基于GBDT的列车晚点时长预测模型研究  

Research on Prediction Model of Train Delay Time Based on GBDT

作　　者：廖璐 张亚东[1] 葛晓程 郭进[1] 禹倩 LIAO Lu;ZHANG Yadong;GE Xiaocheng;GUO Jin;YU Qian(School of Information Science and Technology,Southwest Jiaotong University,Chengdu 611756,China;Institute of Railway Research,University of Huddersfield,UK)

机构地区：[1]西南交通大学信息科学与技术学院,成都611756 [2]哈德斯菲尔德大学铁路研究所,英国

出　　处：《铁道标准设计》2021年第8期149-154,176,共7页Railway Standard Design

基　　金：国家自然科学基金青年基金项目(61703349);中国铁路总公司科技研究开发计划课题(N2018G062,K2018G011)。

摘　　要：准点率是衡量英国铁路服务质量的关键指标之一。列车晚点时长的准确预测,尤其是旅客列车,对调度指挥工作具有重要意义。基于对英国WCML列车实际运行数据分析,提出一种列车晚点时长预测方法:通过时间事件图分析影响晚点的因素,交叉验证法确定模型重要参数,以列车的车站晚点偏差值为自变量建立晚点时长预测的GBDT模型。对测试数据集的预测结果表明,允许误差3min以内,模型预测精度达到99.76%。决策树模型、随机森林模型、基于默认参数的GBDT模型预测结果对比表明,调整参数后的GBDT模型预测精度更高,性能更优。同时基于预测模型的重要度排序,得到影响列车晚点的关键因素为列车前一站的晚点偏差值。Punctuality is one of the key measurements of the quality of rail services in UK.Accurate prediction of the delay time for trains,especially passenger trains,is of great significance for traffic management and dispatching processes.Based on a big set of real data from UK’s West Coast Main Line(WCML),this paper proposes an approach to have a better prediction of train’s delay time:use a Time-event Graph to identify the factors which will affect the delay and use Cross Validation method to determine important parameters of model,take the timetable deviation at station as an independent variable to establish the GBDT model to have a prediction of the delay time.The prediction on the test data set shows that the accuracy of model prediction is 99.76%with a permissible error of 3 minutes.The prediction result comparison of the decision tree model,the random forest model,and the GBDT model based on default parameters show that the adjusted GBDT model has higher prediction accuracy and optimal performance.Based on the sensitivity analysis of the model,the key factor affecting the train delay time is the delay at the previous station.

关 键 词：铁路运输 晚点时长预测 GBDT 时间事件图 列车运行实绩 

分 类 号：U292.4[交通运输工程—交通运输规划与管理]