Calibration of Decision-Based Crowd-Behaviour Model

Abstract

Various methods of calibration are used depending on the model type, application, and individual preferences. While there is no universally applicable method, statistical techniques became popular in recent decades. Introduced calibration concept consists of separate calibration episodes to avoid choosing only a few metrics to describe the whole system and a high computational time increasing exponentially with the number of parameters. These episodes are designed to be separated from each other and to cover one type of pedestrian behaviour captured by some model parameters. The design of the calibration quantities; estimate of the needed simulation time to get stationary results; and the number of iterations by Chebyshev's inequality influencing the quality of the results are discussed. Furthermore, hypothesis testing (James' test) is used to compare the model and experimental data. This calibration process can be applied for any pedestrian model; this paper deals with its application on the crowd-behaviour phase in the author's decision based model.