The invention discloses a multi-task urban traffic flow prediction method based on a graph neural network, which studies a new problem of predicting multi-scale (fine-grained and coarse-grained) traffic flow for the first time, specifically, a road map is given, and a coarse-grained road map is firstly constructed based on topological proximity and traffic flow similarity between nodes (road links); then, a cross-scale graph convolution Cross-Scale GCN is provided to extract traffic flow characteristics of fine granularity and coarse granularity, and the traffic flow characteristics are fused. According to the method, time features are extracted by using an LSTM (Long Short Term Memory) added with Intra-Attention and Inter-Attention, and in order to ensure the consistency of prediction results of two kinds of scale data, structural constraints are introduced. According to the scheme, excellent performance is embodied in the aspect of fine-grained and coarse-grained flow prediction, and the prediction accuracy is improved.
本文公开了一种基于图神经网络的多任务城市交通流量预测方法,首次研究了预测多尺度(细粒度和粗粒度)交通流量的新问题,具体来说,给定道路图,我们首先基于节点(道路链接)之间的拓扑接近度和交通流相似度构建粗粒度道路图;然后提出了一种跨尺度图卷积Cross‑Scale GCN来提取细粒度和粗粒度的交通流量特征,并将它们进行融合。利用加入了Intra‑Attention与Inter‑Attention的LSTM提取时间特征,为了保证两种尺度数据预测结果的一致性,引入了结构约束。该方案在细粒度和粗粒度流量预测方面的同时体现出优异性能,提高了预测的准确率。
Multi-scale traffic flow prediction method based on graph convolutional neural network
一种基于图卷积神经网络的多尺度交通流量预测方法
07.05.2021
Patent
Elektronische Ressource
Chinesisch
Traffic flow prediction method based on graph convolutional network
Europäisches Patentamt | 2025
|Europäisches Patentamt | 2023
|Traffic flow prediction method based on multi-view dynamic graph convolutional network
Europäisches Patentamt | 2022
|Short-term traffic flow prediction method based on integrated graph convolutional neural network
Europäisches Patentamt | 2024
|