Physics-informed Generative AI for Power Outage Scenario Synthesis

Power outage scenarios generation and evaluation are the foundation of intelligence restoration strategy development, which further strengthens the self-healing capabilities of the power system. However, few studies have explicitly focused on power outage scenarios, and a complete procedure has yet to be established. Based on the Multi-head Attention for Tabular data-Denoising Diffusion Probabilistic Model (MAT-DDPM), an extension method for power outage scenarios is presented in the paper. First, basic power outage scenarios are generated based on a whole-process power outage simulation scheme. Then, MAT-DDPM is proposed to rapidly generate diverse power outage scenarios, refining the DDPM structure by designing an input module and integrating a multi-head attention mechanism. Finally, an enhancement process for the large language model is designed to evaluate outage scenarios, thereby generating analysis reports and providing feedback. Simulation results confirm that the outage scenarios generated by MAT-DDPM are diverse and satisfy the power system operating constraints with high probability. The evaluation model provides valuable analytical insights into power outage scenarios, helping operators systematically understand critical outage information.