Scholarlink

<p>Short-term traffic flow prediction is a prerequisite for optimized decision-making in intelligent transportation systems. It plays a critical role in improving road network efficiency and travel service quality. In traffic flow data collected by road network sensors under regular sampling, actual observations often suffer from data quality issues such as missing values, noise, and operating condition drift. To address this problem, this study introduces a Neural Ordinary Differential Equation module based on the spatial graph convolution and temporal convolution structures of Spatio-Temporal Graph Neural Network, and constructs a new method for short-term traffic flow prediction on road networks. Experiments on the public PeMS04 and PeMS08 traffic flow datasets show that the proposed method achieves Mean Absolute Errors of 5.0 vehicles/5&#xa0;min and 5.8 vehicles/5&#xa0;min for 20-min and 60-min predictions, respectively. The results indicate that the proposed method effectively suppresses error accumulation and phase misalignment. It maintains robustness under multi-step rolling prediction, peak-hour fluctuations, and high missing-rate scenarios, providing a technical reference for freeway monitoring and congestion warning.</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 Analysis of Neural Ordinary Differential Equations and Graph Neural Networks for Traffic Flow Prediction

  • 期刊论文

  • 发表日期：

    2026年7月9日

查看全文
  • Xiaoning Zhu

摘要

Short-term traffic flow prediction is a prerequisite for optimized decision-making in intelligent transportation systems. It plays a critical role in improving road network efficiency and travel service quality. In traffic flow data collected by road network sensors under regular sampling, actual observations often suffer from data quality issues such as missing values, noise, and operating condition drift. To address this problem, this study introduces a Neural Ordinary Differential Equation module based on the spatial graph convolution and temporal convolution structures of Spatio-Temporal Graph Neural Network, and constructs a new method for short-term traffic flow prediction on road networks. Experiments on the public PeMS04 and PeMS08 traffic flow datasets show that the proposed method achieves Mean Absolute Errors of 5.0 vehicles/5 min and 5.8 vehicles/5 min for 20-min and 60-min predictions, respectively. The results indicate that the proposed method effectively suppresses error accumulation and phase misalignment. It maintains robustness under multi-step rolling prediction, peak-hour fluctuations, and high missing-rate scenarios, providing a technical reference for freeway monitoring and congestion warning.