Physics-Conditioned Generative Adversarial Networks for State Estimation in Active Power Distribution Systems with Low Observability

A novel method is proposed to address the issue of low-observability in Distribution System State Estimation (DSSE). We first use the historical data at the unobservable locations to construct and train proper Generative Adversarial Network (GAN) models to compensate for lack of direct real-time measurements. We then integrate the trained GAN models, together with the direct synchronized measurements at the observable locations, into the formulation of the DSSE problem. In this regard, we simultaneously take advantage of the forecasting capabilities of the GAN models, the available real-time synchronized measurements, and the DSSE formulations based on physical laws in the power system. As a result, on one hand we conduct a physics-conditioned estimation of the unknown power injections at the unobservable locations; and on the other hand, we also achieve a complete DSSE solution for the understudy low-observable active power distribution system.