Study Of Optimization Algorithm About Spherical Model Of Semivariogram In Geostatistics

In order to describe accurately space randomness and structural about natural phenomenonsuited for regionalized variables theory, aim to the problems of the weighted linearprogramming, considering space anisotropic, proposed the weighted quadraticprogramming and the combining algorithm constituted of the genetic algorithm and thepattern search, optimized spherical model with double structures, contrasted the fittingeffect of three optimization model by concrete example.Firstly, devided experimental datas about spherical model with double structures into twoand a half by the weighted linear programming, considered the weight based on thereciprocal of lag distance and fitted separately. I got the conclusion that many variablereplacement increase mapping error, the weighted linear programming only has a betterfitting effect for some datas, a worse fitting effect for the whole datas, it can not describespace variability accurately.Secondly, devided experimental datas about spherical model with double structures intotwo and a half by the weighted quadratic programming, considered the weight based on thereciprocal of lag distance and fitted separately. Comparing the weighted quadraticprogramming with the weighted linear programming, I got the conclusion that it has abetter fitting effect for some datas, a best fitting effect for a few datas. therefore, it reducesthe overall fitting errors but it can not describe space variability accurately.Thirdly, fitted experimental datas about spherical model with double structures by thegenetic algorithm and got a better fitting effect, Considered the fitting result as a initialpoint and optimized again by the pattern search method, I got the conclusion that it isimpossible that the combining algorithm is only suitable for some datas and is not suitalbefor another datas. It devided the error into each data segment and reduce the global error. Inaddition, it won’t ignore spatial variation in small scale and describe spatial informationaccurately.The iterations and consumed resource of the weighted linear programming and theweighted quadratic programming are less than the combining algorithm constituted of the genetic algorithm and the pattern search method. But from the fitting effect, the combiningalgorithm is better than the weighted linear programming and the weighted quadraticprogramming.The weight is beneficial to the fitting precision of the theoretical model of semivariogram,the computing methods about the weight contain two methods. One weight is based on thereciprocal of lag distance. Another is based on the quantity of sampling point. Optimizedspherical model based on the different weighted by the combining algorithm constituted ofgenetic algorithm and pattern search and contrasting their fitting effect, I got theconclusion that the fitting precision of the optimization algorithm based on the weight ofthe reciprocal of lag distance is better than the optimization algorithm based on the weightof the quantity of sampling point.