Abstract With well-defined graphs, graph convolution based spatiotemporal neural networks for traffic prediction have achieved great performance in numerous tasks. Compared to other methods, the networks can exploit the latent spatial dependencies between nodes according to the adjacency relationship. However, as the topological structure of the real road network tends to be intricate, it is difficult to accurately quantify the correlations between nodes in advance. In this paper, we propose a graph convolutional network based adaptive graph learning algorithm (AdapGL) to acquire the complex dependencies. First, by developing a novel graph learning module, more possible correlations between nodes can be adaptively captured during training. Second, inspired by the expectation maximization (EM) algorithm, the parameters of the prediction network module and the graph learning module are optimized by alternate training. An elaborate loss function is leveraged for graph learning to ensure the sparsity of the generated affinity matrix. In this way, the expectation maximization of one part can be realized under the condition that the other part is the best estimate. Finally, the graph structure is updated by a weighted sum approach. The proposed algorithm can be applied to most graph convolution based networks for traffic forecast. Experimental results demonstrated that our method can not only further improve the accuracy of traffic prediction, but also effectively exploit the hidden correlations of the nodes. The source code is available at https://github.com/goaheand/AdapGL-pytorch.

    Highlights An adaptive graph learning algorithm (AdapGL) for traffic prediction based on spatiotemporal neural networks is proposed. The prediction network module can be improved with more accurate spatial dependencies. The updated graph can help exploit the hidden correlations of nodes, which is valuable for transportation management and control applications. The convergence of the EM-based AdapGL algorithm can be guaranteed, and this method can be applied to not only existing graph convolution based spatiotemporal prediction networks, but also other kinds of tasks. A novel Parameterized Graph Learning module (PGL) is designed, and the whole traffic prediction framework is divided into two parts, which are a PGL module to generate a possible graph and a Prediction Network module (PN) to get the final result, respectively. A graph set is built by some pre-defined graphs (e.g., distance and correlation coefficient). An alternate training approach is leveraged for the optimization process to replace the traditional end-to-end training. Specifically, in the training process for the PN module, a pre-defined graph generated by the graph set is fixed as the expectation of the optimal structure. Then, in the training process of the PGL module, the optimized PN module is fixed to generate a more effective adjacency matrix. The two training processes run one by one in a circle, and a well-designed loss function is utilized to ensure the sparsity of the generated graph. A weighted average approach is designed to update the current graph structure to control the stability of the training process. Experiments on three GCN-based neural networks and four real-world datasets are carried out for the multi-step traffic flow prediction. The results demonstrate that our algorithm can generate a more reliable graph and improve the performance of the tested networks significantly.


    Access

    Check access

    Check availability in my library

    Order at Subito €


    Export, share and cite



    Title :

    AdapGL: An adaptive graph learning algorithm for traffic prediction based on spatiotemporal neural networks


    Contributors:
    Zhang, Wei (author) / Zhu, Fenghua (author) / Lv, Yisheng (author) / Tan, Chang (author) / Liu, Wen (author) / Zhang, Xin (author) / Wang, Fei-Yue (author)


    Publication date :

    2022-03-20




    Type of media :

    Article (Journal)


    Type of material :

    Electronic Resource


    Language :

    English





    Transfer Learning in Traffic Prediction with Graph Neural Networks

    Huang, Yunjie / Song, Xiaozhuang / Zhang, Shiyao et al. | IEEE | 2021



    Multiadaptive Spatiotemporal Flow Graph Neural Network for Traffic Speed Forecasting

    Xu, Yaobin / Liu, Weitang / Mao, Tingyun et al. | Transportation Research Record | 2022


    Graph Attention Convolutional Network: Spatiotemporal Modeling for Urban Traffic Prediction

    Song, Qingyu / Ming, RuiBo / Hu, Jianming et al. | IEEE | 2020