The invention relates to a road traffic safety prediction method based on a dynamic graph attention space-time network. According to the method, firstly, a dynamic graph attention module is provided, and the weight of the structure is dynamically adjusted through the dependency relation of road space features and surrounding important feature buildings. Secondly, the invention also provides a space-time convolution module, and the module is formed by connecting a graph convolution layer with two time convolution layers. The number of parameters is reduced by reducing and increasing feature dimensions in a space-time convolution layer respectively. The time convolution layer comprises two extended convolutions with different activation functions, and outputs of the two extended convolutions are connected so as to extract time features; the graph convolution layer realizes scale compression and feature compression by performing scale ascending and descending on a channel, and spatial features are extracted by using topological properties of a traffic network. And finally, a full-connection network is used when external data such as weather is processed. The results of the three modules are subjected to feature fusion and then output final prediction data through a final full connection layer. According to the invention, the potential traffic accident risk is reduced by improving the accuracy of traffic accident prediction.

    本发明涉及一种基于动态图注意力时空网络的道路交通安全预测方法。在本方法中,首先提出了一个动态图注意力模块,该结构通过对道路空间特征和其周围重要特征建筑的依赖关系动态调整权重。其次,本发明还提出了一个时空卷积模块,该模块由图卷积层连接两个时间卷积层组成。通过在时空卷积层分别降低和升高特征维度,以减少参数的数量。时间卷积层中是带有不同激活函数的两个扩展卷积,并对其输出进行连接,用来提取时间特征;图卷积层通过对通道进行尺度的升降,实现尺度压缩和特征压缩,利用交通网络的拓扑性质提取空间特征。最后,处理天气等外部数据时使用了全连接网络。三个模块的结果进行特征融合后经过最后的全连接层实现输出最终预测数据。本发明通过提高交通事故预测的准确性来减少潜在的交通事故风险。


    Zugriff

    Download


    Exportieren, teilen und zitieren



    Titel :

    Road traffic safety prediction method based on dynamic graph attention space-time network


    Weitere Titelangaben:

    一种基于动态图注意力时空网络的道路交通安全预测方法


    Beteiligte:
    JIA ZHAOHONG (Autor:in) / YUAN JIE (Autor:in) / YANG BO (Autor:in)

    Erscheinungsdatum :

    2024-03-22


    Medientyp :

    Patent


    Format :

    Elektronische Ressource


    Sprache :

    Chinesisch


    Klassifikation :

    IPC:    G08G Anlagen zur Steuerung, Regelung oder Überwachung des Verkehrs , TRAFFIC CONTROL SYSTEMS / G06F ELECTRIC DIGITAL DATA PROCESSING , Elektrische digitale Datenverarbeitung / G06N COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS , Rechnersysteme, basierend auf spezifischen Rechenmodellen / G06Q Datenverarbeitungssysteme oder -verfahren, besonders angepasst an verwaltungstechnische, geschäftliche, finanzielle oder betriebswirtschaftliche Zwecke, sowie an geschäftsbezogene Überwachungs- oder Voraussagezwecke , DATA PROCESSING SYSTEMS OR METHODS, SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL, SUPERVISORY OR FORECASTING PURPOSES



    Traffic flow prediction method and system based on time-space synchronization dynamic graph attention network

    XIA DAWEN / LIN ZHAN / LI HUAQING et al. | Europäisches Patentamt | 2024

    Freier Zugriff

    Road network traffic flow prediction method based on gated loop graph convolution attention network

    REN YILONG / CHEN YUE / MA TIAN et al. | Europäisches Patentamt | 2023

    Freier Zugriff

    Traffic flow prediction method and device based on space-time attention graph neural network

    HUANG WENBING / YUAN JIRUI / TIAN CHUJIE et al. | Europäisches Patentamt | 2024

    Freier Zugriff

    Traffic speed prediction method and device based on space-time attention graph convolutional network

    XIA YINGJIE / TIAN RUI | Europäisches Patentamt | 2021

    Freier Zugriff

    Traffic flow prediction method and system of attention time-space synchronization graph convolutional network

    XIA DAWEN / WEI XIAODUO / LI HUAQING et al. | Europäisches Patentamt | 2023

    Freier Zugriff