Research And Application Of Spatial Data Analysis Based On Kriging And Support Vector Regression

With the progress and development of science and technology,a large amount of data is collected and organized in specific spatial locations.Through the research and analysis of spatial data,we can better measure the degree of influence of different spa-tial positions on a certain point.In addition,spatial data also has the characteristics of spatial correlation and spatial heterogeneity.The common spatial interpolation method is Kriging.Since Support Vector Machine(SVM)is suitable for both large sample data and small sample data,and can avoid problems such as dimensional disasters.SVM has at-tracted more and more scholars' attention,and SVM has been widely used in the research of classification and regression problems and so on.At present,some scholars have com-bined Kriging and support vector machine for research.According to the decomposition of regionalized variables,it can be understood as the phenomenon observed at different scales,and referring to the practices of others.When selecting the ? value of the support vector machine,controlling it within the square root of the sill value can better use spa-tial information.In this paper,a two-scale SVM model based on Kriging optimization is proposed and applied to the study of precipitation data in Xinjiang.Firstly,this paper builds a two-scale SVM model based on Kriging optimization,where the two-scale SVM model is reflected in the selection of a two-scale kernel func-tion,and in practical applications,two different combinations of kernel functions are selected as the kernel function of this model.Secondly,the variogram function is solved based on the existing data,and the value of ? is controlled within the square root of the sill value,and then the optimal penalty parameter and the kernel parameters and the value of ? are selected by the 10-fold cross-validation method.Finally,the required Lagrangian multiplier is obtained by solving the quadratic programming problem,and the threshold in the model is further solved.Substituting the Lagrangian multiplier and the selected kernel function and threshold into the solved regression model can obtain the correspond-ing fitting value.After that,a simulation experiment is designed and compared with the support vector regression model fitting result by calculating the mean square error(MSE)and the goodness-of-fit value R2.This paper uses a two-scale SVM model based on Kriging optimization to fit and analyze the precipitation of Xinjiang's 66 observation stations from 2013 to 2016.By selecting the six meteorological elements of longitude,latitude,average wind speed,av-erage air pressure,average air temperature,and average humidity,a two-scale SVM model based on Kriging optimization for precipitation was constructed.The results show that when the kernel function is selected properly,the two-scale SVM model based on Kriging optimization improves the accuracy of precipitation fitting at each station than the support vector regression model,and provides a new idea for the study of precipitation prediction(interpolation)at unknown stations.