Spatial–Temporal Traffic Modeling With a Fusion Graph Reconstructed by Tensor Decomposition

Abstract

Accurate spatial-temporal traffic flow forecasting is essential for helping traffic managers take control measures and drivers to choose the optimal travel routes. Recently, graph convolutional networks (GCNs) have been widely used in traffic flow prediction owing to their powerful ability to capture spatial-temporal dependencies. However, designing the spatial-temporal graph adjacency matrix, which is essential to the success of GCNs remains an open question. This paper proposes a GCN-based traffic flow forecasting method that reconstructs the binary adjacency matrix via tensor decomposition. We first reformulate the spatial-temporal fusion graph adjacency matrix into a three-way adjacency tensor. Then, we use Tucker decomposition to reconstruct the adjacency tensor, encoding more informative and global spatial-temporal dependencies. Finally, we propose multiple Spatial-temporal Tensor Graph Convolution layers that assemble a Spatial-temporal Synchronous Graph Convolutional module for localized spatial-temporal correlations learning and a Dilated Convolution module for global correlations learning in parallel. This enables the comprehensive spatial-temporal dependencies of the road network to be aggregated and learned. Experimental results on four open-access datasets demonstrate that the proposed model outperforms state-of-the-art approaches in terms of prediction performances.