The invention relates to the technical field of intelligent traffic data processing, in particular to a large-scale traffic data complementing method and system based on space-time semantic understanding, and the method comprises the steps: carrying out the preprocessing of to-be-complemented traffic data, and generating preprocessing result data comprising incomplete traffic flow data, geographic edges and label semantic tensors; and inputting the preprocessing result data into a pre-trained graph converter model, converting a traffic data completion problem into an incomplete traffic data semantic description slice recovery problem by using the graph converter model, and obtaining a traffic data completion result through semantic description slice recovery. According to the method, the GT-TDI model is used for estimating flow data in a large-scale network and an extreme missing scene, semantic description is introduced to help the GT-TDI model to capture the spatial-temporal correlation at a network level with proper granularity, more accurate interpolation estimation can be provided, and particularly, under an extreme true condition with a relatively large missing rate, the method has the advantages of being high in robustness and high in robustness. The traffic data complementation performance can be improved by accurately capturing the space-time correlation.
本发明涉及智能交通数据处理技术领域,特别涉及一种基于时空语义理解的大规模交通数据补全方法及系统,通过对待补全的交通数据进行预处理,生成包括不完全交通流量数据、地理边缘及标签语义张量的预处理结果数据;将预处理结果数据输入至预训练的图转换器模型,利用图转换器模型将交通数据的补全问题转化为不完整的交通数据语义描述切片恢复问题并通过语义描述切片恢复获取交通数据补全结果。本发明利用GT‑TDI模型估算大规模网络和极端缺失场景中的流量数据,同时引入语义描述帮助GT‑TDI模型在网络级别以适当的粒度捕获时空相关性,能够提供更准确的插补估算,尤其是在缺失率较大的极端确实情况下,能够通过准确捕获时空相关性来提高交通数据补全性能。
Large-scale traffic data completion method and system based on space-time semantic understanding
基于时空语义理解的大规模交通数据补全方法及系统
2024-02-06
Patent
Electronic Resource
Chinese
British Library Conference Proceedings | 2004
|A tensor-based method for missing traffic data completion
Online Contents | 2013
|A tensor-based method for missing traffic data completion
Elsevier | 2012
|IEEE | 2004
|