.png){kind=icon}
.png){kind=icon}
Return
Dynamic graph neural networks for improving the reliability of traffic flow prediction: Progress and prospects
.png){kind=icon}
PRE
AI
B
R
X
Z
Y
M
DOI：10.1016/j.ress.2025.112142.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Traffic flow prediction is the core research direction of intelligent transportation systems. Its accuracy and reliability directly affects traffic congestion management and accident prevention. Dynamic Graph Neural Networks (DGNNs) can effectively model spatial-temporal heterogeneity and improve prediction performance by dynamically updating the parameters of the adjacency matrix. However, there is no comprehensive discussion among scholars about the performance of DGNN in traffic flow prediction studies to unify the caliber. In this paper, 50 studies are screened since 2020 by a traffic meta-analysis approach with a statistical analysis framework. It aims at exploring whether DGNNs outperform GNNs, DGNNs and which model components perform well in traffic flow prediction tasks in real-world application scenarios. The results of the analysis show that DGNNs significantly outperform traditional GNNs in traffic flow prediction. Besides, time-series convolutional networks can assist DGNNs in modeling complex time-series-dependent features in traffic flows. This traffic meta-analysis reveals the effectiveness of the key components of dynamic graph modeling at the theoretical level. Meanwhile, it provides model selection and design guidance for optimal traffic flow management at the practical level.
Journal
R
IF:
11
Papers:
9.0K
Citations:
4.2W
Organization
Cited Papers