ProfileSR-GAN: A GAN Based Super-Resolution Method for Generating High-Resolution Load Profiles

This paper presents a novel two-stage load profile super-resolution (LPSR) framework, ProfileSR-GAN, to upsample the low-resolution profiles (LRLPs) high-resolution (HRLPs). The LPSR problem is formulated as Maximum-a-Posteriori problem. In first-stage, GAN-based model adopted restore high-frequency components from LRLPs. To reflect load-weather dependency, aside LRLPs, weather data added an input model. second-stage, polishing network guided by outline loss and switching novelly introduced remove unrealistic power fluctuations in generated HRLPs improve point-to-point matching accuracy. evaluate realisticness of HRLPs, new set shape evaluation metrics developed. Simulation results show that: i) ProfileSR-GAN outperforms state-of-the-art methods all shape-based can achieve comparable performance with those accuracy, ii) after applying convert LRLPs downstream task, non-intrusive monitoring, be significantly improved. demonstrates that effective mechanism for restoring downsampled time-series sets improves tasks require HR inputs.