Calibration of microscopic traffic simulators using travel times and trajectories.

Abstract

Traffic network simulation has become the solution for estimating the performance of existing and future projects. It provides this framework through the incorporation of several driver behavior models, which through their interaction create a realistic representation of individual drivers. As such, calibration of these sub-models is necessary to emulate reality. In this work, two methodologies are presented that use trajectories (or derivatives, from the Next-Generation Simulation (NGSIM) data set and a GPS study, respectively) to perform automated calibration. The techniques for the respective calibrations are sampling and optimization. Both methodolgies utilize Latin Hypercube Sampling, but the second methodology is paired with a meta-heuristic optimization, called the Firefly Algorithm for further performance improvement.