Abstract:In complex and volatile transportation traffic situations, accurately forecasting traffic flow can greatly facilitate to manage and plan real-time traffic. However, due to complex road map and traffic conditions, forecasting temporal traffic is a highly challenging task. In literature, most existing works rely on various spatial–temporal models solely built on the topological structure of road networks, to capture both spatial and temporal dependencies of traffic flow at multiple sensing sites. However, besides the road topological relationship, the dynamics of real-world traffic is indispensably affected by the complex but unknown inter-dependency of multiple sites. Based on the above consideration, this paper proposed a novel traffic forecast framework based on multiple-view graph construction and learning, and spatiotemporal graph neural network (STGNN). Specifically, our work's contributions are following. First, considering transfer entropy (TE)-based relationship graph can characterize the causal influence from one sensing sites to another, the traditional undirected topological graph is transformed into a directed graph through multiplying with the constructed TE relationship graph, which can capture both the topological relationship, and latent casual influence from one site to another, and greatly facilitate to accurately forecast traffic. Meanwhile, a learnable graph matrix is trained to complement the TE-enhanced causality graph. Second, the proposed STGNN-based forecasting framework smoothly integrates recurrent neural network and GNN that works on directional and learnable graphs, so-called DL-GNN. In detail, the aggregation process in DL-GNN is formulated as three parts: aggregation on indegree and outdegree directions of the causality graph, and on learnable graph to fully extract the sites' representations. The comprehensive experiments on real traffic data demonstrate that our proposal outperforms other state-of-the-art STGNN-based traffic forecasting schemes.

Adaptive Spatio-temporal Graph Neural Network for traffic forecasting

Traffic Flow Forecasting Based on Adaptive Graph Structure for Spatio-Temporal Attention Mechanism Network

Do We Really Need Graph Neural Networks for Traffic Forecasting?

Spatio-Temporal Graph Neural Network for Traffic Prediction Based on Adaptive Neighborhood Selection

Spatial-Temporal Adaptive Graph Convolution with Attention Network for Traffic Forecasting

Spatial-temporal graph neural network based on gated convolution and topological attention for traffic flow prediction

Automated Spatio-Temporal Synchronous Modeling with Multiple Graphs for Traffic Prediction

Learning Dynamics and Heterogeneity of Spatial-Temporal Graph Data for Traffic Forecasting

Dynamic Multi-Granularity Spatial-Temporal Graph Attention Network for Traffic Forecasting

Adaptive spatial-temporal graph attention networks for traffic flow forecasting

Make More Connections: Urban Traffic Flow Forecasting with Spatiotemporal Adaptive Gated Graph Convolution Network

Spatial-Temporal Interactive Dynamic Graph Convolution Network for Traffic Forecasting

Spatial-Temporal Interactive Graph Neural Network for Traffic Forecasting.

A Graph Convolutional Stacked Temporal Attention Neural Network for Traffic Flow Forecasting

TF-MVGNN: an accurate traffic forecasting framework based on spatial–temporal graph neural network through exploiting multiple-view graph construction and learning

Spatio-Temporal Graph Convolutional Networks: A Deep Learning Framework for Traffic Forecasting

STNN: A Spatial-Temporal Graph Neural Network for Traffic Prediction.

Spatio-Temporal Attention-based Graph Convolution Networks for Traffic Prediction

Spatial-Temporal Fusion Graph Neural Networks for Traffic Flow Forecasting

Adaptive Hybrid Spatial-Temporal Graph Neural Network for Cellular Traffic Prediction

Graph learning-based spatial-temporal graph convolutional neural networks for traffic forecasting