汽车工程 ›› 2021, Vol. 43 ›› Issue (11): 1662-1672.doi: 10.19562/j.chinasae.qcgc.2021.11.012

• • 上一篇    下一篇

基于多源数据的车流量时空预测方法

胡杰(),龚永胜,蔡世杰,黄腾飞   

  1. 1.武汉理工大学,现代汽车零部件技术湖北省重点实验室,武汉  430070
    2.武汉理工大学,汽车零部件技术湖北省协同创新中心,武汉  430070
    3.新能源与智能网联车湖北工程技术研究中心,武汉  430070
  • 收稿日期:2021-06-17 修回日期:2021-07-24 出版日期:2021-11-25 发布日期:2021-11-22
  • 通讯作者: 胡杰 E-mail:auto_hj@163.com
  • 基金资助:
    湖北省科技重大专项(2020AAA001)

A Spatio-Temporal Prediction Method of Traffic Flow Based on Multi-Source Data

Jie Hu(),Yongsheng Gong,Shijie Cai,Tengfei Huang   

  1. 1.Wuhan University of Technology,Hubei Key Laboratory of Advanced Technology for Automotive Components,Wuhan  430070
    2.Wuhan University of Technology,Hubei Collaborative Innovation Center for Automotive Components Technology,Wuhan  430070
    3.Hubei Research Center for New Energy & Intelligent Connected Vehicle,Wuhan  430070
  • Received:2021-06-17 Revised:2021-07-24 Online:2021-11-25 Published:2021-11-22
  • Contact: Jie Hu E-mail:auto_hj@163.com

摘要:

为提高车流量的预测精度,本文中从外部特征、时间特征和空间特征的角度综合考虑了多因素对车流量的影响,提出了一种基于多源数据和时空预测的车流量预测方法。在外部特征方面,深入探索了日期、天气和兴趣点特征对车流量影响;在时间特征方面,提出了基于时间卷积网络(TCN)的时间序列预测框架,并以近邻周期和日周期为基线分别建立时间预测模型;在空间特征方面,提出了基于图表示学习的空间特征提取方法,实现了相邻路网节点间的空间相关性特征提取。结果表明,与多种现有预测方法相比,该方法在提升预测精度的同时改善了中长时车流量预测性能。

关键词: 车流量, 多源数据, 深度学习, 时空预测

Abstract:

In order to enhance the accuracy of traffic flow prediction, a traffic flow prediction method based on multi-source data and spatio-temporal prediction is proposed in this paper with comprehensive considerations of the effects of various factors on traffic flow from the perspectives of external features, time features and spatial features. In terms of external features, the influence of the features of date, weather and point of interest on traffic flow is explored in depth. In terms of time features, a time series prediction framework based on time convolution network is put forward, and the time prediction models are established with the neighbor cycle and daily cycle as base lines respectively. In terms of spatial features, a spatial feature extraction method based on graph representation learning is proposed to fulfill the extraction of the spatial correlation feature between adjacent road network nodes. The results show that compared with various existing prediction methods, the method adopted improves the prediction performance of medium- and long-term traffic flow with high prediction accuracy.

Key words: traffic flow, multi-source data, deep learning, spatio-temporal prediction