ISSN 1000-1239 CN 11-1777/TP

计算机研究与发展 ›› 2022, Vol. 59 ›› Issue (2): 282-293.doi: 10.7544/issn1000-1239.20210901

所属专题: 2022空间数据智能专题

• 人工智能 • 上一篇    下一篇



  1. 1(东南大学计算机科学与工程学院 南京 211189);2(东南大学网络空间安全学院 南京 211189) (
  • 出版日期: 2022-02-01
  • 基金资助: 

Spatial-Temporal Graph Neural Network for Traffic Flow Prediction Based on Information Enhanced Transmission

Ni Qingjian1, Peng Wenqiang1, Zhang Zhizheng1, Zhai Yuqing2   

  1. 1(School of Computer Science and Engineering, Southeast University, Nanjing 211189);2(School of Cyber Science and Engineering, Southeast University, Nanjing 211189)
  • Online: 2022-02-01
  • Supported by: 
    This work was supported by the National Key Research and Development Program of China (2018YFB1004300).

摘要: 交通问题不仅影响人们的出行,同时也会带来环境污染以及安全等问题,准确的交通流预测是构建智能交通系统、预防和缓解交通问题的关键.目前的预测方法大多没有考虑到交通流动态的时空相关性、周期性以及线性与非线性等特点.在充分考虑上述因素的基础上,提出一种基于信息增强传输的时空图神经网络模型,主要包含多特征注意力模块、信息增强传输模块、时间注意力模块以及线性与非线性融合模块.其中,多特征注意力模块捕获多种交通特征之间的内在联系,考虑交通流的周期性;信息增强传输模块充分利用了交通网络信息,以增强交通网络的信息传输能力,进而挖掘出复杂动态的空间依赖关系;时间注意力模块负责自适应地提取不同时间间隔之间的依赖关系;线性与非线性融合模块则同时考虑了数据的线性与非线性特征.论文在真实数据集上进行了大量对比实验,实验结果表明,对比目前较为先进的基线方法,提出的方法在交通流的预测性能方面,体现了较为明显的优势.

关键词: 交通流预测, 图神经网络, 时空, 信息增强, 注意力

Abstract: Traffic problems not only affect people’s travel, but also bring environmental pollution and safety problems. Accurate traffic flow prediction is the key to build an intelligent transportation system to prevent and alleviate traffic problems. Most of the current methods do not take into account the dynamic spatial-temporal correlation, periodicity, linear and nonlinear characteristics of traffic flow. In this paper, an information enhanced transmission spatial-temporal graph neural network is proposed, which mainly contains a multi-feature attention module, an information enhanced transmission module, a temporal attention module, and a linear-nonlinear fusion module. The multi-feature attention module captures the intrinsic connection between multiple traffic features and takes into account the periodicity of traffic flow. The information enhanced transmission module makes full use of the information in the traffic network, enhancing the information transmission capability of the traffic network and mining the complex dynamic spatial dependencies. The temporal attention module adaptively extracts the dependencies between different time intervals. Furthermore, the linear-nonlinear fusion module considers both linear and nonlinear data features. A large number of comparative experiments are conducted on real-world datasets, and the experimental results demonstrate that the method proposed in this paper has more obvious advantages in terms of traffic flow prediction performance compared with the state-of-the-art baselines.

Key words: traffic flow prediction, graph neural network, spatial-temporal, information enhanced, attention