Agent-based Learning Approach to Electric Vehicle Routing Problem with Vehicle-to-Grid Supply and Battery Swapping

作者：A. Narayanan, Prasant Misra, Ankush Ojha, Abhinav Gupta, Supratim Ghosh, A. Vasan · 发表于：COMAD/CODS · 年份：2023 · DOI：10.1145/3570991.3571049 · 被引用次数：13 · 研究领域：Computer Science

Electric vehicles (EV) are well suited for last-mile delivery fleets due to better operational costs and lower emissions. However, capital costs of EVs/captive chargers; lack of public charging; and time to charge could be potential roadblocks. To overcome these roadblocks, EV batteries can be used to increase revenue by vehicle to grid (V2G) sale of electricity; and battery swapping (SWP) to handle charging availability and speed. We consider the problem of vehicle routing for an EV fleet with V2G and SWP. The constraints include loading capacity and delivery time windows; and the objective of minimize the overall cost of delivery. We complement existing approaches with a learning agent (LA) that scales to large problem sizes involving hundreds of customers, discharge stations, and battery swapping points. Using two representative datasets (Solomon and Homberger) and a postal delivery network from the city of Bangalore, we compare LA against a genetic algorithm (GA) and three other baselines. Our experimental evaluation shows that LA is 5.65 times faster than GA, while GA is up to more accurate than LA. The LA is able to scale to problem instances with 400+ nodes while the GA is unable to scale beyond 200 nodes.