Scholarlink

<p>The complex movement patterns of individuals significantly shape urban structure. However, existing models for predicting mobility flows rely on historical mobility flow data and often fail in its absence. This study proposes an effective and simple model, the adaptive gravity model, that generates origin-destination (OD) flows without using historical OD flow data. It addresses the difficulty of obtaining a suitable distance function in the absence of historical flow data by treating the basic gravity model as a soft-supervised signal to guide the learning of an adaptive distance function. Experiments conducted in New York, Chicago, Los Angeles (USA), Nanjing (China), and Porto (Portugal) demonstrate that the proposed model significantly outperforms the baselines, with an average improvement of 34.58%. The proposed model demonstrates increasing advantages over the baseline as zonal-level traveler volume rises, making it particularly effective for large cities where transport planning is of higher importance. In short, this study proposes a simple and effective method that provides strong support for generating mobility flows in the absence of historical data across diverse urban regions.</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

An Adaptive Gravity Model for Generating Mobility Flows without Historical Flow Data

  • 期刊论文

  • 发表日期：

    2026年3月30日

查看全文
  • Xucai Zhang,
  • Tanhua Jin,
  • Xiaobing Wei,
  • Yeran Sun,
  • Nico Van de Weghe,
  • Frank Witlox,
  • Haosheng Huang

摘要

The complex movement patterns of individuals significantly shape urban structure. However, existing models for predicting mobility flows rely on historical mobility flow data and often fail in its absence. This study proposes an effective and simple model, the adaptive gravity model, that generates origin-destination (OD) flows without using historical OD flow data. It addresses the difficulty of obtaining a suitable distance function in the absence of historical flow data by treating the basic gravity model as a soft-supervised signal to guide the learning of an adaptive distance function. Experiments conducted in New York, Chicago, Los Angeles (USA), Nanjing (China), and Porto (Portugal) demonstrate that the proposed model significantly outperforms the baselines, with an average improvement of 34.58%. The proposed model demonstrates increasing advantages over the baseline as zonal-level traveler volume rises, making it particularly effective for large cities where transport planning is of higher importance. In short, this study proposes a simple and effective method that provides strong support for generating mobility flows in the absence of historical data across diverse urban regions.