Consensus Multi-Agent Reinforcement Learning for Volt-VAR Control in Power Distribution Networks

Volt-VAR control (VVC) is a critical application in active distribution network management system to reduce losses and improve voltage profile. To remove dependency on inaccurate incomplete models enhance resiliency against communication or controller failure, we propose consensus multi-agent deep reinforcement learning algorithm solve the VVC problem, which determines operation schedules for regulators, on-load tap changers, capacitors. The problem formulated as networked Markov decision process, solved using maximum entropy framework novel communication-efficient strategy. proposed allows individual agents learn group policy local rewards. Numerical studies IEEE test feeders show that our matches performance of single-agent benchmark. In addition, shown be efficient resilient.