基于残差时序图卷积网络的交通速度预测

Traffic Speed Prediction Based on Residual Time Series Graph Convolutional Network

准确的交通速度预测是现代智慧交通系统中重要的组成部分,对解决交通拥堵和保障公众出行安全具有重要的意义。针对现有的交通预测模型存在对交通速度中长期预测任务效果不是很好的问题,提出一个新的基于残差时序图卷积网络深度学习框架RSATCN。首先利用可学习的遮罩矩阵和图卷积网络相结合来捕捉空间特征,再利用时间注意力提取时间序列的动态相关性,最后用残差时序网络捕捉时间特征和速度特征。在两个真实世界的数据集上的实验表明,提出的模型在预测交通速度中长期任务方面优于最新的基线。

Accurate traffic speed prediction is an important part of the modern intelligent transportation system,which is of great significance for solving traffic congestion and ensuring the safety of public travel.Aiming at the problem that the existing traffic prediction models are not very effective for medium and long-term traffic speed prediction tasks,a new deep learning framework RSATCN based on residual time series graph convolutional networks is proposed.Firstly,a learnable mask matrix and a graph convolutional network were combined to capture spatial features,secondly,then temporal attention was used to extract the dynamic correlation of time series,and finally a residual time series network was used to capture temporal and speed features.Experiments on two real-world datasets show that the proposed model outperforms state-of-the-art baselines on the medium and long-term task of predicting traffic speeds.