Submit

Login Login

Register Register

Home / Journals / CMC / Online First / doi:10.32604/cmc.2026.088700
Journal Menu
Special Issues
Table of Content

All issues

Online First

2026

2025

2024

2023

2022

2021

2020

2019

2018

2017

2016

2015

2014

2013

2012

2011

2010

2009

2008

2007

2006

2005

2004

Open Access

ARTICLE

A Proximity-Aware Road Network–Integrated Trajectory Graph Neural Network for Road-Level Traffic Flow Prediction

Qun Zhang1,2, Yusheng Mei1, Ziyue Wang1, Yingzhe Liu1, Zhuofeng Zhao1,*
1 North China University of Technology, Beijing, China
2 Institute of High Energy Physics, Chinese Academy of Sciences, Beijing, China
* Corresponding Author: Zhuofeng Zhao. Email: email
(This article belongs to the Special Issue: Advanced Networking Technologies for Intelligent Transportation and Connected Vehicles)

Computers, Materials & Continua https://doi.org/10.32604/cmc.2026.088700

Received 08 July 2026; Accepted 14 August 2026; Published online 03 September 2026

Abstract

Accurate road-level traffic flow prediction remains challenging in heterogeneous urban networks, where traffic interactions extend beyond direct road connectivity. Existing graph-based models can exploit road topology or vehicle trajectories but often overlook geometric continuity and dependencies between non-adjacent yet functionally related road segments. To address this gap, this study proposes a Proximity-Aware Road Network–Integrated Trajectory Graph Neural Network (PA-RNTrGNN). The model integrates trajectory-derived movement transitions, ordered road geometry, proximity-aware structural relations, and bidirectional traffic status information. Specifically, map-matched trajectories are converted into transition matrices to preserve directional movement causality; road geometries are encoded as ordered sequences using a unidirectional gated recurrent unit (GRU); and a proximity-aware structural matrix is constructed from distance decay, directional similarity, and road-type compatibility. Experiments on the Beijing, Chengdu, and Porto datasets yield overall mean absolute error (MAE) values of 4.491, 8.537, and 0.593. Compared with Multi-Spatio-Temporal Fusion Graph Recurrent Network (MSTFGRN), the best-performing baseline in terms of overall MAE, PA-RNTrGNN reduces MAE by 4.30%, 33.92%, and 2.15% on the three datasets, while also achieving the lowest mean peak-hour MAE. The largest improvement on Chengdu highlights the effectiveness of combining trajectory-driven propagation with proximity-aware structural modeling for large and heterogeneous road networks.

Keywords

Traffic flow prediction; vehicle trajectories; proximity-aware constraints; graph neural networks; road geometry; spatiotemporal modeling

  • 7

    View

  • 1

    Download

  • 0

    Like

Related articles
Copyright© 2026 Tech Science Press
© 1997-2026 TSP (Henderson, USA) unless otherwise stated

Share Link