Scholarlink

<p>This study solves the train timetable rescheduling (TTR) problem from a brand-new perspective. We assume that train traffic controllers take three main actions, i.e., adjusting dwelling times, running times, and train orders, to reschedule the timetable in real-time dispatching. To raise the interpretability of rescheduling models, we propose a graph neural network&#xa0;(GNN) approach to map the train timetable data into evolution graphs that fit the paradigm of train operation processes. Based on graphs, two experiments from node and edge perspectives were investigated based on train operation data, i.e., (1) node experiment: train dwelling times and running times are predicted; and (2) edge experiment: an algorithm based on evolution graph, called overtaking identification algorithm&#xa0;(OIA), is proposed to identify train overtaking based on the consequences of the node experiment. Timetable rescheduling solutions are obtained by integrating the GNN, OIA, and train operation constraints. Experimental results show that the proposed approach has a satisfactory predictive performance. Timetable rescheduling cases under diverse&#xa0;delay scenarios are examined, showing that the proposed method is superior to other three standard rule-based benchmarks regarding train delays of the disturbed train groups under the given scenarios. Additionally, the model exhibits high efficiency in the three timetable rescheduling scenarios, demonstrating the model’s applicability in real-time train dispatching.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Solving the railway timetable rescheduling problem with graph neural networks

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年9月27日

查看全文
  • Ping Huang,
  • Zihuan Peng,
  • Zhongcan Li,
  • Qiyuan Peng

摘要

This study solves the train timetable rescheduling (TTR) problem from a brand-new perspective. We assume that train traffic controllers take three main actions, i.e., adjusting dwelling times, running times, and train orders, to reschedule the timetable in real-time dispatching. To raise the interpretability of rescheduling models, we propose a graph neural network (GNN) approach to map the train timetable data into evolution graphs that fit the paradigm of train operation processes. Based on graphs, two experiments from node and edge perspectives were investigated based on train operation data, i.e., (1) node experiment: train dwelling times and running times are predicted; and (2) edge experiment: an algorithm based on evolution graph, called overtaking identification algorithm (OIA), is proposed to identify train overtaking based on the consequences of the node experiment. Timetable rescheduling solutions are obtained by integrating the GNN, OIA, and train operation constraints. Experimental results show that the proposed approach has a satisfactory predictive performance. Timetable rescheduling cases under diverse delay scenarios are examined, showing that the proposed method is superior to other three standard rule-based benchmarks regarding train delays of the disturbed train groups under the given scenarios. Additionally, the model exhibits high efficiency in the three timetable rescheduling scenarios, demonstrating the model’s applicability in real-time train dispatching.