Optimal Scheduling for EV Charging Stations with Varying Power Limits Based on MADDPG Algorithm

Charging behaviors of electric vehicles (EVs) are subject to a high degree of randomness, which are difficult in accurate forecast. With the growing of the number of EVs, real-time scheduling for the charging station becomes more important. By collecting real-time information of plugged-in EVs and using the Multi-agent Deep Deterministic Policy Gradient (MADDPG) algorithm, a decentralized charging scheduling method for charging stations is proposed. Firstly, the model of the system and the energy boundary (EB) are formulated for decentralized collaborative scheduling of charging piles. Secondly, a multi-agent environment suitable for MADDPG is constructed, so that the charging pile can make the real-time charging/discharging strategies based on local information. Finally, case studies illustrate that the proposed decentralized charging scheduling method can decide scheduling strategies in real time without relying on forecast information of EVs, effectively reducing charging costs and responding to changes in operating scenarios.