Optimal Coordination for Multiple Network-Constrained VPPs via Multi-Agent Deep Reinforcement Learning

Abstract

This paper proposes a multi-agent deep reinforcement learning method to coordinate multiple microgrids owned virtual power plants (VPPs) connected in the active distribution network. A communication-based independent twin delayed deep deterministic policy gradient method is proposed to implement a partially observable Markov game for coordinating multiple VPPs (agents). The decentralized training process can protect the privacy of the agent-based entity participating in the framework, and each entity can share equal rights with simultaneous decision-making. Internal markets are proposed in our framework that the price contains the information of energy trading and network information. The price quota curve based external market is proposed for further exploiting distributed energy resource operation potential to enable VPPs with market power to be price-makers. The proposed method has been verified in a modified IEEE test system to prove its effectiveness. The results confirm that the proposed framework can simultaneously improve the operation cost and maintain network operational safety.