Research On Description And Modeling Of Crowd Flow Dynamics

Advancements of urbanization propose new challenges of pedestrian safety management,especially in the large public transportation area like metro stations.Transportation authority currently lacks an understanding of pedestrian dynamics and thus are unable to accurately assess crowd safety risks and subsequently establish regulations to manage crowds.Therefore,modeling and analysis of crowd motion are in imperative demand.The thesis focused on the modeling of crowd motion and proposed a series of exploration,which began with literature review.We summarized the history and development of this topic and discussed the advantages and disadvantages of different methods.The specific statements on social force model(SFM),a typical microscopic model in crowd motion,and machine learning were introduced.SFM is widely accepted in the research of microscopic crowd motion.However,the consideration of group behavior is not included in SFM.As group behavior is such a common phenomenon,in this study,we proposed a modified social force model that incorporates group behavior(GB-SFM)and demonstrated the validity of the proposed GB-SFM by comparing simulation results from the original SFM and proposed GB-SFM with true field observations.In addition,we used the GB-SFM to estimate various pedestrian movement efficiency and safety parameters using different crowd volumes and paired pedestrian ratios.The modification and promotion of SFM brought an understanding of two main drawbacks resulting in its intrinsic logic.Therefore,a new model of crowd motion based on different framework was proposed in this study.With the help of machine learning,the relationship of pedestrian velocity and surroundings were explored and applied into the new model.This study also implemented the new model in our in-house simulation code and validated its reliability.The results of this thesis will bring a deeper understanding of crowd motion and improve the risk management of pedestrian fluid in public urban space.This is also the first time that machine learning is used in modeling of crowd motion,which provide a new idea to solve this problem.