The invention relates to a dynamic graph convolution traffic flow prediction method based on spatial-temporal characteristics, and the method employs an STF-SDAM-DGCN model which is composed of a spatial-temporal embedding module, a spatial-temporal characteristic fusion module, a weighted fusion layer, and an external factor module. The method comprises the following steps: firstly, performing period division on historical traffic flow data, fusing the historical traffic flow data with space-time embedding information learned by a space-time embedding module to serve as input data of a space-time feature fusion module, and constructing a sparse directed adjacency matrix by using an attention mechanism, asymmetric convolution and a Zero-Softmax function; inputting the data into a spatial-temporal feature fusion module, and fusing the data with external influence factors in an external factor module to obtain historical traffic flow spatial-temporal feature data fused with the external factors; the data is then transmitted into a weighted fusion layer, a space-time embedding module transmits predicted future time step information into the layer, future traffic flow representation is generated according to historical traffic flow characteristics obtained by an encoder, the future traffic flow representation serves as decoder input, prediction is carried out through a decoder, and a final future traffic flow prediction result is obtained through a full connection layer.

    本发明涉及一种基于时空特征的动态图卷积交通流预测方法,采用由时空嵌入模块、时空特征融合模块、加权融合层和外部因素模块组成的STF‑SDAM‑DGCN模型,首先对历史交通流数据进行周期划分后与时空嵌入模块学习到的时空嵌入信息融合作为时空特征融合模块的输入数据,同时使用注意力机制、非对称卷积以及Zero‑Softmax函数构建稀疏有向邻接矩阵;将数据输入时空特征融合模块中,同时与外部因素模块中的外部影响因素融合后得到融合外部因素的历史交通流时空特征数据;数据随后传入加权融合层,时空嵌入模块将预测的未来时间步信息也传入该层,依据编码器得到的历史交通流特征生成未来交通流表示,并将其作为解码器输入,通过解码器进行预测,经过全连接层得到最终的未来交通流预测结果。


    Access

    Download


    Export, share and cite



    Title :

    Dynamic graph convolution traffic flow prediction method based on spatial-temporal characteristics


    Additional title:

    一种基于时空特征的动态图卷积交通流预测方法


    Contributors:
    LIU ZHANWEN (author) / WANG YANG (author) / LI WENQIAN (author) / NIU YIDAN (author) / YANG NAN (author) / CHENG JUANRU (author) / XUE ZHIBIAO (author) / FAN JIN (author) / JIA XIAOHANG (author) / ZHAO BINYAN (author)

    Publication date :

    2023-11-10


    Type of media :

    Patent


    Type of material :

    Electronic Resource


    Language :

    Chinese


    Classification :

    IPC:    G08G Anlagen zur Steuerung, Regelung oder Überwachung des Verkehrs , TRAFFIC CONTROL SYSTEMS / G06N COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS , Rechnersysteme, basierend auf spezifischen Rechenmodellen



    Dual Dynamic Spatial-Temporal Graph Convolution Network for Traffic Prediction

    Sun, Yanfeng / Jiang, Xiangheng / Hu, Yongli et al. | IEEE | 2022



    Traffic prediction method based on dynamic graph convolution

    FAN JIN / WENG WENCHAO / TIAN HAO et al. | European Patent Office | 2023

    Free access

    Dynamic graph convolution traffic speed prediction method

    LIU QILIANG / YUAN HAOTAO / YANG LIU et al. | European Patent Office | 2020

    Free access

    Spatial-Temporal Position-Aware Graph Convolution Networks for Traffic Flow Forecasting

    Zhao, Yiji / Lin, Youfang / Wen, Haomin et al. | IEEE | 2023