通信学报 ›› 2021, Vol. 42 ›› Issue (3): 54-64.doi: 10.11959/j.issn.1000-436x.2021018

• 学术论文 • 上一篇    下一篇

多源异构数据融合的城市私家车流量预测研究

刘晨曦, 王东, 陈慧玲, 李仁发   

  1. 湖南大学信息科学与工程学院,湖南 长沙 410082
  • 修回日期:2020-12-13 出版日期:2021-03-25 发布日期:2021-03-01
  • 作者简介:刘晨曦(1999- ),女,回族,安徽亳州人,湖南大学博士生,主要研究方向为智能交通、轨迹数据挖掘、城市计算。
    王东(1964- ),男,江西九江人,博士,湖南大学教授、博士生导师,主要研究方向为车载网络、智能交通、网络性能测试、无线网络协议与性能分析。
    陈慧玲(1997- ),女,湖南衡阳人,湖南大学硕士生,主要研究方向为轨迹数据挖掘、信息管理、自然语言处理。
    李仁发(1957- ),男,湖南郴州人,博士,湖南大学教授、博士生导师,主要研究方向为计算机体系结构、嵌入式计算、无线网络、虚拟与仿真。
  • 基金资助:
    国家自然科学基金资助项目(61272061);地理信息工程国家重点实验室开放基金资助项目(SKLGIE2018-M-4-3)

Study of forecasting urban private car volumes based on multi-source heterogeneous data fusion

Chenxi LIU, Dong WANG, Huiling CHEN, Renfa LI   

  1. College of Computer Science and Electronic Engineering, Hunan University, Changsha 410082, China
  • Revised:2020-12-13 Online:2021-03-25 Published:2021-03-01
  • Supported by:
    The National Natural Science Foundation of China(61272061);The Open Fund of State Key Laboratory of GeoInformation Engineering(SKLGIE2018-M-4-3)

摘要:

通过有效地捕获城市私家车出行的时空特征,提出一种多源异构数据融合的私家车流量预测模型。首先,融合私家车轨迹和城市区域数据表征城市私家车的出行分布;其次,通过多视角时空图建模私家车出行和城市区域之间的动态关联,设计了多图卷积-注意力网络以提取车流量演变的时空特征;最后,进一步融合时空特征与天气等外部特征,联合预测私家车流量。在长沙市和深圳市采集的真实数据上进行验证,实验结果表明,与现有的模型相比,所提模型的均方根误差约降低了11.3%~20.3%,平均绝对百分误差约降低了10.8%~36.1%。

关键词: 多源异构数据, 兴趣区域, 图神经网络

Abstract:

By effectively capturing the spatio-temporal characteristics of urban private car travel, a multi-source heterogeneous data fusion model for private car volume prediction was proposed.Firstly, private car trajectory and area-of-interest data were integrated.Secondly, the spatio-temporal correlations between private car travel and urban areas were modeled through multi-view spatio-temporal graphs, the multi-graph convolution-attention network (MGC-AN) was proposed to extract the spatio-temporal characteristics of private car travel.Finally, the spatio-temporal characteristics and external characteristics such as weather were integrated for joint prediction.Experiments were conducted on real datasets, which were collected in Changsha and Shenzhen.The experimental results show that, compared with the existing prediction model, the root mean square error of the MGC-AN is reduced 11.3%~20.3%, and the average absolute percentage error is reduced 10.8%~36.1%.

Key words: multi-source heterogeneous data, area of interest, graph neural network

中图分类号: 

No Suggested Reading articles found!