Integrating Multi-Graph Convolutional Networks and Temporal-Aware Multi-Head Attention for Lane-Level Traffic Flow Prediction in Urban Networks

Conference Paper (2025)
Author(s)

Fengmei Sun (Tongji University)

Hong Zhu (Tongji University)

Keshuang Tang (Tongji University)

Yingchang Xiong (Tongji University)

Chaopeng Tan (TU Delft - Traffic Systems Engineering)

Zhixian Tang (The Hong Kong Polytechnic University)

Research Group
Traffic Systems Engineering
DOI related publication
https://doi.org/10.1109/ITSC58415.2024.10920109
More Info
expand_more
Publication Year
2025
Language
English
Research Group
Traffic Systems Engineering
Bibliographical Note
Green Open Access added to TU Delft Institutional Repository ‘You share, we take care!’ – Taverne project https://www.openaccess.nl/en/you-share-we-take-care Otherwise as indicated in the copyright section: the publisher is the copyright holder of this work and the author uses the Dutch legislation to make this work public. @en
Pages (from-to)
1878-1884
ISBN (electronic)
979-8-3315-0592-9
Reuse Rights

Other than for strictly personal use, it is not permitted to download, forward or distribute the text or part of it, without the consent of the author(s) and/or copyright holder(s), unless the work is under an open content license such as Creative Commons.

Abstract

The urban signalized road network, characterized by its dynamic and complex nature due to frequent signal control adjustments and unpredictable demand fluctuations, presents significant challenges for predicting lane-level traffic flow. This study introduces the innovative MGCN-TAMA model, which addresses these challenges by integrating multi-graph convolutional networks with a temporal-aware multi-head attention mechanism. The proposed model employs three types of adjacency matrices-a geographical matrix, a signal matrix, and an attention matrix-to capture the complex spatial dependencies among various traffic approaches. Additionally, the model utilizes temporal-aware multi-head attention to discern the nonlinear correlations in traffic variations over time. Tested on a real-world dataset from Tongxiang City, the MGCN-TAMA model significantly outperforms traditional models. Notably, in the first 30-minute prediction interval, our model achieves the lowest Mean Absolute Error, with 2.5649 vehicles per 5-minute span. These results underscore the effectiveness of combining graph-based methods with advanced attention mechanisms to enhance the accuracy of predicting lane-level traffic volumes in urban networks.

Files

Integrating_Multi-Graph_Convol... (pdf)
(pdf | 1.1 Mb)
- Embargo expired in 20-09-2025
License info not available