Pedestrian Flow Evacuation Simulation Research Based On Two-layer Relationship Mechanism

The rapid increases in pedestrians and crowded urban infrastructures have made it a significance to explore the pedestrian flow dynamic.Pedestrian traffic congestion brings great inconvenience to people’s travels.Moreover,travel delays result in great losses to the social economy and living quality of residents.In addition,in several large public places,such as the square in an assembly,crosswalk or subway station during rush hours,and a crowded concert scene,safe evacuation is a big issue in emergency cases,such as fire and gas leak.The lack of scientific and effective guidance for personnel evacuation results in chaos,which causes stampede and threaten the safety of the lives and properties of pedestrians.Traditional evacuation drills entail enormous manual labor,material resources,and financial resources,and simulating random events and the subconscious interests of pedestrians for their locations is difficult and inefficient.Using computer simulation technology to simulate the pedestrian flow with various population compositions in both daily travel and emergency evacuation situation can minimizes the cost while achieving better simulation effect.However,most of the existing model ignore the complex relationship among pedestrians,thus prevents from simulating the team-group aggregation pattern and collective collision avoidance.Therefore,based on two-layer relationship mechanism,this paper proposes a modified social force model to simulate the self-organization phenomena in real life.Besides,the existing model has a defect in macro path optimization,some existing methods has shortcomings such as slow convergence speed and poor fitting of real path.Thus,based on two-layer relationship mechanism,we propose a modified social force model to drive pedestrians’ motion in collision avoidance.Meanwhile,combined with the upper level macro neural network based Q-Learning path optimization algorithm,it is applied to simulate the self-organized phenomenon of pedestrian flow and evacuate under emergency situations.The main work and innovation of this paper are summarized as follows:1.In order to reproduce the collective motion in pedestrian flow simulation,a two-layer relationship mechanism is proposed.In pedestrian flow simulation,considering that leading and social relationship are existed at the same time,the leading relationship is placed on the upper level,and the social relationship is placed in the lower layer.Thus,two-layer relationship mechanism is formed.In addition,considering the effect of visual field on pedestrian motion,group visual field sharing is proposed to describe the communication between members of a group.Meanwhile,the influence of other pedestrians is removed,which is the premise of self-organization phenomena.2.Considering the team-group pattern is real life,an aggregation algorithm is proposed.Based on the two-layer relationship mechanism,this paper adds an aggregation force into the force formula of social force model.The internal force among the members of team-group in pedestrian flow is more detailed descripted.Thus we can reproduce the team-group pattern in real pedestrian flow.3.There are a lot of friction and collision among pedestrians because of the lack of previous collision avoidance strategy.Thus,this paper comes up with a collective collision avoidance strategy.With this method,pedestrians in group can make previous collision avoidance with other pedestrians who are in their visual field.In this way,the risk of collision is reduced,so the efficient of evacuation is improved.Meanwhile,the safety of pedestrians is guaranteed.It can also reproduce the previous avoidance in real life.4.A Q-Learning path optimization algorithm based on artificial neural network is proposed.According to the selected target,the leader chooses the right path from the optimized path library based on artificial neural network Q-Learning algorithm.And the followers carry out collision avoidance and path following by using the improved social force model.The perceptual performance of the group to the environment information is improved and the pedestrian path finding in the real scene is more realistic fitting.Supported by the National Natural Science Foundation of China(item No.61472232,61272094,etc.),the above theoretical research results are applied to the simulation of pedestrian flow self-organization phenomena and the optimization of the real environment path.The pedestrian flow evacuation simulation system realizes two functional modules,which are micro pedestrian collision avoidance and macro path optimization.By taking into account many environmental factors and real scene information,a virtual scene that corresponds to the real scene is built.By comparing evacuation time with pedestrian density under different scale crowd with different relationship density,the influence on pedestrian evacuation efficiency is obtained,and the optimal value of pedestrian relationship density for evacuation is obtained.On the basis of this optimal value,we compare the pedestrian speed and flow with other classic models and improved models under the same pedestrian density,drawing the conclusion that the improved social force model is effective and practical.At the same time,the optimization path is added to the macro guidance to improve the fitting of the real scene evacuation.Finally,the simulation results are compared with the real pedestrian flow video data to verify that the model can realistically reproduce the self-organizing phenomena of pedestrian flow in real life.