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。
Traffic flow prediction method based on cyclic space-time attention mechanism
一种基于循环时空注意力机制的交通流量预测方法
2024-02-20
Patent
Elektronische Ressource
Chinesisch
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
Europäisches Patentamt | 2024
|Dynamic space-time neural network traffic flow prediction method based on attention mechanism
Europäisches Patentamt | 2023
|Ground road lane-level traffic flow prediction method based on space-time attention mechanism
Europäisches Patentamt | 2021
|Traffic flow prediction method based on space-time attention network
Europäisches Patentamt | 2023
|Europäisches Patentamt | 2023
|