交通运输系统工程与信息 ›› 2025, Vol. 25 ›› Issue (1): 173-187.DOI: 10.16097/j.cnki.1009-6744.2025.01.017

• 系统工程理论与方法 • 上一篇    下一篇

延误场景下列车速度曲线与动态调度联合优化方法

林俊亭*,李茂林,邱晓辉   

  1. 兰州交通大学,自动化与电气工程学院,兰州730070
  • 收稿日期:2024-07-17 修回日期:2024-12-07 接受日期:2024-12-11 出版日期:2025-02-25 发布日期:2025-02-24
  • 作者简介:林俊亭(1978—),男,河北海兴人,教授,博士。
  • 基金资助:
    国家自然科学基金(52162050)。

AJoint Optimization Method of Train Speed Curves and Dynamic Scheduling Under Delay Scenarios

LIN Junting*, LI Maolin, QIU Xiaohui   

  1. School of Automation and Electrical Engineering, Lanzhou Jiaotong University, Lanzhou 730070, China
  • Received:2024-07-17 Revised:2024-12-07 Accepted:2024-12-11 Online:2025-02-25 Published:2025-02-24
  • Supported by:
    National Natural Science Foundation of China (52162050)。

摘要: 为使发生延误的高速列车能够快速恢复正常运营,同时满足停车精度、准时性、节能性及调度实时性等多方面的要求,综合考虑一体化模型在平衡多个目标时面临的多重非线性约束问题,以及非一体化模型需分别求解多个独立模型的局限性,本文提出一种列车动态调度与速度曲线的联合优化方法。首先,基于参考系统的约束,应用集成内在好奇心模块和优先经验回放机制的双决斗深度强化学习算法(Intrinsic Curiosity Module Prioritized Experience Replay Dueling Double Deep Q-Network, ICM-PER-D3QN)优化列车速度曲线模型,保证列车的停车精度、准时性和节能性,并将此数据用作联合模型训练的基础;其次,采用ICM-PER-D3QN算法求解列车的动态调度模型,缓解列车延误并确保调度的实时性;最后,基于列车在站间区间的运行信息,使用集成长短期记忆网络的卷积神经网络完成列车速度曲线与动态调度的联合。实验环境选择京沪高铁的一段下行线路,设置3组延误场景验证所提方法的有效性。仿真结果表明,在联合优化模型下,列车的平均调度时长为0.92s,列车动态调度结果与速度曲线的平均匹配度为98.89%,平均匹配时长为0.0014s。此外,相较于仅基于动态调度模型的未优化速度曲线,平均牵引能耗降低了9%,平均总延误时间降低了6.38%。

关键词: 铁路运输, 联合方法, 深度学习, 速度曲线, 动态调度, 强化学习

Abstract: To enable delayed high-speed trains to quickly resume normal operation while satisfying requirements such as the exactness of train stop, punctuality, energy efficiency, and real-time scheduling, this paper proposes a joint optimization method for dynamic scheduling and speed curves, aiming to addressing the complexities of the integrated models in balancing nonlinear constraints and the limitations of solving multiple independent models separately in non-integrated models. First, based on the constraints of the reference system, the Intrinsic Curiosity Module Prioritized Experience Replay Dueling Double Deep Q-Network (ICM-PER-D3QN) is applied to optimize the train speed curve model, ensuring the exactness of train stop, punctuality, and energy efficiency. This data is used as the foundation for joint model training. Second, the ICM-PER-D3QN algorithm is utilized to solve the dynamic scheduling model of trains, alleviating delays and ensuring the real-time performance of scheduling. Utilizing train operational data in sections, a convolutional neural network integrated with a long short-term memory network is used to jointly optimize train speed curves and dynamic scheduling. The experimental environment is based on downward line of the Beijing Shanghai High-Speed Railway, with three delay scenarios designed. Simulation results show that, under the joint optimization model, the average scheduling time of trains is 0.92 s, the average matching accuracy between dynamic scheduling results and speed curves reaches 98.89%, and the average matching time is 0.0014 s. In addition, compared to the unoptimized speed curves based solely on the dynamic scheduling model, the average traction energy loss is reduced by 9%, and the average total delay time is reduced by 6.38%.

Key words: railway transportation, joint method, deep learning, speed curve, dynamic scheduling, reinforcement learning

中图分类号: