Analysis And Evaluation Of Various Interpolation Methods Based On Drilling Soil Heavy Metal Data

In the process of national urbanization,the demand for urban land continues to expand.The original industrial enterprises gradually move out of the urban area,and their sites are converted into urban construction land or other land.However,due to industrial production and other reasons,the soil and groundwater in the site are polluted by heavy metals.Affect the regional ecological environment.In order to obtain the contaminated area and the remediation range,it is necessary to sample and analyze the contaminated area and interpolate to determine its spatial structure,and then realize the treatment and remediation of the contaminated field area.Constructing a high-precision interpolation method has practical significance and practical value.This article takes a decommissioned medical chemical plant in Taizhou City,Zhejiang Province as the research object,using Ordinary Kriging(OK)and Inverse Distance Weighted(IDW)three weights(p=2,p =3,p=4),and Radial Basis Function Neural Network(RBFNN)three interpolation methods to interpolate and analyze the concentration data of five heavy metals in the region: Cr,Cu,Ni,Pb,and Zn.Evaluation.For the conditions of the Kriging method,the use of Cox-Box to transform the original data can make the data approximate to a normal distribution,and the variance function of 5 metals is fitted.The results are Cr,Ni,Pb,and Pb in the5 metals.Zn is a spherical model,Cu is an exponential model;IDW method uses 20 as the number of neighborhood search points,and interpolates the three weights of IDW(p=2,p=3,p=4),and the results show three weights,The accuracy is p=2,p=3,p=4;the RBFNN method uses K-means clustering to determine the hidden layer center during interpolation and fitting.The number of centers k is calculated using the elbow method,and the Gradient descent trains its output layer parameters,and it is found through experiments that the variability of its original data has a greater impact on the loss function of the model training.For this reason,the three methods are compared through K-fold cross-validation.The value of k is 10,and Root Mean Square Error(RMSE),Mean Absolute Error(MAE)and average relative error(Mean Relative Error,MRE)three indicators for accuracy evaluation.The experimental results show that RBFNN shows high accuracy under the three accuracy indicators,which proves that RBFNN is the best among the three methods for heavy metal data in this field,and ordinary kriging is better than IDW.Three-dimensional modeling analysis was performed on the interpolation results of the three methods.The results showed that the Kriging method can show high accuracy when the data meets the assumptions.The IDW result is poor in spatial performance,and the RBFNN interpolation result is more spatially.It has continuity,and its model performs well on the spatial characteristics of the data.