Short-term Forecasts for Transport Models

The new national traffic management centre VM-CH (Verkehrsmanagement Schweiz) will start its operations in 2008 and cover the entire Swiss motorway network. An efficient management is possible only if one knows the actual traffic situation and if one is able to estimate how this situation will evolve in the short-term. The goal of this study is to identify the best possible methods and their parameterization, in order to apply short-term forecasts of traffic flow q [vhc/h] and speed v [km/h] to the online traffic model of VM-CH. Many methods were left aside from the beginning, because of their complexity and/or because there is no guarantee of better results. The forecast covers the next two hours, which represents a travel distances up to roughly 200 km on motorways. The forecast consists of two distinct steps. First, one must find the best fitting DVC (daily variation curves) for the past period. Second, one has to apply this DVC to the forecast period. The past horizon is two hours, symmetrically to the future one. A general remark: The quality of short-term traffic forecasts is limited by the inherent uncertainty of the traffic data itself. The precision of the forecast cannot be better than the precision of the traffic data. Measured in terms of mean relative error (MRE), this limit is around 10% for a time horizon of two hours. As long as random fluctuations of traffic lie within this range of precision, there is no hope of obtaining better results.

  • Corporate Authors:


    BERN,   Switzerland  CH-3003
  • Authors:
    • DE RHAM, C
    • SCHWARZ, R
  • Publication Date: 2007-12


  • English

Media Info

  • Pagination: 49P
  • Serial:
    • Issue Number: 1206

Subject/Index Terms

Filing Info

  • Accession Number: 01104607
  • Record Type: Publication
  • Source Agency: Ecole Polytechnique Fédérale de Lausanne (EPFL)
  • Files: ITRD
  • Created Date: Jul 17 2008 12:49PM