A Deep Reinforcement Learning Approach to Battery Optimal Management for Urban EV Charging and Swapping Network with Uncertain Demand

作者：Zixiang Zhu, Zhihua Chen, Jing Zhang, Yiwei Su · 年份：2025 · DOI：10.1109/ciotsc67482.2025.11413238 · 研究领域：Electric Vehicles and Infrastructure、Advanced Battery Technologies Research、Smart Grid Energy Management

The remarkable increasing production and use of electric vehicles (EVs) has caused operation management issue for urban EV charging and swapping network. The current battery swapping stations are facing daily uncertain and high-load refueling demands with uneven spatiotemporal distribution. Therefore, this paper constructs a dynamic decision-making model based on Markov Decision Process (MDP) for the battery swapping mode of centralized charging. The Double Deep Q-learning (DDQN) algorithm is exploited to learning the optimal charging schemes with considering fluctuations of electricity bill, battery quantity and random swapping demand. The decision-making process is simplified through a greedy algorithm to reduce complexity and improve efficiency. The effectiveness of the model and algorithm are verified through simulation. The proposed approach outperforms the traditional method by about 11% in terms of total attenuation revenue.