Research on wind power forecasting using a VMD-DE optimized bias-corrected transformer-GRU model

Abstract

To address the issues of systematic bias and non-stationarity in wind power forecasting, this paper proposes a bias-corrected Transformer-GRU hybrid forecasting model optimized by Variational Mode Decomposition (VMD) and the Differential Evolution (DE) algorithm. First, VMD is employed to adaptively decompose the original power sequence, extracting multi-scale frequency-domain features to reduce the non-stationarity of the sequence. Then, a hybrid network that integrates the self-attention mechanism of Transformer with the temporal modeling capability of GRU is constructed, and a bias correction module is introduced to achieve dynamic compensation of prediction errors. Based on experimental data from an actual wind farm, the prediction results indicate that the forecasting model outperforms traditional models in terms of mean squared error (MSE), mean absolute error (MAE), root mean squared error (RMSE), and the coefficient of determination (R²), with the R² exceeding 0.9815, thereby validating the effectiveness and superiority of the model in short-term wind power forecasting.