Statistical Validation of Traffic Simulation Models

Abstract

Traffic simulation models support detailed analysis of the dynamics of traffic phenomena and are important tools for analysis of transportation systems. In order to evaluate correctly the impact of different traffic management schemes, simulation models must be able to replicate reality adequately. Model validation (i.e., the process of checking to what extent the model replicates reality) is discussed. The role of validation is defined within the scope of model development and calibration, and the framework for performing the validation is discussed. A hierarchy of statistical methods to validate different types of simulation outputs against observed data is examined. Also, a validation method is proposed on the basis of statistical tests on metamodels fitted to the observed and simulated data. A case study illustrates the applicability of the various methods.