Multi-Agent Deep Reinforcement Learning Based Scheduling Approach for Mobile Charging in Internet of Electric Vehicles
Linfeng Liu, Zhuo Huang, Jia Xu
Abstract
Mobile charging stations (MCSs) have become an indispensable complement of fixed charging stations. In the regions where fixed charging stations are sparsely deployed or even absent, the main concern is that how to properly schedule MCSs to charge the electric vehicles with insufficient electricity (EVCs). In this paper, we focus on the scheduling of idle MCSs and pending EVCs. To increase the charging revenue of MCSs and enhance the proportion of successfully charged EVCs, we schedule idle MCSs to proactively track some EVCs with potential charging demand, and schedule pending EVCs to approach some busy MCSs for potential charging opportunities. To this end, a Scheduling Approach based on Multi-Agent Deep Reinforcement Learning (SA-MADRL) is proposed to train the scheduling models for agents (idle MCSs and pending EVCs). In SA-MADRL, the agents obtain the local observations to make the scheduling decisions. Both idle MCSs and pending EVCs can independently make the scheduling decisions, and thus SA-MADRL can realize the fully distributed scheduling and has a good scalability. Extensive simulations and comparisons demonstrate the performance superiority of SA-MADRL, i.e., the charging revenue of MCSs can be significantly increased, and the proportion of successfully charged EVCs can be effectively enhanced.