Predicting traffic propagation flow in urban road network with multi-graph convolutional network

• Published in: Complex & intelligent systems Vol. 10; no. 1; pp. 23 - 35
• Main Authors: Yang, Haiqiang, Li, Zihan, Qi, Yashuai
• Format: Journal Article
• Language: English
• Published: Cham Springer International Publishing 01.02.2024; Springer Nature B.V; Springer
• Subjects: Artificial neural networks; Communications traffic; Complexity; Computational Intelligence; Data Structures and Information Theory; Engineering; Graph convolutional network; Multilayers; Original Article; Propagation; Roads & highways; Spatial data; Spatial–temporal features; Traffic flow; Traffic models; Traffic prediction; Traffic propagation flow; Traffic volume; Upstream; Traffic prediction; Graph convolutional network; Spatial–temporal features; Traffic propagation flow
• ISSN: 2199-4536; 2198-6053
• DOI: 10.1007/s40747-023-01099-z

Traffic volume propagating from upstream road link to downstream road link is the key parameter for designing intersection signal timing scheme. Recent works successfully used graph convolutional network (GCN) and specific time-series model to forecast traffic flow by capturing the spatial–temporal features. However, accurately predicting traffic propagation flow ( tpf ) is challenging, since the classical GCN model only considers the influence of adjacent road link. In complex urban road network, specific traffic propagation flow ( tpf ) is affected by various spatial features, such as adjacent tpf , which influences from tpf with same upstream link and tpf with same downstream link. Thus, we proposed a multi-graph learning-based model named TPP-GCN (traffic propagation prediction-graph convolutional network) in this paper to predict the traffic propagation flow in urban road network. The TPP-GCN model captures not only the temporal features but also multi-spatial features based on multi-layer convolution. We validated the model using real-world traffic flow data derived from taxi GPS data in Shenzhen, China. Finally, we compare and evaluate the proposed model with the existing models across several prediction scales.