The invention provides a traffic state prediction model (ASTCN) based on adaptive space-time embedding, which effectively learns space-time information of a traffic state through trainable time embedding and a graph adjacency matrix, and utilizes a gating fusion mechanism to control the proportion of different time embedding so as to improve prediction precision. According to the method, the future traffic state change on a traffic road network is effectively predicted, the influence degree of weekly time embedding on prediction precision is found to be maximum, day time embedding is second than day time embedding, and month time embedding is minimum in time embedding visualization, meanwhile, a new graph adjacency matrix construction method is provided, and the prediction accuracy is improved. An ablation experiment finds that the influence degrees of the two different graph adjacency matrixes in the method on the prediction precision of the traffic speed and the traffic flow are different, and the ASTCN shows excellent performance in the prediction results of the traffic speed data set and the traffic flow data set by fusing the two graph adjacency matrixes.
本发明提出了一种基于自适应时空嵌入的交通状态预测模型(ASTCN),其通过可训练的时间嵌入和图邻接矩阵有效学习了交通状态的时空信息,利用门控融合机制控制不同时间嵌入的比例以提高预测精度,有效的预测了交通路网上的未来交通状态变化,在对时间嵌入的可视化中发现周时间嵌入对预测精度的影响程度最大、天时间嵌入次之、月时间嵌入则最小,同时,提出一种新的图邻接矩阵构造方法,通过消融实验发现该方法中的两种不同的图邻接矩阵对交通速度和交通流量的预测精度影响程度不同,并通过融合两种图邻接矩阵使得ASTCN在交通速度和交通流量数据集的预测结果上都表现出优异的性能。
Traffic state prediction method based on adaptive spatio-temporal information
基于自适应时空信息的交通状态预测方法
2024-11-29
Patent
Electronic Resource
Chinese
Traffic flow prediction method for traffic flow spatio-temporal data information
European Patent Office | 2024
|Spatio-Temporal AutoEncoder for Traffic Flow Prediction
IEEE | 2023
|Traffic flow prediction method based on spatio-temporal information fusion and storage medium
European Patent Office | 2024
|