Optimal Energy Management of Energy Internet: A Distributed Actor-Critic Reinforcement Learning Method

Abstract

Owning to the capacity constraints and the uneven distribution of resources, energy management problem in energy internet is a major concern. To cope with the variations and complexity of large scale energy management, a distributed actor-critic reinforcement learning based method is proposed for optimal energy management. First, the intelligent action is decided in the distributed agent to alleviate the pressure on centralized intelligent computing. The distributed action of each agent is based on its neighbour information, and an actor-critic reinforcement learning algorithm is applied for dealing with the continuous action space. Then, aiming at the supply-demand balance, the action is adjusted based on global information exchange. After action adjustment, the corresponding rewards are sent to each agent. Finally, the modified action is executed in each agent under the condition of the supply-demand balance. And received rewards are utilized to update each agent. Simulation driven by Pecan Street Inc.s Dataport demonstrates that the proposed intelligent distributed method is effective.