FTDGT: Federated Temporal Dense Granular Transformer‐Based Wind Power Forecasting in Medium and Long Term

Power forecasting is pivotal for ensuring the stability and efficiency of electrical systems. With the rapid advancements in deep learning techniques, their application to power forecasting has gained considerable momentum. This paper introduces a novel approach, integrating the strengths of federated temporal dense granular transformer (FTDGT) architecture for wind power forecasting tasks. Our primary contribution lies in harnessing these deep learning techniques, coupled with meticulous feature selection, to enhance forecasting precision. This research employed two diverse datasets, each presenting its unique set of challenges, to validate the robustness of the proposed model. Our results are encouraging. They demonstrate the effectiveness of the FTDGT framework and emphasize the benefits of careful feature selection. Compared to current models, the proposed FTDGT model enhanced by feature selection shows potential as a benchmark for performance prediction. FTDGT shows a significant performance improvement over the competitor model, with average improvements in root mean square error (RMSE), mean absolute error (MAE), and R 2 of up to 12%, 15.5%, and 9.2%, respectively. FTDGT’s ability to achieve exceptional results for many client models through feature engineering and different data sets emphasizes its versatility and resilience. This research facilitates future data security efforts in deep learning and energy efficiency by highlighting the interplay between sophisticated neural network topologies and feature selection methods.