A Data-Driven Multi-Agent PHEVs Collaborative Charging Scheme Based on Deep Reinforcement Learning

This paper investigates the collaborative charging problem of plug-in hybrid electric vehicles (PHEVs) in a residential area charging station (CS) under an interaction mechanism. The target is to coordinate the charging strategies of all PHEVs to minimize charging cost of PHEVs on the one hand, and avoid the load peak of CS formed by charging behavior of PHEVs on the other hand. At the same time, the usage habits of PHEV owners, the uncertainty of time-varying electricity price in power market, and operation constraints of PHEVs and CS are also taken into account in this paper. To this end, this paper proposes a data-driven multi-agent PHEVs collaborative charging scheme based on deep reinforcement learning (DRL) method. Firstly, PHEVs are respectively constructed as agents. The interaction among PHEVs in the CS are formulated as a Markov Game. Then, a multi-agent PHEVs charging scheduling algorithm based on a multi-agent deep deterministic policy gradient (MADDPG) method is developed, in which each agent (i.e., PHEV) expects to minimize its cost by choosing the optimal charging strategy during the charging horizon. The proposed method can learn from data mining and gradually grasp the system operation rules by input and output data. Simulations are presented at last to validate the effectiveness of the proposed method.