Assessing data-driven state estimation performance in low-voltage distribution grids

Monitoring Low-Voltage (LV) distribution grids is becoming increasingly important as Distributed Energy Resources (DERs) increase grid utilisation and therefore the frequency of critical states. To monitor these grid states in real-time, the application of State Estimation (SE) techniques is required to overcome limited real-time measurement infrastructure. Furthermore, electrical LV grid models are often inaccurate or unreliable. Data-driven SE approaches can address these challenges by solely using historical smart meter measurement data, i.e., without requiring an accurate grid model. In this paper, we systematically assess the performance of data-driven SE applied to the estimation of voltage magnitudes at smart meter nodes. For cases showing low accuracy, we also propose a data-driven optimisation heuristic for the systematic placement of additional real-time measurement devices. Simulation experiments on benchmark LV grids with different topologies, measurement configurations, and DER shares show that data-driven SE can achieve a maximum estimation error of less than 0.014 p.u. for all examined scenarios.