Design And Implementation Of Crowd Density Prediction System Based On Ensemble Learning

With the improvement of people’s quality of life,more and more diversified public areas have become the choice of entertainment and work places for modern people.However,the social security problems in public areas are also happens frequently because of the growth of this trend.As people go out more often,it is significant to control the visitors flowrate for the work of public safety.Nowadays,COVID-19 is not well controlled in the world,especially in densely populated areas.Forecasting crowd density efficiently and accurately can provide reference indicators for security work,and then to reduce the probability of virus transmission and stampede accidents.Traditional crowd density detection methods mainly include real-time video detection and crowd flow modeling statistics,but there are also some defects,such as it is difficult to predict the trend of crowd flow in the future and lack of research on inter regional correlation.This paper is for research the law of the development of pedestrian flow in key public areas of the city,build a prediction model of this problem,and develop a regional pedestrian flow prediction system on this basis.This paper conducts two studies: the research of pedestrian flow prediction model and the development of prediction system.(1)In the work of establishing the prediction model of crowd density,this paper studies the regular pattern of data development with the means of data analysis,and constructs new features according to these laws.Finally,it is proved that these features are effective through experiments.This paper mainly uses xgboost,lightgbm and catboost to build a single model of pedestrian flow forecast,and then optimizes the parameters of these models.In the end,it establishes a weighted fusion model.(2)In the work of system development.Firstly,this paper analyzes whether the system can be developed,and then it analyzes the audience types and basic functions of the system.It determines the framework of the system,describes the function of each module and design the database table.Finally,this paper establishes a crowd density prediction system based on the above and shows the function of the system.In this paper,the data set records the population density index of each area every hour,but there is population mobility between each area.For this kind of spatially related data,before building the algorithm model,the traditional method is to input the information of the surrounding area together,but this method is easy to ignore the information of the area in the whole space.The innovation of this paper is to construct the directed weighted graph corresponding to each hour through the connection strength between the public areas,stores the information of the edge-weighted directed graph through the edge list,and then build the regional spatial features through the node2 vec algorithm based on random walk,integrates the learned spatial feature information into an n-dimensional vector,and finally takes the vector input ensemble learning model to improve the prediction effect of the model.