Leveraging Battery Electric-Bus Charging Networks for Resilient Shared EV Charging via Deep Reinforcement Learning

Abstract

The rapid electrification of urban transportation has increased dependence on public electric-vehicle (EV) charging infrastructure, making it more vulnerable to frequent and severe disruptions. To address this issue, this study proposes utilizing underused battery electric-bus (BEB) charging networks by dynamically reallocating surplus depot chargers for public EV charging. We introduce an adaptive shared-charging coordination framework to increase the resilience of public charging services. This coordination problem is formulated as a Markov decision process (MDP) that jointly optimizes BEB charging schedules and shared charger allocation under uncertainty. To enable real-time decision-making without requiring precise forecasts of future system states, an on-policy deep reinforcement-learning (DRL) approach based on the asynchronous advantage actor-critic (A3C) algorithm is developed. A case study using real-world data from Beijing during a major urban flood demonstrates the effectiveness of the proposed adaptive shared-charging coordination framework. The results reveal that our approach significantly mitigates degradation in public charging service performance, accelerates recovery to normal operating levels, enhances user accessibility, and supports grid stability. Under an extreme scenario with only 25% of public chargers operational, the proposed strategy limits revenue losses to just 3.49%, compared with losses of 53.34% under conventional operations. Additionally, the A3C-based approach demonstrates notable training efficiency and achieves a favorable balance between short-term responsiveness and long-term system performance when benchmarked against a perfect-information optimization model, proximal policy optimization (PPO), and a greedy heuristic. These findings highlight the substantial potential of BEB charging networks as critical resilience resources for urban public EV charging infrastructure during extreme disruption events.