Research And Application Of Data-Driven Model For Human Mobility Simulation

With the continuous advancement of urbanization,urban planning,construction and operation management are being transformed toward digitalization,systematization,intelligentization and refinement.It is of great significance to study and simulate the flow patterns of people within the city limits to improve the level of urban planning,construction and management.Therefore,this paper conducts a research on crowd mobility simulation model based on data driven in a variety of scenarios,and provides support for strategic deduction of urban planning,construction and operation management in various application scenarios by generating crowd movement trajectories within the city scope.This paper studies the simulation model of crowd mobility based on data driving for the city scope and the scene of small towns or theme parks.The main work and innovations are summarized as follows:Firstly,this paper proposes ExterPOISim,a crowd mobility simulation algorithm based on generative adversarial network framework,which adopts dual attention mechanism and POI embedding for the task of crowd mobility simulation within the city.ExterPOISim model enhances the ability of the model to capture crowd movement rules by introducing double attention feature extraction network and POI embedding network,and at the same time,it introduces speed loss to constrain the generator,thus improving the performance of crowd trajectory simulation.Finally,the proposed algorithm is compared with the baseline method based on the real trajectory dataset,and the effectiveness of the algorithm is verified.Second,a crowd mobility simulation algorithm based on generative adversarial simulation learning ExterGAIL,is proposed for crowd mobility simulation tasks in small towns or theme parks.Through the design of action space to ensure that the generated crowd movement trajectory to meet the spatial continuity.In addition,the prior knowledge of crowd movement is introduced to enhance the ability of the strategy generator to capture crowd movement rules,and the action mask dictionary is used to limit the possible action output of the strategy generator.In the training,the pre-training process is used to make the model learn the crowd movement law from the real crowd mobility trajectory before formal training,so as to speed up the training of the model.Finally,experiments based on public datasets are conducted to verify the performance of the proposed ExterGAIL algorithm to generate crowd movement trajectories.