The surge in popularity of electric vehicles (EVs) has created a need for adaptable and flexible charging infrastructure. Intelligent Charging Robots (ICRs) have emerged as a promising solution to overcome issues faced by fixed charging stations, such as insufficient coverage, station occupancy, spatial constraints, and strain on the power grid. Nonetheless, optimizing the operational efficiency of ICRs presents a significant challenge. This study focuses on optimizing the scheduling of ICRs in a public parking facility through Deep Reinforcement Learning (DRL) methods. We first introduce the Intelligent Charging Robots Scheduling Problem (ICRSP) that maximizes either the number of EVs served (MN) or the total output electricity of ICRs (ME), and establish the corresponding mathematical model. Then, a DRL framework based on the Transformer structure is proposed to tackle ICRSP by integrating decisions of ICR assignment and EV sequencing to enhance solution quality. Furthermore, we devise a masking mechanism in the decoder to manage ICRs’ self-charging behavior during the charging service. Finally, experimental results validate the effectiveness of the proposed DRL approach in providing efficient scheduling solutions for large-scale ICRSP instances. The comparative analysis of MN-ICRSP and ME-ICRSP models offers valuable insights for ICRs operation scheduling, aiding in balancing operator revenue and customer satisfaction.

中文翻译：

---

为电动汽车调度智能充电机器人：一种深度强化学习方法

电动汽车 （EV） 的普及催生了对适应性强且灵活的充电基础设施的需求。智能充电机器人 （ICR） 已成为一种很有前途的解决方案，可以克服固定充电站面临的问题，例如覆盖范围不足、电站占用率、空间限制和电网压力。尽管如此，优化 ICR 的运营效率是一项重大挑战。本研究的重点是通过深度强化学习 （DRL） 方法优化公共停车设施中 ICR 的调度。我们首先介绍了智能充电机器人调度问题 （ICRSP），该问题使服务的电动汽车数量 （MN） 或 ICR 的总输出电力 （ME） 最大化，并建立了相应的数学模型。然后，提出了一种基于 Transformer 结构的 DRL 框架，通过整合 ICR 分配和 EV 排序决策来解决 ICRSP 问题，以提高求解质量。此外，我们在解码器中设计了一种屏蔽机制，以管理 ICR 在充电服务期间的自充电行为。最后，实验结果验证了所提出的 DRL 方法在为大规模 ICRSP 实例提供高效调度解决方案方面的有效性。MN-ICRSP 和 ME-ICRSP 模型的比较分析为 ICR 的运营调度提供了有价值的见解，有助于平衡运营商收入和客户满意度。