A Data-driven Method For Simulating Group Behavior And Intervention Strategies

Swarm is a pervasive natural phenomenon in biological world,whose group motion behaviors have features of self-organization,coordination,emergence.It is fittable for researching and reference.Based on RVO,we research on the problem of group movement behavior simulation and intervention using Genetic Algorithm optimization method and K-means clustering method.The main work and contributions are summarized as follows:(1)Applying RVO model and Genetic Algorithm optimization method,a data-driven method for simulating group movement behavior is proposed.First of all,a framework for simulating group movement behavior is established;Then,based on it,a data-driven simulation method is designed.Datas of real-world group behavior are extracted using UWB;The behaviors of simulated group are modeled based on RVO.A framework for optimizing paramenters in RVO is proposed using Genetic Algorithm.Finally,in a simulated enviornment established by simulink and C++,the proposed method is validated,and tracking error of group motion behavior satisfies precision requirement.(2)Refer that intervention approach of population congestion behavior,an external intervention method,which combines density prediction with virtual obstacle,is proposed.1)In order to avoid the congestion and disorder among group with high-desity behavior,a population density prediction method based on K-means clustering method is proposed: First of all,we implement RVO motion model to predict the movement state of the group at the next time;Then,define a function that determines whether there is a high-density group;finally,the K-means clustering method is used to determine the high-density region.2)After determining the high-density region,this region is grouped by the method of arranging the virtual obstacle to reduce the density of the region.First of all,we again implement the K-means clustering method to divide the high-density regions into two categories;Then,we select the appropriate method to calculate the classification line;Finally,a virtual obstacle is set at the posotion of the classification line to achieve the purpose of grouping.3)In addition,the collective behavior is guided by setting the virtual barrier at the classification line.Through simulation test,the proposed method is validated,and the time for passing obstacles is shorten.(3)A simulation system is established using UWB and ROS,and the proposed algorithms in chapter 3 and chapter 4 are tested on it.Fistly,a data-driven method for simulating group movement behavior is carried on simulation system: we extract the Datas of real-world group behavior based on UWB;Then use RVO to simulate the behaviors of real-world group;Secondly,an external intervention method based on density prediction with virtual obstacle is carried out based on ROS.Through the simulation test,the proposed method is further confirm the validity.