Adaptive automatic driving (AAD) for metro trains is currently hindered by the limited diversity of real-world operational data and the high costs associated with its acquisition. To address these challenges, this study proposes a comprehensive four-stage framework that integrates expert operational knowledge, thereby facilitating scalable AAD deployment with minimal dependency on empirical datasets. The methodology initially leverages expert-defined motion constraints, such as acceleration profiles and speed thresholds, to synthesize 2.9 million high-fidelity driving curves that capture complex speed-distance dynamics across diverse track conditions. These trajectories are then processed through a fuzzy multi-criteria decision-making system informed by domain expertise to select optimal curves that balance energy efficiency, passenger comfort, and stopping precision. Subsequently, the selected curves are transformed into structured parametric representations to train a random forest model for precise trajectory recovery and reconstruction. Empirical validation across various operational scenarios (with typical travel durations of 170-238 s) demonstrates that the framework achieves a 12% reduction in energy consumption and 95% passenger satisfaction, while aligning with stopping and arrival time targets of Formula: see text m and Formula: see text s, respectively. Furthermore, the model yields a 98.2% reconstruction accuracy with a 76% reduction in data costs compared to traditional sampling methods. By effectively bridging the gap between expert experience and data-driven learning, this framework addresses real-world adaptability gaps, enabling robust AAD deployment in low-infrastructure environments while remaining fully compatible with existing automatic train operation (ATO) systems.
Huang et al. (Wed,) studied this question.