作者投稿和查稿 主编审稿 专家审稿 编委审稿 远程编辑

计算机工程 ›› 2023, Vol. 49 ›› Issue (2): 303-313. doi: 10.19678/j.issn.1000-3428.0063777

• 开发研究与工程应用 • 上一篇    下一篇

基于时空注意力网络的动态高速路网交通速度预测

邹国建1,2, 赖子良1,2, 李晔1,2   

  1. 1. 同济大学 道路与交通工程教育部重点实验室, 上海 201804;
    2. 同济大学 交通运输工程学院, 上海 201804
  • 收稿日期:2022-01-17 修回日期:2022-03-16 发布日期:2022-07-05
  • 作者简介:邹国建(1993-),男,博士研究生,主研方向为智慧高速运行云平台关键算法;赖子良,硕士研究生;李晔(通信作者),教授、博士。
  • 基金资助:
    国家自然科学基金(71961137006)。

Traffic Speed Prediction Based on Spatio-Temporal Attention Network for Dynamic Expressway Network

ZOU Guojian1,2, LAI Ziliang1,2, LI Ye1,2   

  1. 1. The Key Laboratory of Road and Traffic Engineering, Ministry of Education, Tongji University, Shanghai 201804, China;
    2. College of Transportation Engineering, Tongji University, Shanghai 201804, China
  • Received:2022-01-17 Revised:2022-03-16 Published:2022-07-05

摘要: 交通速度是影响高速路网通行效率和安全的重要指标,精准预测高速路网交通速度可以减少交通事故和通行时间,预先为交通控制提供有价值的参考信息,对高速公路管理具有重要意义。基于时空注意力网络,提出一种由数据和长期预测任务驱动面向动态高速路网的交通速度预测模型(ST-ANet)。通过图注意力网络提取高速路网的动态空间关联特征,使用长短期记忆网络提取输入数据的时间关联特征。在此基础上,采用基于多头自注意力机制的时间注意力网络计算历史输入数据和预测值之间的相关性,并利用密集连接和层归一化方法进一步提升模型性能。基于中国宁夏回族自治区银川市高速路网监测数据进行实验,结果表明,与GCN-LSTM模型相比,ST-ANet模型预测未来1 h、2 h和3 h内高速路网交通速度的平均绝对误差分别降低4.0%、3.6%和3.9%。

关键词: 交通工程, 高速路网交通速度预测, ST-ANet预测模型, 时间注意力, 空间注意力, 长短期记忆网络

Abstract: Traffic speed is an important indicator of the efficiency and safety of expressway network traffic.The accurate prediction of expressway network traffic speed can reduce traffic accidents and transit time, provide valuable reference information for traffic control in advance, and is significant for expressway management.This paper proposes a speed prediction model based on Spatio-Temporal Attention Network(ST-ANet) for dynamic expressway networks driven by data and long-term prediction tasks, named ST-ANet.A Graph Attention Network(GAT) is used to extract the dynamic spatial correlation features of the expressway network and a Long Short-Term Memory(LSTM) network is used to extract the temporal correlation features of the input data.Consequently, a temporal attention network based on the multihead self-attention mechanism is used to focus on the correlation between historical input data and predicted values.In addition, this study also employs tricks to improve model performance, including dense connections and layer batch normalization methods.The evaluation experiments used monitoring data from the expressway network in Yinchuan City, the Ningxia Hui Autonomus Region, China.The experimental results show that compared with the GCN-LSTM model, the ST-ANet model reduces the Mean Absolute Error(MAE) of the expressway network traffic speed prediction in the next 1, 2, and 3 h by 4.0%, 3.6%, and 3.9%, respectively.

Key words: transportation engineering, expressway network traffic speed prediction, ST-ANet prediction model, temporal attention, spatial attention, Long Short-Term Memory(LSTM) network

中图分类号: