The invention discloses a space-time adaptive dynamic graph convolutional network traffic flow prediction method, which can capture the time dependence of traffic flow by introducing a gated time convolutional network and using expansion causal convolutional networks with different granularities, and the receptive field size of the expansion causal convolutional networks is exponentially increased along with the increase of the number of hidden layers. Under the support of stack expansion causal convolution, the STADGCN can efficiently and effectively process time-space diagram data with a long-range time sequence, and a self-adaptive mixed graph convolution module composed of a static self-adaptive graph learning module, a dynamic graph learning module and a space gating fusion module is designed. Static adaptive graph learning and dynamic graph learning are respectively utilized to capture global spatial change and local spatial change, and dynamic spatio-temporal characteristics in traffic flow historical data are fully captured through fusion of a spatial gating fusion module, so that accurate traffic flow prediction is facilitated.
本发明公开了时空自适应动态图卷积网络交通流预测方法,能够通过引入门控时间卷积网络,通过使用不同粒度的扩张因果卷积网络来捕捉交通流的时间依赖性,扩张因果卷积网络的感受野大小随着隐藏层数量的增加呈指数增长,在堆叠扩张因果卷积的支持下,STADGCN能够高效且有效地处理具有长程时间序列的时空图数据,设计由静态自适应图学习、动态图学习和空间门控融合模块组成的自适应混合图卷积模块,分别利用静态自适应图学习和动态图学习捕获全局空间变化和局部空间变化,通过空间门控融合模块融合充分捕捉交通流历史数据中的动态时空特征,方便进行交通流准确预测。
Space-time adaptive dynamic graph convolutional network traffic flow prediction method
时空自适应动态图卷积网络交通流预测方法
2024-09-10
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 |
Space-time ARMA graph convolutional network traffic flow prediction method
European Patent Office | 2024
|European Patent Office | 2023
|Traffic flow prediction method based on adaptive dynamic fusion graph convolutional network
European Patent Office | 2024
|European Patent Office | 2025
|European Patent Office | 2024
|