Time-based vs. distance-based sampling in probe vehicle data: Implications for travel time estimation
2012 (English)In: Proceedings of 17th International Conference of Hong Kong Society for Transportation Studies, Hong Kong, 2012Conference paper (Refereed)
This paper analyses the role of the sampling protocol for travel time estimation with low frequency probe vehicle data by likelihood-based methods, such as maximum likelihood or Bayesian estimation. In the literature, there are reported cases where the vehicle positions are sampled at either certain time intervals, say, every minute, or at certain distance intervals, say, every 500 meters. We show that whether sampling is distance-based or time-based determines the proper formulation of the likelihood function. Furthermore, an incorrect likelihood formulation (for example, treating sampling as distance-based when it is time-based in actuality) often leads to biased parameter estimates. For the special case in which the path is partitioned into segments with constant, independent travel speeds we derive explicit formulas for the likelihood function for each of the two sampling principles. We also study the consistency and bias of the estimators in numerical experiments.
Place, publisher, year, edition, pages
Hong Kong, 2012.
travel time, estimation, sampling, GPS, probe
Transport Systems and Logistics
IdentifiersURN: urn:nbn:se:kth:diva-115985ScopusID: 2-s2.0-84896871412OAI: oai:DiVA.org:kth-115985DiVA: diva2:588566
17th International Conference of Hong Kong Society for Transportation Studies
TSC import 2110 2013-01-15 QC 201306122013-01-152013-01-152013-06-12Bibliographically approved