Statistical methods for the calibration and validation of simulation models

Jaume Barceló

Technical University of Catalonia
Department of Statistics and Operations Research
Jaume.barcelo@upc.edu

From a methodological point of view it is widely accepted that simulation is a useful technique to provide an experimental test bed to compare alternate system designs, replacing the experiments on the physical system by experiments on its formal representation in a computer in terms of a simulation model. Model calibration and validation is inherently an statistical process in which the uncertainty due to data and model errors should be account for. This lecture presents explicit methods to take into account the autocorrelation dependencies between traffic data, and the specific time dependencies characteristics of traffic data whose emulation is one of the main abilities of microscopic simulation. The proposal is illustrated with case studies from applications of microscopic traffic simulation where the calibration of route choice models becomes a critical component, from the analysis guidelines for calibration are also proposed in the route based simulation.



2005-05-23