Building Function Inference And Dynamic Population Distribution Simulation Based On Multi-source Spatiotemporal Data

Since the Reform and 0pening Up,with the rapid development of urbanization,China has experienced profound changes in the way of production and life,social economy and geographical space.China’s overall urbanization rate has increased from 17.9% in 1979 to 60.6%in 2019,and the urban population has increased from 179 million to 848 million.Urbanization refers to a series of processes in which the population gathers to cities and promotes the development of cities.This series of processes will affect the population,ecology and economic development of cities,especially the population urbanization and urban spatial expansion.In2015,the urbanization rate of Guangzhou has reached 85.53%,and Tianhe District has achieved100% population urbanization.Since the 1980 s,Tianhe District has developed rapidly.On the one hand,it facilitates the living,work,commuting and recreation of urban residents;on the other hand,due to the increasing population year by year,the mode of economic development is not in harmony with the population growth rate,It brings great pressure to the limited resources and environmental capacity of the city.Excessive expansion of urban space and rapid population growth and uneven spatial distribution lead to unbalanced regional development,contradictions between people and land,traffic jams,excessive consumption of resources and other problems.The cost of urban operation and maintenance is rising,and the social living environment is relatively tense.For a long time,the field of urban spatial geography and urban planning has paid close attention to the changes and expansion of urban spatial structure and the spatial distribution characteristics and laws of urban population in the process of urbanization.It is of great significance for urban planning and management to fully and timely understand the building function and the spatial distribution of urban population and the interaction between them by using the existing technical means.With the development and transformation of network information technology,remote sensing and GIS technology and big data,the availability of POI data,LBS data and building outline data with strong objectivity and current situation and rich semantic information is enhanced,The human activity information represented by high spatial-temporal resolution LBS data from mobile terminals provides a good spatial-temporal big data information source and analysis means for understanding urban population behavior;It pays more attention to the accuracy of micro level and provides better insight;The law of mass data is helpful to find the relationship reflected by the limitation of samples;It is possible to better understand the architectural function and the spatial distribution characteristics of urban dynamic population.At present,there have been relevant scholars’ research on the use of social media data,mobile signaling data,floating car data,bus smart card data and other LBS big data to identify or infer urban functional areas and building functions.However,most of them focus on the street,plot,traffic district,grid and other large spatial scales,lacking a more detailed understanding of the scale.The data obtained by spatialization of population data is mainly through the periodic census in specific administrative units,which is time-consuming and not updated in time.The world’s mainstream population kilometer grid data includes gridded population of the world(GPW)data set,global rural urban mapping project(GRUMP)data set,land scan data set and Worldpop data set,Like the census data of a specific range of administrative units,the spatial scale is rough,static and single,so it is difficult to obtain more precise information of urban dynamic population distribution in the statistical unit.To sum up,this paper takes Tianhe District of Guangzhou City as an example,based on LBS population density data TUD,building outline data,POI data and other multi-source geographic space-time big data,proposes the combination of principal component analysis(PCA)and K-means clustering algorithm to infer the building function under the influence of human space-time activities;The random forest algorithm is used to simulate the dynamic distribution of urban spatial structure.The research results can help us to understand the heterogeneity of architectural functions and the characteristics and laws of urban dynamic population distribution from a more micro perspective.The main conclusions are as follows:(1)Principal Component Analysis can be used to extract the features reflecting the law of human activities,and K-means clustering algorithm can be used to cluster the extracted features,which can effectively identify the building function.The accuracy of the clustering results is not less than 83%,the highest accuracy of cluster 2: urban village,residential area category,reached 92.76%;The human activity information represented by TUD under buildings can effectively identify and infer the unique spatial heterogeneity characteristics of urban villages in Tianhe District.(2)Tianhe District presents the characteristics of building function classification,which is mainly residential and employment,supplemented by business and education.Different types of buildings cluster show different characteristics of population change;The inference classification of building function in Tianhe District has the feature of mixed function,and the functional division of residential type is relatively independent.For example,cluster 2: urban spatial structure represented by residential buildings such as villages in the city and residential districts;Cluster 4: structural characteristics of mixed functions such as science and education,work and high-grade residential districts.(3)The random forest algorithm can effectively simulate the dynamic distribution of urban population in Tianhe District,and the generalization performance of the model is more than84%.The best generalization performance of the model is 10:00-18:00 on weekdays,and its generalization performance score is 88.61%;The generalization performance scores of 10:00-18:00 on rest day and 21:00-1:00 on rest night were 87.33% and 86.75%,respectively;The generalization performance score of 21:00-1:00 at night is the lowest,still reaching 84.68%.And the root mean square error(RMSE),mean relative error(MRE)and mean absolute error(MAE)of the four models in different periods are relatively low.(4)In the results of four different periods of urban dynamic population simulation,the regional economic center is the second most important factor,and the commercial building,accommodation services,building area are important factors,the effect is observably significant.The major influencing factors of 10:00-18:00 on weekdays and rest days were business building,distance to regional economic center,distance to road and distance to subway station.The total contribution of the four factors was 51.48% and 46.02% respectively;In addition to the regional economic center,the important influence of 21:00-1:00 on working days and rest days at night is the common factor of accommodation service,building area and residential area,and the total contribution of the three factors is 35.71% and 34.98% respectively.The research results of this paper will provide necessary planning evaluation and adjustment information for urban planners,provide scientific and favorable decision-making reference for the optimal allocation of urban infrastructure,and support the sustainable and healthy development of social economy.