.png){kind=icon}
.png){kind=icon}
Return
Traffic prediction using an active causality recurrent graph convolutional network
.png){kind=icon}
PRE
AI
DOI：10.1016/j.eswa.2025.129506.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
The quality of our daily lives is significantly influenced by traffic conditions, highlighting the importance of incorporating complex spatiotemporal dependencies in interconnected traffic data for effective prediction. Although recent advancements have demonstrated prediction accuracy using graph convolutional networks, their depends heavily on the accuracy of the graph structures that represent the spatial relationships within the traffic network. To address this challenge, we introduce a novel approach to traffic prediction, the Active Causal Recurrent Graph Convolution Network (ACRGCN), as shown in Fig. 2. ACRGCN offers a new framework that effectively integrates a causal-embedded approach for traffic prediction, leveraging both structural and feature information from correlated traffic time series. Additionally, it incorporates a time-varying dynamic Bayesian network to capture the intricate spatiotemporal topology of traffic data. The model extracts spatiotemporal dependencies from traffic signals using the Active Causality Graph Recurrent Module (ACGRM), while efficiently modeling nonlinear traffic propagation patterns. Furthermore, ACRGCN employs a deep learning-based module that functions as a hyper-network, progressively generating dynamic causal graphs. Finally, extensive experiments on multiple real-world traffic graph datasets validate ACRGCN, and the results demonstrate its superiority over state-of-the-art method
Journal
IF:
7.5
Papers:
2.9W
Citations:
10.2W
Organization
View more organizations