基于贝叶斯网络的多制式轨道交通列车运行冲突检测研究  

Train Operation Conflict Detection in Multi-Standard Rail Transit Based on Bayesian Networks

作　　者：钟兴莉 黄平 彭其渊[1,2] 温傈文[1,2] ZHONG Xingli;HUANG Ping;PENG Qiyuan;WEN Liwen(School of Transportation and Logistics,Southwest Jiaotong University,Chengdu 611756,Sichuan,China;National United Engineering Laboratory of Integrated and Intelligent Transportation,Southwest Jiaotong University,Chengdu 611756,Sichuan,China)

机构地区：[1]西南交通大学交通运输与物流学院,四川成都611756 [2]西南交通大学综合交通运输智能化国家地方联合工程实验室,四川成都611756

出　　处：《铁道运输与经济》2025年第4期128-140,共13页Railway Transport and Economy

基　　金：国家重点研发计划项目(2022YFB4300502);国家自然科学基金项目(72301221)。

摘　　要：为提升铁路调度决策水平,基于领域知识及数据驱动技术,结合预测和推断方法进行列车运行冲突检测研究。从单列车、相邻列车以及多列车层面分析冲突演化过程,考虑多制式轨道交通列车共线运行的场景,结合冲突判定规则,提出冲突检测贝叶斯网络基本模型结构;在此基础上,基于列车运行数据,通过结构学习和参数学习对模型进行结构和参数优化;最终采用准确率、精确率、召回率和F1分数等指标对检测模型进行评估,得到各线路冲突检测平均准确率达81%,平均召回率达86%,F1分数达83%。与常用的冲突检测模型对比表明:本研究构建的贝叶斯网络模型能够较为准确地检测冲突,且误判率低。模型主要优势在于可以通过贝叶斯网络结构解释各变量之间的因果关系;此外,通过该模型所预测的列车晚点推演数据判定冲突发生地点可有效提升冲突检测模型效果,为列车运行冲突消解奠定基础。To improve the decision-making level of railway dispatching,train operation conflict(TOC)detection was studied based on domain knowledge and data-driven technologies through prediction and inference.The conflict evolution process was analyzed from the level of single train,adjacent trains,and multiple trains.Furthermore,by considering the scenario of multi-standard rail transit trains running on the same line,the basic model structure of Bayesian Networks(BNs)for TOC detection was proposed according to the conflict determination rules.Based on the train operation data,the structure and parameters of the model were optimized by structure learning and parameter learning.Evaluation indicators such as accuracy,precision,recall,and F1 score were used to assess the detection model.The results show that the average detection accuracy for TOC on each line is 81%;the average recall is 86%,and the average F1 score is 83%.A comparison with commonly used TOC detection models shows that the proposed BN model demonstrates higher accuracy and lower misclassification rates in TOC detection.The main advantage of the model is that the causal relationship between the variables can be explained through the BN structure.In addition,determining the location of the conflict by the train delays predicted by the model enhances the performance of the model,which lays a decision-making basis for TOC mitigation.

关 键 词：铁路运输 列车运行冲突检测 贝叶斯网络 多制式轨道交通 领域知识 数据驱动 

分 类 号：U298.1[交通运输工程—交通运输规划与管理]