Reinforcement learning based mobile energy disseminator dispatching for on-road electric vehicle charging

The rapid growth of electric vehicles (EVs) exacerbates the existing challenges of vehicle range anxiety and charging infrastructure inadequacy. To address these concerns, promising solutions have emerged, including vehicle-to-vehicle dynamic wireless charging (V2V-DWC) technologies, particularly Mobile Energy Disseminators (MEDs). A MED is a large vehicle bearing non-contact technology equipment that charges participating EVs in its proximity. This paper establishes a platform for reliable simulation of the energy transfer between the MED and the EV, develops models that describe EV battery consumption and V2V charging dynamics, and develops an AI-based framework for MED dispatching. The first component of the developed platform is a realistic reinforcement learning (RL) based highway environment termed ``ChargingEnv.'' This component incorporates reliable charging simulation that accounts for two common practical issues in wireless charging deployment: charging panel misalignment and realistic EV battery consumption patterns. The second component involves multiple deep RL benchmark models trained in ``ChargingEnv'' to maximize the MED deployment charging service quality within charging resource constraints. Findings from numerical experiments demonstrate that the model trained in the framework substantially alleviates battery depletion concerns. The framework can serve as a tool for road agencies or private-sector investors to efficiently plan their MED deployments.