Bayesian Calibration for Activity Based Models

We consider the problem of calibration and uncertainty analysis for activity-based transportation simulators. ABMs rely on statistical models of traveler's behavior to predict travel patterns in a metropolitan area. Input parameters are typically estimated from traveler's surveys using maximum likelihood. We develop an approach that uses Gaussian process emulator to calibrate an activity-based model of a metropolitan transplantation system. Our approach extends traditional emulators to handle high-dimensional and non-stationary nature of the transportation simulator. Our methodology is applied to transportation simulator of Bloomington, Illinois. We calibrate key parameters of the model and compare to the ad-hoc calibration process.