Abstract The past decade has witnessed a significant growth of electric vehicles (EVs) deployment in public and private transportation sectors. Dynamic electric vehicle routing aims to plan the routes of EVs to serve dynamically generated customers’ requests while ensuring the battery level by visiting the recharge stations. This paper studies the dynamic electric vehicle routing problem with simultaneous pick-up and delivery and soft time windows (DEVRP-STW). We design a graph-based spatio-temporal multi-agent reinforcement learning (GMIX) framework consisting of a spatio-temporal graph attention network (ST-GAT) and a value decomposition-based multi-agent reinforcement learning algorithm (Graph-QMIX). Graph-QMIX derives a multi-agent soft policy gradient based on QMIX and Soft Actor–Critic and utilises a graph-based mixing network proposed to enhance the interaction of agents. Extensive experiments including a case study using real-world taxi data from New York City and a simulation study are conducted. The proposed GMIX outperforms the baseline algorithms on the request–response ratio (RRO), the average travelling distance per completed request (ATR), the out of energy ratio (OER) and the average waiting time per completed request (AWR).
Highlights This paper studies the dynamic electric vehicle routing problem with simultaneous pick-up and delivery and soft time windows (DEVRP-STW). We design a graph-based spatio-temporal multi-agent reinforcement learning (GMIX) framework consisting of a spatio-temporal graph attention network (ST-GAT) and a value decomposition-based multi-agent reinforcement learning algorithm (Graph-QMIX). The proposed GMIX outperforms the baseline algorithms on the request–response ratio (RRO), the average travelling distance per completed request (ATR), the out-of-energy ratio (OER) and the average waiting time per completed request (AWR).
GMIX: Graph-based spatial–temporal multi-agent reinforcement learning for dynamic electric vehicle dispatching system
2022-09-12
Article (Journal)
Electronic Resource
English
Multi-agent reinforcement learning for order-dispatching via order-vehicle distribution matching
European Patent Office | 2023
|MULTI-AGENT REINFORCEMENT LEARNING FOR ORDER-DISPATCHING VIA ORDER-VEHICLE DISTRIBUTION MATCHING
European Patent Office | 2020
|MULTI-AGENT REINFORCEMENT LEARNING FOR ORDER-DISPATCHING VIA ORDER-VEHICLE DISTRIBUTION MATCHING
European Patent Office | 2020
|Dynamic Dispatching Control Model Based on Multi-Agent System
British Library Conference Proceedings | 2009
|