Low-Voltage Grid Control Based On Data-Driven State Estimation and Reinforcement Learning

• Published in: 2024 IEEE Power & Energy Society General Meeting (PESGM) pp. 1 - 5
• Main Authors: Ozlemis, Hakan, Mora, Edwin, Schubert, Jano, Mohapatra, Anurag, Duckheim, Mathias, Niessen, Stefan
• Format: Conference Proceeding
• Language: English
• Published: IEEE 21.07.2024
• Subjects: autonomous distribution grids; data-driven; Distributed power generation; Low voltage; Reactive power; real-time control; Real-time systems; Reinforcement learning; Robustness; State estimation; Substations; Uncertainty; Voltage control
• ISSN: 1944-9933
• DOI: 10.1109/PESGM51994.2024.10689213

As Distributed Energy Resources (DERs) become more prevalent in Low-Voltage (LV) grids, distribution system operators face the challenge of implementing reliable, novel control strategies that combine grid monitoring and automated decision-making. The lack of accurate electrical grid models and adequate automation infrastructure at LV level compounds this challenge. In this context, we propose a two-stage, model-free, data-driven controller for DER-dominated LV grids. The proposed controller uses feedback from real-time measurements at the substation to determine suitable curtailment signals and power factors for downstream, individually dispatchable DERs. The controller integrates a model-free state estimator that feeds an online-trained reinforcement learning agent to minimize voltage band violations and DER interventions. Extensive simulations on a benchmark LV grid show that the proposed controller can reduce voltage violations by 98% compared to two status- quo voltage regulation methods, while curtailing distributed generation by only 5.2% and consumption by less than 2%.