A Methodology for the Microscopic Calibration of Agent-Based Pedestrian Simulation Models

Computer simulation of pedestrian dynamics has gained recent interest as a promising tool for analyzing pedestrian behavior. However, the calibration and the validation of model parameters represent one of the major issues that limit the use of existing simulation models in many applications. The objective of this study is to introduce a methodology for calibrating model parameters on the microscopic (individual) level, using actual pedestrian data. The goal is to select model parameters that minimize the error between pedestrian trajectories resulting from the simulation and the actual pedestrian trajectories. The proposed approach, using genetic algorithms, was applied to calibrate a recently developed agent-based pedestrian simulation model, using four sets of pedestrian data from three different cities. The results showed that the proposed calibration approach leads to accurate simulated trajectories that reflect pedestrian behaviour during different interactions.