposes a novel deep hybrid short-term wind power prediction model: PCC-GWO-VMD-BiGRU-Attention. The model integrates Pearson correlation coefficient (PCC), variational mode decomposition (VMD), gray wolf optimization (GWO), and attention mechanism optimized bidirectional gated recurrent unit (BiGRU-Attention) to enhance prediction accuracy and robustness, while quantifying prediction uncertainty through deep ensemble methods. First, PCC is used to identify the key factors affecting wind power, thereby improving the model's computational performance. Second, GWO can dynamically adjust the key parameters [K, α] in VMD for optimal decomposition of the input time series, thereby improving prediction accuracy. Finally, the BiGRU-Attention model is utilized to extract global temporal features in historical sequences, while the attention mechanism focuses on key information in the sequences to further enhance prediction performance. Experimental results show that, compared to other deep learning models, this model achieves the highest accuracy in short-term wind power prediction with an root mean square error of 0.2677 Megawatt (MW), mean absolute error of 0.1509 Megawatt (MW), mean square error of 0.0717 Megawatt (MW), and coefficient of determination (R2) of 0.9605. This method has significant practical value and contributes to ensuring safe operation of wind farms.