From theory to practice: Gaussian process metamodels for the sensitivity analysis of traffic simulation models. A case study of the Aimsun mesoscopic model.

Abstract

This paper discusses a metamodel-based technique for model sensitivity analysis and applies it to the Aimsun mesoscopic model. Throughout the paper it is argued that the application of sensitivity analysis is crucial for the true comprehension and correct use of the traffic simulation model while also acknowledging that the main obstacle to an extensive use of the most sophisticated techniques is the high number of model runs they usually require. For this reason we have tested the possibility of performing sensitivity analysis not on a model but on its metamodel approximation. Important issues arising when estimating a metamodel have been investigated and commented on in the specific application to the Aimsun model. Among these issues is the importance of selecting a proper sampling strategy based on low discrepancy random number sequences and the importance of selecting a class of metamodels able to reproduce the inputs-ouputs relationship in a robust and reliable way. Sobol sequences and Gaussian process metamodels have been recognized as the appropriate choices. The proposed methodology has been assessed by comparing the results of the application of variance-based sensitivity analysis techniques to the simulation model and to a metamodel estimated with 512 model runs, on a variety of traffic scenarios and model outputs. Results confirm the powerfulness of the proposed methodology and also open up to a more extensive application of sensitivity analysis techniques to complex traffic simulation models.