The invention discloses a traffic flow prediction method based on a cyclic space-time attention mechanism, and belongs to the field of traffic flow prediction. The prediction method comprises the following steps: firstly, processing original data from a sensor through a data preprocessing module to obtain a traffic flow data set; secondly, inputting the divided data set into a spatial-temporal feature extraction module based on a cyclic spatial-temporal attention mechanism, and fusing the learned spatial-temporal features; and finally, inputting the obtained spatial-temporal characteristics into a prediction module and predicting traffic flow data in a period of time in the future. In the data filling process of the data preprocessing stage, a forward filling method is designed according to the characteristics of traffic flow data, and the calculation overhead is greatly reduced. According to the method, sliding self-attention position codes with circulation characteristics are designed, the problem that traffic flow prediction is poor in performance on long dependence is solved, distributed training on multiple GPUs is supported through a parallel computing framework, the computing parallelism higher than that of a single GPU is provided, and the computing efficiency is improved.

    本发明公开了一种基于循环时空注意力机制的交通流量预测方法,属于交通流量预测领域。预测方法包括:首先,将来自传感器的原始数据经数据预处理模块得到交通流量数据集;其次,将划分后的数据集输入到基于循环时空注意力机制的时空特征提取模块,融合学习到的时空特征;最后,将得到的时空特征输入到预测模块并预测未来一段时间的交通流量数据。在数据预处理阶段的数据填充过程中,根据交通流量数据的特点设计了前向填充法,大幅降低了计算开销。本方法设计了带有循环特性的滑动自注意力位置编码,解决了交通流量预测在长依赖上表现不佳的问题,并通过并行计算架构骨架支持多个GPU上的分布式训练,提供高于单个GPU的计算并行度,提高运算效率。


    Access

    Download


    Export, share and cite



    Title :

    Traffic flow prediction method based on cyclic space-time attention mechanism


    Additional title:

    一种基于循环时空注意力机制的交通流量预测方法


    Contributors:
    XU XIAO (author) / ZHANG LEI (author) / LIU BAILONG (author) / LIANG ZHIZHEN (author) / JIANG HAIFENG (author) / ZHU CHUANQI (author) / ZHANG ZENGZHI (author)

    Publication date :

    2024-02-20


    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 / G06F ELECTRIC DIGITAL DATA PROCESSING , Elektrische digitale Datenverarbeitung / G06N COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS , Rechnersysteme, basierend auf spezifischen Rechenmodellen



    Traffic flow prediction method based on space-time attention mechanism

    WANG JIAYING / YANG HENG / SHAN JING et al. | European Patent Office | 2024

    Free access

    Dynamic space-time neural network traffic flow prediction method based on attention mechanism

    MENG XIANGFU / XU RUIHANG / FAN HONGYU et al. | European Patent Office | 2023

    Free access

    Ground road lane-level traffic flow prediction method based on space-time attention mechanism

    KONG XIANGJIE / YU KAIFENG / SHEN GUOJIANG et al. | European Patent Office | 2021

    Free access

    Traffic flow prediction method based on space-time attention network

    MA CHUANG / YAN LI / LIU SHUAIWU et al. | European Patent Office | 2023

    Free access

    Traffic flow prediction method of space-time convolution fusion probability sparse attention mechanism

    ZHANG HONG / CHEN LINBIAO / CHEN LINLONG et al. | European Patent Office | 2023

    Free access