The increase in photovoltaic generation, electric vehicle charging, and heat pump demand challenges the operational limits of low-voltage distribution grids. This necessitates effective curtailment methods that can function under sparse observability, noisy measurements, and imperfect grid models. Unlike previous end-to-end reinforcement learning approaches, this work decouples congestion detection and control by combining a random forest violation pre-classifier with an actor-critic controller, assessing its robustness to measurement noise and grid parameter mismatch. The framework is tested on a real low-voltage grid using synthetic future operating scenarios characterized by low observability and controllability. With accurate grid parameters, the controller achieves a 98.9% reduction in total violation magnitude, maintaining nearly unchanged performance under the tested measurement noise settings. Grid model mismatch poses greater challenges, yet the controller still mitigates most violations under the tested mismatch conditions.
Blogger's Review: This study presents a robust reinforcement learning framework that effectively enhances the resilience of low-voltage grids against noise and model mismatches, showcasing the potential of machine learning in power system management.