Analysis Of Occupational-residential Distribution And Prediction Of Crowd Flow In Cities

With the rapid development of big data mining and its other related technologies,the concept strategy of "smart city" has also entered people’s vision.Nowadays,in the urban construction,all kinds of sensing and monitoring facilities are becoming more and more complete,and the obtained information and data about cities are becoming more and more comprehensive.The core idea of "smart city" is to analyze the data obtained from the daily operation of the city,so as to obtain the data that is useful for the overall planning and scheduling of the city.The research of population flow density in this paper refers to the index of population flow in a certain area,which is used to measure the population density in a region.By observing the population density,the activity characteristics and behaviors of people can be analyzed and predicted.Crowd flow has guiding significance to intelligent traffic,intelligent security and other aspects.For example,it can predict some emergencies including traffic congestion、crowd gathering,which can help city managers make rational use of urban resources.Therefore,the research on crowd flow is of great significance and has reference value for other subjects of "smart city".In this paper,the mobile signaling data of Changchun is regarded as the main research object.At the same time,taxi GPS data,weather information data and urban point of interest(POI)is taken as assist research data to improve data accuracy.Because mobile signaling data can extract location information,and has the advantages of large sample size,wide coverage,small intervention and high objectivity,mobile signaling data is particularly suitable for us to extract crowd traffic density in various urban areas.In order to conduct behavioral analysis of urban population in Changchun and establish the model for the flow characteristics of crowd density,this paper first analyzes the occupational-residential distribution of residents in Changchun,and then proposes a crowd flow prediction method by using multi-source data.The main work of this paper is as follows:1.Firstly,through feature extraction and data processing technology,we analyze and process the existing mobile signaling data and taxi GPS data of Changchun,so as to turn the data into features required for modeling.In addition,this paper also make use of web crawler to extract relevant POI data and weather information data,and processes them to extract features according to specific analysis rules.2.Based on the analysis of Changchun mobile signaling data of five consecutive working days in July,the locations of employment and residence of Changchun residents are identified and summarized.Also,different regions of the occupation and residence situation is analyzed according to the division of the administrative region.Through the analysis of occupational-residential location,it can be clearly seen that Changchun is in good jobs-housing conditions,and the daily commuting distance of residents is very healthy.Also,the employment distribution in Changchun urban areas is even and there are no overcrowding of posts and the appearance spatial distribution of employment centers,which will provide spatial analysis basis for crowd density prediction.3.The prediction method aimed at crowd flow is proposed.Due to the temporal continuity and spatial influence of the crowd flow density,we make a grid division of urban areas.In order to improve the prediction accuracy of the model,the grids are clustered according to time series and space factors,and the different clustering was trained separately.In machine learning training,we start from the idea of building prediction model in data competition,and mainly use the XGBOOST and LightGBM algorithm in gradient promotion tree.By comparing with other machine learning algorithms,an excellent algorithm with the equilibrium between efficiency and accuracy is selected to conduct model training for the multi-source and multi-dimensional features,and then the model is integrated with clustering method to establish model.4.Finally,experimental verification has done.Through the analysis of the average relative error between the predicted result and the real value,the experimental results prove the accuracy and effectiveness of the urban crowd flow prediction method based on multi-source data.