Eur. Phys. J. B (2014) 87: 27
https://doi.org/10.1140/epjb/e2014-40699-x

Regular Article

Quantitative comparison between crowd models for evacuation planning and evaluation

¹ School of Computer Engineering, Nanyang Technological University, Nanyang Avenue, 639798 Singapore, Republic of Singapore
² Division of Physics and Applied Physics, School of Physical and Mathematical Sciences, Nanyang Technological University, 21 Nanyang Link, 637371 Singapore, Republic of Singapore
³ Computational Science, University of Amsterdam, Science Park 904, 1098 XH Amsterdam, The Netherlands
⁴ Complexity Program, Nanyang Technological University, 60 Nanyang View, 639673 Singapore, Republic of Singapore
⁵ National Research Institute ITMO, 197101 St. Petersburg, Russia

^a e-mail: vaisagh1@e.ntu.edu.sg

Received: 23 July 2013
Received in final form: 17 October 2013
Published online: 3 February 2014

Abstract

Crowd simulation is rapidly becoming a standard tool for evacuation planning and evaluation. However, the many crowd models in the literature are structurally different, and few have been rigorously calibrated against real-world egress data, especially in emergency situations. In this paper we describe a procedure to quantitatively compare different crowd models or between models and real-world data. We simulated three models: (1) the lattice gas model, (2) the social force model, and (3) the RVO2 model, and obtained the distributions of six observables: (1) evacuation time, (2) zoned evacuation time, (3) passage density, (4) total distance traveled, (5) inconvenience, and (6) flow rate. We then used the DISTATIS procedure to compute the compromise matrix of statistical distances between the three models. Projecting the three models onto the first two principal components of the compromise matrix, we find the lattice gas and RVO2 models are similar in terms of the evacuation time, passage density, and flow rates, whereas the social force and RVO2 models are similar in terms of the total distance traveled. Most importantly, we find that the zoned evacuation times of the three models to be very different from each other. Thus we propose to use this variable, if it can be measured, as the key test between different models, and also between models and the real world. Finally, we compared the model flow rates against the flow rate of an emergency evacuation during the May 2008 Sichuan earthquake, and found the social force model agrees best with this real data.