A Platform for Deploying Multi-agent Deep Reinforcement Learning in Microgrid Distributed Control

Abstract

Distributed control strategies have been attracted significant attention due to numerous advantages over traditional centralized control strategies. The development of deep reinforcement learning method provides a novel approach to control grid without knowing the system's parameters. The training and validating process with grid simulation as environment have been supported by several toolboxes. In this paper, a platform based on redis NoSQL database is proposed to the deploy the multi-agent system of deep reinforcement learning algorithms for control microgrid in a distributed manner. The accuracy of agent implementation under realistic condition with physical communication network can be evaluated with the proposed platform. The distributed control in islanded DC microgrid using Deep Deterministic Policy Gradient is introduced as an use case to show the operation of the platform.