摘 要：针对交通流数据建模时空特性挖掘不足的问题，提出了 STGAN 网络。运用时空图卷积和注意力机制挖掘道路网络时空规律。注意力机制使得网络对相邻道路和历史时间数据的关注度不同，其分组注意力卷积的机制能够使得网络训练摆脱路网空间拓扑规模的限制，并使模型可运用在较大规模的路网上。实验表明，STGAN 模型在宁波高、快速路和主干路上速度预测误差比 DCRNN 小，在宁波数据集上预测速度表现出良好的精度。
关键词：公路运输；速度预测；时空依赖；注意力；图卷积
DOI:10.19850/j.cnki.2096-4706.2023.05.031
基金项目：宁波市交通运输局科技项目（202117）
中图分类号：TP39；U495 文献标识码：A 文章编号：2096-4706（2023）05-0128-04
Research on Road Speed Prediction Based on Spatiotemporal Convolution Attention Network—A Case Study of Ningbo's Main Road Network
HU Zheng, LIN Yang, ZENG Qiulin, SHU Tai, WU Xiaobin
(Ningbo Transportation Development Research Center, Ningbo 315042, China)
Abstract: STGAN network is proposed to solve the problem of insufficient mining of spatiotemporal characteristics of traffic flow data modeling. The spatiotemporal graph convolution and attention mechanism is used to mine the spatiotemporal laws of road network. The attention mechanism makes the network pay different attention to adjacent roads and historical time data. Its grouping attention convolution mechanism can make the network training get rid of the restriction of the spatial topological scale of the road network, and make the model can be applied to a large scale road network. The experiment shows that the STGAN model has a smaller speed prediction error than DCRNN in Ningbo high-speed, expressway and trunk roads, and the prediction speed shows good accuracy on Ningbo data set.
Keywords: road transportation; speed prediction; spatiotemporal dependence; attention; graph convolution
参考文献：
[1] 冉斌．世界智能交通进展与趋势 [J].中国公路，2018（14）：22-23.
[2] ASIF M T，DAUWELS J，CHONG Y G，et al. Spatiotemporal Patterns in Large-Scale Traffic Speed Prediction [J].IEEE Transactions on Intelligent Transportation Systems，2014，15（2）：794-804.
[3] 李树彬，孔祥科，李青桐，等 . 考虑混沌特性的 PSRXGBoost 短期交通流预测（英文） [J]. 东南大学学报：英文版，2022，38（1）：92-96.
[4] YANG B L，SUN S L，LI J Y，et al. Traffic flow prediction using LSTM with feature enhancement [J].Neurocomputing，2019，332：320-327.
[5] 朱凯利，朱海龙，刘靖宇，等 . 基于图卷积神经网络的交通流量预测 [J]. 智能计算机与应用，2019，9（6）：168-170+177.
[6] LI Y G，YU R，SHAHABI C，et al. Diffusion Convolutional Recurrent Neural Network: Data-Driven Traffic Forecastingg [J/OL]. arXiv:1707.01926 [cs.LG].[2022-09-28].https://arxiv.org/abs/1707.01926.
[7] WU Z H，PAN S R，LONG G D，et al. Graph WaveNet for Deep Spatial-Temporal Graph Modeling [J/OL]. arXiv:1906.00121 [cs.
LG].[2022-09-29].https://arxiv.org/abs/1906.00121.[8] RASAIZADI A，ARDESTANI A，SEYEDABRISHAMI S. Traffic management via traffic parameters prediction by using machine learning algorithms [J].IJHCM（International Journal of Human Capital Management），2021，6（1）：57-68.
作者简介：胡睁（1985—），男，汉族，浙江宁波人，高级工程师，硕士，研究方向：交通信息化。