中国安全科学学报 ›› 2019, Vol. 29 ›› Issue (S1): 20-26.doi: 10.16265/j.cnki.issn1003-3033.2019.S1.005

• 安全系统学 • 上一篇    下一篇

高速铁路列车晚点时间实时预测的神经网络模型*

黄平1,2,3, 文超**1,2,3 副教授, 李忠灿1,2, 杨宇翔1,2, 彭其渊1,2 教授   

  1. 1 西南交通大学 综合交通大数据应用技术国家工程实验室,四川 成都 610031;
    2 西南交通大学 综合交通运输国家地方联合工程实验室,四川 成都 610031;
    3 滑铁卢大学 铁路研究中心,加拿大 滑铁卢 N2L3G1
  • 收稿日期:2019-03-03 修回日期:2019-05-25 出版日期:2019-06-30 发布日期:2020-10-28
  • 通讯作者: ** 文 超(1984—),男,江西宜春人,博士,副教授,硕士生导师,主要从事铁路运输组织优化、交通大数据应用等方面的研究。E-mail:wenchao@swjtu.cn。
  • 作者简介:黄 平 (1990—),男,四川宜宾人,博士研究生,研究方向为铁路运输组织优化、数据挖掘、机器学习等。E-mail:huangping129@my.swjtu.edu.cn。
  • 基金资助:
    国家自然科学基金资助(71871188);国家重点研发计划(2017YFB1200701);西南交通大学博士研究生创新基金资助(D-CX201827)。

A neural network model for real-time prediction of high-speed railway delays

HUANG Ping1,2,3, WEN Chao1,2,3, LI Zhongcan1,2, YANG Yuxiang1,2, PENG Qiyuan1,2   

  1. 1 National Engineering Laboratory of Integrated Transportation Big Data Application Technology, Southwest Jiaotong University, Chengdu Sichuan 610031, China;
    2 National United Engineering Laboratory of Integrated and Intelligent Transportation, Southwest Jiaotong University, Chengdu Sichuan 610031, China;
    3 Railway Research Centre, University of Waterloo, Waterloo N2L3G1, Canada
  • Received:2019-03-03 Revised:2019-05-25 Online:2019-06-30 Published:2020-10-28

摘要: 准确地预测高速列车晚点时间对提高高速铁路实时调度指挥水平及运输服务质量有重要意义。以武汉-广州高速铁路(HSR)列车运行实绩数据为基础,建立基于循环神经网络(RNN)的列车晚点预测模型。该模型中,按照列车实际运行顺序输入RNN以利用其反馈机制学习到相邻列车间相互作用关系。基于平均绝对误差(MAE)以及平均绝对百分误差(MAPE)评估模型的预测能力。结果表明:提出的深度学习模型预测精度明显高于人工神经网络、支持向量回归及马尔科夫等已有列车晚点时间预测模型。

关键词: 高速铁路(HSR), 列车运行实绩, 晚点预测, 循环神经网络(RNN), 列车相互作用

Abstract: Accurately forecasting train delays has great significance for improving the real-time dispatching ability and the quality of transport service. By using the train operation records of Wuhan-Guangzhou HSR, a deep learning model based on RNN was established to predict train delays. In this model, the trains were fed into RNN according to their operating orders to use the feed-back mechanism of RNN to capture train interaction. The model performance was evaluated based on mean absolute error (MAE) and mean absolute percentage error (MAPE) metrics. The results demonstrate that deep learning model proposed in this paper distinctly outperform other models widely used delay prediction, including Artificial Neural Network, Support Vector Regression, and Markov Model.

Key words: high-speed railway (HSR), train operation records, delay prediction, recurrent neural network (RNN), train interaction

中图分类号: