The invention discloses a traffic flow prediction method based on discrete wavelet transform and a space-time self-attention network, and the method comprises the steps: 1, an initial embedding generation stage: decomposing original data into trend and event items through employing DWT, and enriching the features of input data through employing various kinds of embedded information, and 2, a space-time self-attention learning stage; in the stage, low-frequency traffic flow and high-frequency traffic flow are modeled through a space-time self-attention network, the space-time self-attention network is composed of three modules, a time self-attention module captures a dynamic long-term time pattern, and a geographic space self-attention module and a semantic space self-attention module model short-distance dynamic space dependence and long-distance dynamic space dependence respectively; 3, a self-adaptive polymerization stage; high-frequency flow and low-frequency flow are obtained through the step 1, and the high-frequency flow and the low-frequency flow serve as input; performing the step 2 for many times to obtain output; in order to retain meaningful events and fuse the meaningful events with stable trends, adaptive aggregation is adopted to obtain final output. According to the method, the time self-attention module is used for capturing time dependence, the two space self-attention modules are used for capturing space dependence, and the prediction precision is improved by combining various embedding and information delay perception.
基于离散小波变换和时空自注意力网络的交通流量预测方法,包括:步骤1:初始嵌入生成阶段,使用DWT将原始数据分解为趋势和事件项,使用多种嵌入信息丰富输入数据的特征,步骤2:时空自注意力学习阶段;该阶段通过时空自注意力网络分别对低频和高频的交通流量进行建模,时空自注意力网络由三个模块组成,时间自注意模块捕捉动态长期的时间模式,地理空间自注意模块和语义空间自注意模块分别建模近距离和远距离动态空间依赖;步骤3:自适应聚合阶段;通过步骤1获得了高频和低频流量,将高频和低频流量作为输入;经过多次步骤2得到输出;为了保留有意义的事件并和稳定的趋势融合,采用自适应聚合得到最终输出。本发明利用时间自注意力模块捕获时间依赖,两个空间自注意力模块捕获空间依赖,并结合多种嵌入和信息延迟感知来提升预测精度。
Traffic flow prediction method based on discrete wavelet transform and space-time self-attention network
基于离散小波变换和时空自注意力网络的交通流量预测方法
2024-09-27
Patent
Electronic Resource
Chinese
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 network
European Patent Office | 2023
|Traffic flow prediction method based on space-time causal attention network
European Patent Office | 2024
|Traffic flow prediction method based on time-space diagram attention neural network
European Patent Office | 2025
|Highway traffic flow prediction method based on Transform and graph attention network
European Patent Office | 2023
|Traffic flow prediction method based on space-time diagram wavelet convolutional neural network
European Patent Office | 2022
|