Font Size: a A A

Research On Grouping Algorithm For Group Evacuation Simulation

Posted on:2020-08-13Degree:DoctorType:Dissertation
Country:ChinaCandidate:Y LiFull Text:PDF
GTID:1366330599952304Subject:Management of engineering and industrial engineering
Abstract/Summary:PDF Full Text Request
Frequent public safety accidents(such as trampling,etc)have prompted researchers to pay more attention to features of crowd dynamics,especially evacuation.In general,there are two main research methods for crowd evacuation: the evacuation experiment or exercise and the model simulation.Experiments or exercises are difficult to be performed because of various reasons,and its authenticity needs to be evaluated.Therefore,the computer simulation has gradually become the mains tool for such studies.In fact,it is a challenging and complex task to model and analyze the behavioral features of evacuation.In recent decades,although a lot of achievements have been made and various simulation models have been emerging,there are still some new problems.Social Force Model(SFM)is a classic micro-continuous model.The model has quite a lot of advantages,such as rigorous and universal mathematical formula,calibrated parameters and the comprehensive factors,etc.SFM can not only control the movement of each pedestrian,but also produce interactive impact among the crowd,which can reproduce the phenomenon of many crowd movements.And then,it has been widely used in many directions of crowd dynamics since its publication.However,there are also some flaws such as the lack of consideration of the internal structure of the crowd.In fact,there are not only isolated individuals,but also small groups formed by various social relationships and groups formed by specific scenarios such as exits and routes.In addition,the model also pays little attention to the simulations of groups and their movements within the crowd,and is manifested in the following aspects:(1)The social force model can achieve good results in both simulating an individual and a crowd evacuation,but the model can not reflect the movement state of groups within a crowd well due to its own limitations.In fact,there are a large number of small groups formed by various social relationships and groups composed of some small groups and isolated individuals within the evacuation crowd,which have a significant impact on the evacuation of the whole crowd.Therefore,pre-identification of small groups in the crowd and pre-division of groups according to specific evacuation scenarios are not only the premise of simulating group movement,but also the basis of more realistic simulation of crowd evacuation.(2)The classical sociological model of identifying small groups was established by Mcphail et al.,and it belongs to the field of humanities and social sciences.The model mainly focuses on qualitative theoretical description,and has a large amount of computation and high computational complexity.Although the researches about it are followed by the scholars in the field of science and engineering follow up,most of them pay more attention to the structure of a crowd.And most of attention to small group identification comes from the perspective of computer vision.Therefore,the qualitative research and quantitative analysis of small group identification need to be further discussed.(3)Small groups are usually small,and one group is made of two or three pedstrians,but during evacuation,because of specific scenario factors,such as the number and distribution of exports,many small groups will form some bigger groups to move to different exits.Therefore,it is necessary not only to use clustering algorithm to divide the evacuated crowd into groups,but also to combine the characteristics of specific scenarios,otherwise the results will be inconsistent with the actual situation.(4)The social force model is essentially a simulation model to simulate individual movement,and crowd movement features are the natural manifestation of the self-organizing behavior of individual interaction,so it is difficult to simulate group movements within crowds,and some special modifications for the model are necessary.Therefore,this paper aims to simulate the movement of the internal groups of evacuated people.Firstly,we need to identify the small groups in the crowd,and then use the improved clustering algorithm as a tool to divide the evacuated groups according to social relations and specific evacuation scenarios to obtain the necessary grouping information.Finally an improved social force model based on these informations is proposed,which can be used to simulate crowd evacuation more vividly.The main research contents are as follows:(1)A top-bottom hierarchical clustering algorithm is proposed to identify small groupsMcphail et al.studied the movement characteristics of small groups in the crowd by film recording crowd movements,and proposed a sociological model to identify small groups.However,the three motion features of the model are not easy to be solved,especially the third one which has a high computational complexity and a large amount of computation.After analysis and calculation,the third feature(angle of motion direction)is replaced equivalently as distance increment.The number of valid records is accumulated as an average operator to calculate similarity,which can filter out local outliers and improve the calculation accuracy of similarity between individuals.A top-bottom hierarchical clustering method is proposed to identify small groups on this basis.This clustering algorithm combines greedy strategy and improves the clustering algorithm based on density peak(proposed by Rodriguez et al.).It can automatically d etermine the number of clusters and divide them into two layers: top-level coarse-grained and bottom-level fine-grained,which significantly improves the quality of clustering.This process provides a more accurate basis for quantifying interpersonal relationships,and it also provides small group information for group motion simulation.The group partition is based on the combination of social relations and evacuation scenarios in the next part.(2)A clustering algorithm based on grid density and relationship is proposed to divide the evacuated crowd into groupsThe evacuation scenario studied in this thesis is a planar area with multiple exits.The planar area is first divided by gridding.The interpersonal relationship is quantified according to the similarity,and then a core grid is determined according to the individual density and the relationship between individuals in grids.After,the outlier noise is filtered and the sparse grid is separated,and the core grid center is used to cluster a crowd into groups.This is the clustering algorithm based on grid density and relationship,which actually includes two grid partitioning methods,static uniform grid partitioning and dynamic non-uniform grid partitioning.This process is mainly based on interpersonal relationships and the specific scene to quickly divide grids,which is to improve clustering efficiency and accuracy,and to provide grouping information for future group motion simulation.(3)An improved social force model that can simulate group movement is proposed.Groups are the main components of a crowd,and group movement directly affects the movement features of the whole crowd.Therefore,a simulation model of group motion based on the traditional social force model is presented in this paper.The model takes into account the social relationship between pedestrians,and obtains group attraction based on group information.The attraction is the combination of two forces,one is the interpersonal attraction between individuals in the group;the other is the group leader's guiding attraction to members in the group.Experiments show that group attraction determines the individual's trajectory and reflects the group's state of motion.(4)The validity of the improved social force model is verified in the crowd evacuation simulation platformA virtual scene is built according to the real scene and deployed to the crowd evacuationsimulation platform built by our research group to enhance the realistic effect.The improved social force model which simulates group movement in evacuated crowds is embedded in the simulation platform and compared with the classical social force model in the evacuation simulation experiment.Finally,the simulation results of the model are verified to be more in line with the actual evacuation process than the classical social force model.
Keywords/Search Tags:Crowd Evacuation, Clustering Algorithm, Group Dynamics, Small Group, Social Force Model, Group Motion Simulation
PDF Full Text Request
Related items