Research On Density Peak Clustering Analysis Method For Geographic Spatio-Temporal Raster Data

The mining and analysis of spatio-temporal data is of great significance for discovering the internal potential value of data,exploring the law of spatio-temporal distribution,and supporting the application requirements of decision-making.Geospatio-temporal raster data is an important type of spatio-temporal data,which has unique spatio-temporal attribute characteristics and contains multi-dimensional information such as time,space and ground object attributes.it divides the geographical space into several rows and columns through a regular grid,and records the information of geographical elements changing with time in a fixed spatial location.From the point of view of data mining,spatio-temporal raster data has a wide range of application scenarios and requirements because of its convenient acquisition and update,easy fusion and segmentation.Spatio-temporal clustering analysis is an important means of spatio-temporal data mining technology,which can not only be used as an independent data analysis tool,but also serve to discover the spatio-temporal distribution pattern and law of geographical elements,and reveal the characteristics,association relations and evolution trends of geographical entities.Rodriguez proposed the Fast density Peak Discovery algorithm(Clustering by fast search and find of density peaks,CFSFDP)in2014,which is one of the density clustering algorithms.Because of its ingenious design,the ability to identify clusters of arbitrary shape and the ability to identify outliers,it is favored in many analysis and applications.However,the algorithm still has some problems,such as unable to automatically select clustering center,cluster misclassification caused by single cluster with multi-density peak,and can not take into account the spatio-temporal coupling characteristics of geographic spatio-temporal raster data when clustering applications.As a result,its ability to explore the dynamic changes of elements with time and space is limited.Based on this,firstly,this paper studies the improvement of CFSFDP algorithm for fast finding density peak value from the aspects of automatic selection of clustering center and single cluster with multidensity peak value.Furthermore,considering the spatio-temporal coupling characteristics of data,a geographic spatio-temporal raster data clustering method considering temporal,spatial and non-spatio-temporal attribute constraints is proposed.This paper combs the geographic spatio-temporal raster data and its analysis methods,and puts forward the improvement strategy of density peak clustering algorithm.the main research contents and results are as follows:(1)Research on the improvement of CFSFDP.This paper combs the core ideas,algorithm steps,advantages,and disadvantages of the CFSFDP algorithm,expounds the direction of the algorithm to be improved,and analyzes the shortcomings of the current improvement research.Based on this,an improved algorithm "Adaptive Density Peak Tree Clustering algorithm(Adaptive Density Peak Tree Clustering,ADPTC)" is proposed in this paper.The density peak tree is established by two indexes of the density peak algorithm: "local density" and "repulsive group value",and the density peak tree is cut according to the definition of reachable point,connected point,and cutting point,thus the function of automatically selecting clustering center is realized.It also solves the problem of inaccurate clustering results caused by the peak density of single cluster with multi-density.Finally,through the case comparison,this method can automatically select the clustering center when clustering under the same time complexity,and has a good clustering effect.(2)Research on density peak clustering algorithm considering time,space,and non-spatio-temporal attribute constraints.By combing the structural characteristics and application scenarios of geographic spatio-temporal raster data,the clustering requirements for spatio-temporal raster data are determined.Based on the abovementioned ADPTC algorithm,an improved density peak tree clustering algorithm for spatio-temporal raster data is designed and implemented: "time-space-attribute constrained adaptive density peak tree clustering algorithm ST-ADPTC(SpatioTemporal Adaptive Density Peak Tree Clustering)".By constructing the spatiotemporal attribute neighborhood,introducing the K-nearest neighbor to calculate the repulsion group value,and combining with the adaptive density peak tree,the spatiotemporal raster data clustering analysis considering the spatio-temporal coupling characteristics is realized.The density peak clustering algorithm has the ability to discover the dynamic changes of geographical elements with time and space.Taking the European regional precipitation raster data as experimental data,the ST-ADPTC algorithm is compared with the classical spatio-temporal data clustering analysis algorithm.The results show that ST-ADPTC algorithm can automatically select clustering centers when facing spatio-temporal raster data,and the clustering results are more consistent with the realistic law,can better express spatio-temporal differentiation,and can identify spatio-temporal clusters of various complex shapes.(3)ADPTC?LIB tool construction and experimental verification.In order to verify the practicability of the improved algorithm in this paper,the python toolkit of ADPTC?LIB is constructed to complete the clustering analysis of spatio-temporal raster data through different functional modules,including data clustering trend exploration before clustering,data clustering and providing a variety of visualization methods,to support the application of spatio-temporal raster data clustering analysis.Finally,taking the Asian regional temperature raster data as the experimental data,the spatio-temporal clustering analysis of the experimental data is completed by using the data clustering trend analysis method,ST-ADPTC clustering algorithm and spatiotemporal clustering visualization analysis method constructed by the ADPTC?LIB toolkit.The clustering results show that several spatio-temporal clustering patterns are consistent with the characteristics of temperature distribution in Asia,which can be used to assist in exploring the spatio-temporal distribution pattern and the law of geographical entities.Through the improvement of the fast discovery density peak clustering algorithm,this study provides a tool for discovering the potential rules of spatio-temporal raster data and natural grouping,which is expected to support the extraction and mining of valuable information from spatio-temporal raster data without prior knowledge.Assist in discovering the spatio-temporal evolution patterns of geographical phenomena.