Online Multi-Agent Reinforcement Learning for Decentralized Inverter-Based Volt-VAR Control

The distributed Volt/Var control (VVC) methods have been widely studied for active distribution networks(ADNs), which is based on perfect model and real-time P2P communication. However, the always incomplete with significant parameter errors such communication system hard to maintain. In this article, we propose an online multi-agent reinforcement learning decentralized framework (OLDC) VVC. framework, VVC problem formulated as a constrained Markov game novel soft actor-critic (MACSAC) algorithm. MACSAC used train agents online, so accurate ADN no longer needed. Then, trained can realize optimization using local measurements without OLDC has shown extraordinary flexibility, efficiency robustness various computing conditions. Numerical simulations IEEE test cases not only demonstrate that proposed outperforms state-of-art algorithms, but also support superiority of our in application.