• •    

CCML2021+92: 面向交通流量预测的时空超关系图卷积网络

张永凯1,武志昊2,林友芳1,赵苡积1   

  1. 1. 北京交通大学计算机与信息技术学院
    2. 北京交通大学
  • 收稿日期:2021-06-07 修回日期:2021-06-15 发布日期:2021-06-15
  • 通讯作者: 赵苡积

CCML2021+92: Spatio-Temporal Hyper Relationship Graph Convolutional Networks for Traffic Flow Forecasting

  • Received:2021-06-07 Revised:2021-06-15 Online:2021-06-15

摘要: 交通流量预测是智能交通系统中的重要研究课题,然而,交通对象(如站点、传感器)之间存在的复杂局部时空关系使得这项研究颇具挑战。尽管以往的一些研究将流量预测问题转化为一个时空图预测问题取得了较大的进展,但是它们忽略了跨时空维度的直接关联性。目前仍缺乏一种全面建模局部时空关系的方法。针对这一问题,首先提出一种新颖的时空超图建模方案,通过构造一种时空超关系来全面地建模复杂的局部时空关系;然后提出一种时空超关系图卷积模型(STHGCN)来捕获这些关系用于交通流量预测。在四个公开交通数据集上进行了大量对比实验,相比ASTGCN、STSGCN等时空预测方法,STHGCN在均方根误差(RMSE)、平均绝对误差(MAE)、平均绝对百分比误差(MAPE)三个评价指标上取得了更优的预测效果;不同方法运行时间的对比结果表明,STHGCN有着更快的推理速度。

关键词: 交通流量预测, 局部时空关系, 时空图预测, 超图, 时空超关系

Abstract: Traffic flow forecasting is an important research topic for the intelligent transportation system. However, it is very challenging to accurately forecast the traffic flow due to the complex local spatio-temporal relationships among traffic objects. Although some previous studies have made great progress by transforming the traffic flow forecasting problem into a spatio-temporal graph forecasting problem, they ignore the direct correlations that across spatio-temporal dimensions among traffic objects. At present, there is still lack of a comprehensive modeling approach for the local spatio-temporal relationships. A novel spatio-temporal hypergraph modeling scheme was first proposed to address this problem by constructing a kind of spatio-temporal hyper relationship to comprehensively model the complex local spatio-temporal relationships. Then a Spatio-Temporal Hyper Relationship Graph Convolutional Network (STHGCN) was proposed to capture these relationships for traffic flow forecasting. Extensive comparative experiments were conducted on four public traffic datasets. Compared with the spatio-temporal forecasting methods such as Attention based Spatial-Temporal Graph Convolutional Networks(ASTGCN) and Spatial-Temporal Synchronous Graph Convolutional Networks(STSGCN), STHGCN achieves better results in three evaluation indicators including Root Mean Square Error (RMSE), Mean Absolute Error (MAE) and Mean Absolute Percentage Error (MAPE); the comparison of the running time of different methods shows that STHGCN has faster inference speed.

Key words: traffic flow forecasting, local spatio-temporal relationship, spatio-temporal graph forecasting, hypergraph, spatio-temporal hyper relationship

中图分类号: