Vision Transformer-Based Multi-Scale Wind Field Super-Resolution Reconstruction Model

Accurate wind farm site selection and wind energy resource prediction require high-quality wind field data. However, existing satellite and weather station data are limited by their low resolution, posing significant challenges to the decision-making accuracy. To address these limitations, this paper proposes WFSR-MsFormer, a novel multi-scale vision Transformer architecture specifically designed for the wind field super-resolution task is proposed. In the proposed framework, wind field data are first gridded and expanded to yield multiscale data through downsampling and upsampling. Second, three parallel ViTs are applied to process the data of each scale respectively, where the self-attention mechanism is used to obtain long-range dependencies. Finally, upsampling is performed through interpolation and the channels are merged to generate the super-resolution wind field prediction results. Comparative experiments based on real wind power data demonstrate that the proposed WFSR-MsFormer exhibits excellent accuracy for wind field data reconstruction.