• First study integrating SAV pickup-delivery with DWC under time windows. • Proposes STH-DRL with spatio-temporal attention and adaptive masking. • Achieves superior cost, on-time ratio, and SOC stability across cities. • Validates that 25-29 SAVs can replace 115 taxis in 24-h continuous operation. • Provides quantitative insights for DWC deployment and battery capacity optimisation. Dynamic wireless charging (DWC) technology is rapidly emerging as a promising solution to alleviate range anxiety in electric shared autonomous vehicles (SAVs). However, SAVs’ service quality depends not only on efficient charging but also on adherence to service time windows. Existing routing strategies for static charging typically treat charging and time-window scheduling as independent processes, failing to capture their spatiotemporal interdependence under dynamic charging scenarios. To address the SAV’s Pick-up, Delivery and Dynamic Charging Problem with Time Windows (SPDCP-TW), this study proposes a Spatio-Temporal Heterogeneous Deep Reinforcement Learning (STH-DRL) strategy that minimises route costs and stabilises battery state-of-charge levels while ensuring on-time service. Specifically, we introduce a Spatio-Temporal Heterogeneous Attention mechanism to explicitly encode time-window features and an adaptive multi-constraint masking mechanism to distinguish between early, on-time, and late arrival scenarios. A comprehensive simulation platform, incorporating real-world road networks, traffic flows, and taxis’ travel data, has been developed to evaluate multiple time-window-related metrics. Experimental results across three cities and multiple scales show that, compared with heuristic and state-of-the-art DRL methods, our proposed strategy consistently achieves superior performance in terms of total cost, on-time ratio, and battery state stability. Experiments across three cities and multiple operational scales demonstrate that our approach consistently outperforms heuristic and state-of-the-art DRL methods in terms of total cost, on-time ratio, and battery charging state stability. In 24-h continuous operation tests, the proposed strategy maintains on-time ratios of 90-99% while optimising charging schedules. Furthermore, analysis using real taxi datasets reveals that 9–29 continuously operating SAVs can replace 115 conventional taxis. Finally, sensitivity analysis provides practical insights for DWC deployment and battery capacity optimisation, showing that with 5.5% DWC road coverage, a 25-50 kW power range achieves a balanced trade-off, halving battery capacity, and significantly reducing route costs.
Wang et al. (Wed,) studied this question.