Multi-objective Calibration/Validation of a Microscopic Simulation Platform and the Effect of Goodness-of-Fit Form on Calibration Results

In previous research it was shown that current state-of-the-art calibration methods (single-criteria and weighted summation) are inadequate for calibrating microscopic traffic simulation platforms, since traffic dynamics is a multi-faceted problem wherein speed, volume and density interact with one another. Simulation outputs such as traffic operations (delay, travel time), road safety performance (time-to-collision, etc) and emissions are functions of these vehicular interactions. Calibrating the later measures without calibrating the underlying fundamental attributes does not ensure the validity of the simulation platform outputs. The previous results from a Pareto Archive Dynamically Dimensioned Search (PA-DDS) case study are showcased, along with a new methodology called Non-dominated Sorted Genetic Algorithm (NSGA II) for the three criteria of speed, volume and crash potential index (CPI)/vehicle (a surrogate safety measure). The results are compared to those obtained from a weighted summation calibration using Genetic Algorithm. Different goodness-of-fit measures, Root Mean Squared Percentage Error and Mean Absolute Normalized Error, were tried to gauge their affect on the outcome of calibration results.