Generative Adversarial Networks with Physics-Informed Losses for High-Resolution Load Profile Generation and Inpainting

Accurate load profile data is essential for optimizing energy systems, yet real-world datasets often suffer from low-resolution and missing data. This paper introduces a novel framework that combines Generative Adversarial Networks (GANs) with Physics-Informed Loss (PIL) derived from Physics-Informed Neural Networks (PINNs), which are integrated directly into the generator to address these challenges. This hybrid approach generates high-resolution synthetic load profiles that not only fill missing data but also ensure that the generated profiles adhere to physical laws governing energy systems, such as energy conservation and load fluctuations. By embedding domain-specific physics into the generative process, the model significantly enhances data quality and resolution for low-quality datasets. Experimental results demonstrate notable gains in data accuracy, resolution, and consistency, making this framework an effective tool for energy management systems. The methodology also has broader applicability in other fields, such as inpainting and generating high-resolution data sets in energy systems, industrial processes, or any domain where data must comply with real-world physical constraints or operational requirements.