A CSWF-Attention U-Net-Based Super-Resolution Method for Power Quality Data

This study introduces a deep learning-driven approach for super-resolution reconstruction of low-quality power grid data using an enhanced U-Net model integrated with a Channel-Spatial Weighted Fusion (CSWF) module and attention gate. By refining feature extraction and adaptive weighting mechanisms, the model addresses the limitations of traditional methods in capturing disturbance features like voltage sags and harmonics. Experiments using real power grid datasets demonstrate improved performance across PSNR, SSIM, and MSE metrics compared to classical interpolation and convolutional models.This approach provides a scalable, software-level enhancement pathway for improving power quality monitoring without upgrading existing hardware infrastructure.