Research On The Identification Method Of Urban Functional Area Based On Multi-source Data

With the development of economy and society,the process of urbanization has occurred widely around the world,and China has accelerated the process of urbanization accordingly.The “Thirteenth Five-Year Plan”(2016-2020)states that mega-and mega-cities should speed up their efforts to improve their internationalization and properly disperse non-core functions in central urban areas.The division and identification of urban functional zones is helpful to analyze the current status of urban functional zone utilization,to understand the internal spatial structure of the city,and to provide a basis for decision-making in order to optimize the urban spatial pattern and unblock non-core functions in the central urban area.Traditional urban functional area division research methods have the problems of large subjectivity,slow data update,high labor costs,and lack of local research.The rise of big data and the update of data mining technologies have provided geographical big data and refined analysis techniques rich in semantic information for the study of the division of urban functional areas.This paper studies the identification method of urban functional areas from two kinds of granularity,using multi-source data as the basic data support for urban functional area identification.The 800 m × 800 m grid division in Zhengzhou's downtown area is used as a fine-grained basic research unit to explore the improvement of the recognition accuracy of POI weights on the same scale.The POI data is preprocessed and reclassified,and different types of POI data are given different weights according to the different geographic areas and public awareness of the corresponding geographical entities of the POI data.Then,the functional area recognition feature vector is constructed,and the functional type of the unit is identified by calculating the type-occupying feature vector of each unit,and the recognition result corresponds to six functional areas.Finally,combined with satellite map and planning map sampling comparison verification,the overall recognition accuracy reached 79%,and the Kappa coefficient was 0.75,indicating that the classification consistency is high.In addition to quantitative identification methods for POI data,this study also uses clustering algorithms in machine learning to identify coarse-grained functional areas.Commonly used clustering algorithms include K-means algorithm,Gaussian mixture model,and DBSCAN algorithm.At present,most of the researches on functional area recognition use K-means algorithm,which requires artificially set the number of clusters(K value)and has poor clustering effect on non-spherical data sets.This study compares and analyzes the advantages and disadvantages of three commonly used clustering algorithms and the applicable scenarios in functional area recognition,providing a theoretical basis for the selection of clustering algorithms.The previous research also ignored the influence of different POI weights in cluster recognition,and studied combining POI weights in cluster recognition to further improve the clustering method recognition effect.In addition,previous studies on functional area recognition were mostly holistic studies,lacking individual studies.The specific manifestation is that when the POI data density is unevenly distributed,the results obtained by the quantitative identification method and the clustering identification method of POI are relative and cannot fully reflect the true distribution status of individual functional areas.For example,a complex area of a city center is classified as a business service area,but the density of public service data is much greater than the density of public service data in a public service area in the fringe area of the city.This paper proposes a functional area boundary extraction algorithm based on kernel density estimation from an individual perspective,extracts individual functional area boundaries from a global scale,and accurately extracts the true spatial distribution of individual functional areas.The kernel density estimation method was used to extract the boundaries of functional areas based on the spatial distribution density changes of POI.It is verified that the method has high applicability to different functional area boundaries.The research fills the blank of individual functional area identification research under POI weight assignment,and provides new ideas for functional area boundary extraction methods.