系统管理学报

系统管理学报 ›› 2026, Vol. 35 ›› Issue (4): 917-927.DOI: 10.3969/j.issn.2097-4558.2026.04.003

• • 上一篇    下一篇

面向复杂环境的可解释性列车运行态势预测及自更新方法

陈浩然1，王悉1，王洪伟1，杨欣2，费振豪3，王建英4   

  1. 1. 北京交通大学 先进轨道交通自主运行全国重点实验室 自动化与智能学院，北京 100044；2. 北京交通大学 系统科学学院，北京 100044；3. 卡斯柯信号有限公司上海市铁路智能调度指挥系统工程研究中心，上海 200072；4. 中国铁道科学研究院集团有限公司通信信号研究所，北京 100081
  • 收稿日期:2025-07-02 修回日期:2025-10-09 出版日期:2026-07-28 发布日期:2026-08-03
  • 基金资助:
    国家自然科学基金资助项目(52472334,U2368204,U2469201);北京市自然科学基金资助项目(L241051);中国国家铁路集团
    有限公司科技研究开发计划资助项目(N2024G040);先进轨道交通自主运行全国重点实验室(北京交通大学)自主研究课题资
    助项目(RAO2025ZZ003);中央高校基本科研业务费(2022JBZX034)

Explainable Train Operation Status Prediction and Self-Updating Method for Complex Environments

Chen Haoran1，Wang Xi1，Wang Hongwei1，Yang Xin2，Fei Zhenhao3，Wang Jianying4#br#   

  1. 1. State Key Laboratory of Advanced Rail Autonomous Operation；School of Automation and Intelligence，Beijing Jiaotong University，Beijing 100044，China；2. School of Systems Science，Beijing Jiaotong University，Beijing 100044，China；3. Shanghai Railway Intelligent Dispatch Command System Engineering Research Center，CASCO Signal Ltd.，Shanghai 200072，China；4. Signal and Communication Research Institute，China Academy of Railway Sciences Co.，Ltd.，Beijing 100081，China
  • Received:2025-07-02 Revised:2025-10-09 Online:2026-07-28 Published:2026-08-03

PDF

摘要： 川藏铁路作为我国高海拔复杂环境下的战略干线铁路,面临恶劣气候、客货混运及异常事件频发等多重挑战,其列车运行态势预测与调度管理亟须突破传统模型的局限。针对现有预测方法依赖静态历史数据、实时动态调整能力不足及误差累积等问题,本文提出一种融合概率分布预测与在线自更新能力的可解释性预测模型。该模型通过构建基于多源异构数据的结构化多维特征矩阵,采用梯度提升树实现数值预测与概率分布预测双输出,并设计4层验证框架:基于 SHAP(SHapleyAdditiveexPlanations)特征贡献度分析,建立数据有效性、因果可解释性、模型稳定性及预测一致性验证机制,筛选具备明确致因的新数据以触发增量更新;通过数据缓冲池的增量式决策树扩展,实现模型参数的精准优化。实验结果表明,在概率分布预测能力与其他先进预测模型相当的情况下,本方法的数值预测性能更优;自更新性能方面,预测均方根误差(RMSE)较静态模型降低27.9%,且注入10%噪声数据时误差波动率低于3.2%。本研究不仅能实现预测结果随自更新逐步精确的目标,还可有效避免无效数据输入对模型精度的影响,为复杂环境下的列车运行调度提供兼具预测精度与可解释性的可靠支持。

关键词: 列车运行态势, 预测模型, 自更新, 概率分布预测, 可解释性

Abstract: As a strategic railway line operating in high-altitude and complex environment, the Sichuan-Xizang Railway faces challenges such as harsh climate conditions, mixed passenger-freight operations, and frequent abnormal events. Existing train operation prediction methods rely heavily on static historical data and lack real-time adaptive capabilities, resulting in accumulated prediction errors. To address these issues, this paper proposes an explainable prediction model that integrates probabilistic distribution prediction with online self-updating capabilities. The model constructs structured multidimensional feature matrices from multi-source heterogeneous data and utilizes gradient boosting trees to generate both numerical forecasting and probabilistic predictions. A four-layer validation framework is designed based on SHAP(SHapley Additive exPlanations) feature contribution analysis to establish validation mechanisms for data effectiveness, causal interpretability, model stability, and prediction consistency. New data with clear causal relationships are selected to trigger incremental updates. Through incremental decision tree-tree expansion using a data buffer pool, model parameters are precisely optimized. The experimental results show that the proposed method achieves superior numerical prediction performance while maintaining comparable probabilistic prediction capability relative to advanced baseline models. Regarding self-updating performance, the root mean square error (RMSE) decreases by 27.9% compared with static models, and the error fluctuation rate remains below 3.2% even with 10% noise injection. This paper not only improves prediction accuracy through continuous self-updating, but also effectively prevents invalid data from degrading the model performance, thereby providing reliable and interpretable support for train operation scheduling in complex environments.

Key words: train operation status, prediction model, self-updating, probabilistic distribution prediction, explainability

中图分类号: