Short-term Wind Power Prediction Based on TPE-VMD-Transformer-GRU

To enhance the utilization efficiency and dispatchability of wind power resources, and to improve the economic and low-carbon performance of power systems, accurate forecasting of wind power output is critically important. This study proposes a hybrid short-term wind power forecasting model that integrates Variational Mode Decomposition (VMD), Transformer, and Gated Recurrent Unit (GRU) networks. Additionally, Tree-structured Parzen Estimator (TPE) is used to optimize the model's hyperparameters. In the proposed framework, the wind power series is first decomposed into multiple intrinsic modes using VMD. These modes, along with meteorological features, are then processed through a Transformer module to extract attention-based feature representations. The combined features are subsequently input into a GRU network to capture temporal dependencies and generate forecasts. Extensive experiments conducted on the Shanghai power grid dataset demonstrate that the proposed model outperforms benchmark models across multiple evaluation metrics, exhibiting superior accuracy, robustness, and generalization capability.