
A hybrid agent architecture that combines model-based deep reinforcement learning with imitation learning to overcome the challenges of low sample efficiency and catastrophic forgetting in power grid control applications.


coremsg

hybrid-deep-reinforcement-learning-agent-architecture-combining-model-based-and-imitation-learning-for-resilient-power-grid-control


Hybrid Deep Reinforcement Learning Agent Architecture Combining Model-based and Imitation Learning for Resilient Power Grid Control


title_rewrite


The core message of this article is to propose a robust online voltage control algorithm that can maintain voltage stability in distribution grids even when the exact network topology is unknown. The algorithm combines a nested convex body chasing algorithm to track the set of consistent grid models with a robust predictive controller to adjust reactive power injections accordingly.


robust-online-voltage-control-for-distribution-grids-with-uncertain-topology


Robust Online Voltage Control for Distribution Grids with Uncertain Topology
