Cover Me: Mitigating Multi-Agent System Failure through Reinforcement Learning—A Technology Demonstration

Agent systems are common subjects of investigation in power systems research.Multi-agent systems are robust and can provide guarantees with regards to convergence and solution quality, but cannot show resilience against circumstances not considered at design time.Deep reinforcement learning cannot per se give guarantees, but can adapt to unforeseen situations.The technology demonstration showcases a hybrid of a multi-agent system and a reinforcement-based agent, where the latter safeguards the former against failure.