Scholarlink

<p>The subway system is increasingly considered the backbone of urban transportation in many cities around the world. Accurate prediction of real-time subway passenger flow is crucial for improving demand management and operational efficiency. However, the intricate spatio-temporal distribution of subway passenger data frequently impedes prediction performance, with prior models primarily concentrating on capturing spatio-temporal connections between various stations. Incorporating weather and atmospheric features, as well as comprehensive spatio-temporal dependencies, into the prediction still poses critical challenges. Therefore, we have developed a deep spatio-temporal network fusion model (DSTNFM) that integrates multiple data sources to model and integrate various subway passenger features. The model consists of five modules, each extracting spatio-temporal characteristics of inflow and outflow between different stations, network topology information, and the influence of weather and air quality. Lastly, we perform weighted feature fusion on output data with the same shape. This research focuses on 276 subway lines in Beijing and uses two time granularities, 10&#xa0;min and 15&#xa0;min, for passenger flow prediction. The numerical results display that the DSTNFM model outperforms the baseline model in terms of prediction accuracy. The results of this research offer contributions to the advancement of deep learning methods for predicting subway passenger flow and offer valuable perspectives for the design and administration of urban transportation systems.</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

DSTNFM: urban subway flow prediction based on network scale

  • 期刊论文

  • 发表日期：

    2026年3月25日

查看全文
  • Xiaoting Huang,
  • Changxi Ma,
  • Yongpeng Zhao

摘要

The subway system is increasingly considered the backbone of urban transportation in many cities around the world. Accurate prediction of real-time subway passenger flow is crucial for improving demand management and operational efficiency. However, the intricate spatio-temporal distribution of subway passenger data frequently impedes prediction performance, with prior models primarily concentrating on capturing spatio-temporal connections between various stations. Incorporating weather and atmospheric features, as well as comprehensive spatio-temporal dependencies, into the prediction still poses critical challenges. Therefore, we have developed a deep spatio-temporal network fusion model (DSTNFM) that integrates multiple data sources to model and integrate various subway passenger features. The model consists of five modules, each extracting spatio-temporal characteristics of inflow and outflow between different stations, network topology information, and the influence of weather and air quality. Lastly, we perform weighted feature fusion on output data with the same shape. This research focuses on 276 subway lines in Beijing and uses two time granularities, 10 min and 15 min, for passenger flow prediction. The numerical results display that the DSTNFM model outperforms the baseline model in terms of prediction accuracy. The results of this research offer contributions to the advancement of deep learning methods for predicting subway passenger flow and offer valuable perspectives for the design and administration of urban transportation systems.