物联网学报 ›› 2020, Vol. 4 ›› Issue (3): 78-85.doi: 10.11959/j.issn.2096-3750.2020.00176

• 专题:智慧交通物联网 • 上一篇    下一篇

基于多特征时空图卷积网络的水运通航密度预测

董伟,张磊磊,金子恒,孙伟(),高俊波   

  1. 上海海事大学信息工程学院,上海 201306
  • 修回日期:2020-07-06 出版日期:2020-09-30 发布日期:2020-09-07
  • 作者简介:董伟(1995- ),男,安徽合肥人,上海海事大学硕士生,主要研究方向为注意力机制、海事交通大数据分析|张磊磊(1996- ),男,山西交城人,上海海事大学硕士生,主要研究方向为图卷积神经网络、海事交通大数据分析|金子恒(1997- ),男,上海人,上海海事大学硕士生,主要研究方向为深度学习、时序数据分析|孙伟(1978- ),男,黑龙江哈尔滨人,博士,上海海事大学副教授,主要研究方向为智能感知与优化、海事交通大数据分析等|高俊波(1972- ),男,安徽马鞍山人,博士,上海海事大学副教授,主要研究方向为人工智能、情感倾向分析、机会发现等

Prediction of the waterborne navigation density based on the multi-feature spatio-temporal graph convolution network

Wei DONG,Leilei ZHANG,Ziheng JIN,Wei SUN(),Junbo GAO   

  1. College of Information Engineering,Shanghai Maritime University,Shanghai 201306,China
  • Revised:2020-07-06 Online:2020-09-30 Published:2020-09-07

摘要:

面对港航信息化发展的需求,物联网技术助力我国水运交通感知网络的建设。水运交通大数据分析已成为交通领域研究者和实践者关注的热点。在水运交通中,各港口的通航密度具有非线性、时空相关性和异质性,对其进行精准预测将面临巨大的挑战。提出一种基于多特征时空图卷积网络(MFSTGCN,multi-feature spatio-temporal graph convolution network)的预测方法,解决了水运交通中通航密度的预测问题。MFSTGCN方法从通航量、船舶平均航速和船舶密度3个特征出发,利用空间维图卷积和时间维卷积操作有效捕获通航密度的时空相关性。在某航运平台采集的长江港口船舶自动识别系统(AIS,automatic identification system)数据集上进行实验,结果表明,MFSTGCN 方法的预测效果优于时空图卷积网络(STGCN,spatio-temporal graph convolution network)方法的预测效果。

关键词: 水运交通, 通航密度, 时空相关性, 图卷积网络, 多特征

Abstract:

In the face of the development of the information technology in the port and waterway,the Internet of things (IoT) technology can help to build China’s water transport perception network.The big data analysis of the waterborne transport has become a hot topic for researchers and practitioners in the field of transportation.The navigation density of each port in the water transportation is nonlinear and spatio-temporal correlation,so it is a great challenge to accurately predict it.A multi-feature spatiotemporal graph convolution network (MFSTGCN) was proposed to solve the problem of the traffic density prediction.MFSTGCN effectively captured the spatial-temporal correlation of the ship navigation density data by using the spatial convolution and temporal convolution through three features,which were navigation volume,average ship speed and ship density.The experiment was carried out on the automatic identification system (AIS) data set collected from a shipping platform.The results show that the prediction effect of the MFSTGCN model is better than the spatio-temporal graph convolution network (STGCN) model.

Key words: waterborne traffic, navigation density, spatio-temporal correlation, graph convolution network, multiple feature

中图分类号: