Real-Time Charging and Routing for Electric Long-Haul Trucks via Reinforcement Learning

Le résumé fourni par la source

The electrification of medium and heavy-duty vehicles (MHDVs), particularly in long-haul freight transport, is essential for achieving climate and public health goals but faces significant operational challenges, including limited range, sparse charging infrastructure, and high recharging times. This paper introduces a novel reinforcement learning (RL) framework designed to optimize routing and charging decisions for electric long-haul trucks in realistic, stochastic environments. Leveraging a Proximal Policy Optimization (PPO) agent, our approach integrates Time-of-Use (TOU) electricity pricing, station-specific queueing dynamics, and real-world truck traffic data within a graph-based representation of the national highway network. Charging stations are selected based on proximity to high-voltage transmission lines and key freight corridors, ensuring grid compatibility and route feasibility. The agent receives dynamic observations including state-of-charge (SoC), station pricing, estimated wait times, and energy requirements, and learns to minimize operational costs through adaptive decision-making. Simulation results along a Los Angeles-San Antonio corridor demonstrate that the PPO agent substantially reduces energy costs, travel time, and charging stops while achieving a 100% success rate in route completion. This work underscores the potential of deep reinforcement learning as a scalable and adaptive solution for electrified route logistics and lays the groundwork for future extensions incorporating regulatory constraints, bidirectional energy flows, and environmental factors.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.