Abstract:In traffic forecasting, graph convolutional networks (GCNs), which model traffic flows as spatio-temporal graphs, have achieved remarkable performance. However, existing GCN-based methods heuristically define the graph structure as the physical topology of the road network, ignoring potential dependence of the graph structure over traffic data. And the defined graph structure is deterministic, which lacks investigation of uncertainty. In this paper, we propose a Bayesian Spatio-Temporal Graph Convolutional Network (BSTGCN) for traffic prediction. The graph structure in our network is learned from the physical topology of the road network and traffic data in an end-to-end manner, which discovers a more accurate description of the relationship among traffic flows. Moreover, a parametric generative model is proposed to represent the graph structure, which enhances the generalization capability of GCNs. We verify the effectiveness of our method on two real-world datasets, and the experimental results demonstrate that BSTGCN attains superior performance compared with state-of-the-art methods.

What problem does this paper attempt to address?

This paper attempts to solve two main problems in traffic prediction: 1. **The problem of graph structure definition**: Existing methods based on Graph Convolutional Networks (GCN) usually heuristically define the graph structure through the physical topology of the road network. This approach ignores the potential dependencies in traffic data and fails to accurately describe the relationships between traffic flows. For example, the relationship between two roads with similar trends but far apart geographically is also very important for traffic prediction, but this dependency cannot be captured in a graph pre - defined based on the physical topology. 2. **The problem of graph structure determinacy**: In existing GCN methods, the graph structure is deterministic and lacks consideration of uncertainty. Introducing uncertainty (such as randomly dropping nodes or edges) can enhance the generalization ability of GCN, but existing methods have not conducted in - depth research on this. To solve these problems, the paper proposes a Bayesian Spatio - Temporal Graph Convolutional Network (BSTGCN). This network learns the graph structure end - to - end from the physical topology of the road network and traffic data, thereby discovering a more accurate description of traffic flow relationships. In addition, a parameterized generative model is proposed to represent the graph structure, enhancing the generalization ability of GCN. ### Specific solutions - **Graph structure learning**: BSTGCN regards the graph structure as a sample drawn from a parameterized generative model and infers the posterior probability of the graph structure based on the physical topology of the road network and traffic data. This can more accurately describe the relationships between traffic flows. - **Introducing uncertainty**: By introducing uncertainty (such as randomly dropping nodes or edges), the generalization ability of the model is improved. - **End - to - end learning**: The parameters of the generative model are optimized together with the weights of GCN through the back - propagation algorithm, achieving end - to - end learning. ### Experimental verification The paper verifies the effectiveness of BSTGCN on two real - world datasets. The experimental results show that BSTGCN significantly outperforms existing methods on multiple evaluation metrics, especially in long - term prediction. ### Summary This paper solves the problems of inaccurate graph structure definition and lack of uncertainty in existing GCN methods for traffic prediction by introducing the Bayesian framework and end - to - end learning, improving the accuracy of traffic prediction.

Bayesian Spatio-Temporal Graph Convolutional Network for Traffic Forecasting

A Spatial-Temporal Gated Hypergraph Convolution Network for Traffic Prediction

Bayesian Graph Convolutional Network for Traffic Prediction

Attention Based Graph Covolution Networks for Intelligent Traffic Flow Analysis

Probabilistic spatio-temporal graph convolutional network for traffic forecasting

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

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

Spatio-Temporal Joint Graph Convolutional Networks for Traffic Forecasting

Attention Based Spatial-Temporal Graph Convolutional Networks for Traffic Flow Forecasting

Hierarchical Graph Convolution Network for Traffic Forecasting

Hierarchical Spatio–Temporal Graph Convolutional Networks and Transformer Network for Traffic Flow Forecasting

Local-Global Spatial-Temporal Graph Convolutional Network for Traffic Flow Forecasting

Dynamic Spatio-Temporal Hypergraph Convolutional Network for Traffic Flow Forecasting

Spatial–Temporal-Correlation-Constrained Dynamic Graph Convolutional Network for Traffic Flow Forecasting

3D Graph Convolutional Networks with Temporal Graphs: A Spatial Information Free Framework For Traffic Forecasting

Dual Dynamic Spatial-Temporal Graph Convolution Network for Traffic Prediction

Traffic Graph Convolutional Recurrent Neural Network: A Deep Learning Framework for Network-Scale Traffic Learning and Forecasting

Lanczos Method for Spatio‐temporal Graph Convolutional Networks to Forecast Expressway Flow

Unified Spatio-Temporal Modeling for Traffic Forecasting using Graph Neural Network