Pattern Recognition in Urban Traffic Flows

Accurate predictions of traffic volumes allow authorities to optimize local and network-wide traffic management systems. Traffic flows show variations with respect to different time scales. The authors argue that many of the long- and short-term fluctuations are predictable, because they are recurrent. Due to noise, it is however not trivial to disentangle all temporal patterns. In this study, we propose a novel machine learning method that decomposes a traffic flow time series into multiple interpretable temporal patterns, so-called profiles. We use a data-driven neural network approach that automatically finds (1) the long- and short-term profiles in the presence of noise, (2) the magnitude of each profile at any given time, and (3) the linear combination of these profiles to construct a ‘denoised’ reproduction of the observed traffic volume time series. We apply our approach to traffic volumes in the city of Enschede, the Netherlands, using two years of data..

  • Supplemental Notes:
    • This paper was sponsored by TRB committee ABJ70 Standing Committee on Artificial Intelligence and Advanced Computing Applications.
  • Corporate Authors:

    Transportation Research Board

  • Authors:
    • Eikenbroek, Oskar A L
    • Mes, Martijn
    • Van Berkum, Eric C
  • Conference:
  • Date: 2019


  • English

Media Info

  • Media Type: Digital/other
  • Features: Figures; References;
  • Pagination: 18p

Subject/Index Terms

Filing Info

  • Accession Number: 01697443
  • Record Type: Publication
  • Report/Paper Numbers: 19-01430
  • Files: TRIS, TRB, ATRI
  • Created Date: Mar 1 2019 3:50PM