Mixed Discrete-Continuous Railway Disruption-Length Models with Copulas

Aurelius Zilko

Research output: ThesisDissertation (TU Delft)

184 Downloads (Pure)


The uncertainty of railway disruption length hinders the performance of the Operational Control Centre Rail (OCCR) in Utrecht. One way to model this uncertainty is by representing the disruption length as a probabilistic distribution. A dependence model, taking the form of a joint distribution, between the disruption length and several observable influencing factors is constructed for a particular type of disruption. From the model, the conditional distribution of disruption length can be computed by conditioning the model on the observed values of the influencing factors.

In this thesis, the joint distribution is constructed using the concept of copula and vines. One focus of this thesis is to study this construction when the variables involved are both discrete and continuous. We show that this can still be done, despite the more expensive parameters estimation.

One value from the conditional distribution of disruption length needs to be chosen as the prediction. To investigate the effect of different choices of prediction, the model is tested in four case studies concerning a railway disruption occurring in the area of Houten, the Netherlands. The model is used together with the short-turning and the passenger flow models, developed by the Department of Transport and Planning of Delft University of Technology. Different predictions are made and the impact on the passengers is measured in terms of the total generalized travel time.
Original languageEnglish
QualificationDoctor of Philosophy
  • Redig, F.H.J., Advisor
  • Kurowicka, D., Advisor
Thesis sponsors
Award date30 Jan 2017
Print ISBNs978-94-6186-776-6
Publication statusPublished - 2017


  • railway disruptions
  • Dependence model
  • copula
  • vines
  • railway traffic management
  • Disruption

Fingerprint Dive into the research topics of 'Mixed Discrete-Continuous Railway Disruption-Length Models with Copulas'. Together they form a unique fingerprint.

Cite this