From theory to practice II: a comprehensive approach for the sensitivity analysis of high dimensional and computationally expensive traffic simulation models

摘要

The reliability of traffic model results is strictly connected to the quality of its calibration. A challenge arising in this context concerns the selection of the most influential input parameters. A model Sensitivity Analysis (SA) should be used with this aim. Unfortunately, due to the limitation of time and computational resources, a proper SA is hardly performed in the common practice. A recent study introduced a methodology based on Gaussian process meta-models for the SA of computationally expensive traffic simulation models. Its main limitation was, however, its dependence on the model dimensionality. When the model has more than 15~20 parameters (depending on its regularity), the estimation of a Gaussian process meta-model (also known as Kriging meta-model) may become problematic. In this light, the SA of high-dimensional and computationally expensive models still remains an issue. In the present paper, the Kriging-based approach has been coupled with another recently developed approach (the quasi-OTEE) for the SA of computationally expensive models. The quasi-OTEE SA can be used to identify the whole sub-set of sensitive parameters of a high-dimensional model, and the Kriging-based SA can then be used to refine the analysis and rank the different parameters of the sub-set in a more reliable way. The application of this new SA method is illustrated with the Wiedemann-74 car-following model. Results show that the new method requires 40 times less model evaluations than a standard variance-based SA in identify the influential parameters and their ranks.