Calibration of Hyperparameters for Pedestrian Flow Model Based on Bayesian Optimization

The study of pedestrian flow has attracted considerable interest in the last three decades. Simulation of pedestrian flow provides a convenient approach for researchers to take insights into the dynamics of pedestrian crowd under various situations. The results of simulation can be helpful in emergency predicting and avoiding crowd disasters. In this study, a well-known pedestrian behavior dynamic model, the social force model, is investigated. Bayesian optimization is applied to calibrate the hyperparameters of the model by minimizing the difference of the fundamental diagrams between simulation results and empirical data given in literature. With the hyperparameters calculated by Bayesian optimization, the fundamental diagram of pedestrian flow is nearly the same as the empirical result under various densities.