| As China attaches more and more importance to the problem of soil pollution,a large number of industrial enterprises have been closed or relocated,and the left over site has become an important land resource in the urbanization construction.However,due to historical reasons,the soil in some sites is polluted to different degrees.In order to avoid threats to human health,it is necessary to repair and treat the contaminated sites before development and utilization.Site investigation and pollutant analysis in the early stage are the necessary ways to find out the spatial distribution characteristics of pollutants,the range of remediation and the amount of remediation soil,which is of great significance to the remediation and treatment and safe reuse of contaminated sites in the later stage.In this paper,the residual site of a sulfuric acid plant in Henan province was taken as the research object.The spatial distribution characteristics of heavy metal arsenic in the site were analyzed by KFCM-RBF neural network method.The 3D visualization software was developed to show the pollution status and calculate the range of restoration and the amount of remediation soil.The main research results are as follows:(1)Interpolation study of spatial distribution of heavy metal pollution in soil based on 3D kriging method.Based on the sample data conforming to the normal distribution,the scatter map of the experimental variation function was drawn,and the scatter map was fitted with four theoretical variation function models: linear,spherical,exponential and gauss.The heavy metal content of the unknown soil was predicted by the optimal theoretical variation function model.The experimental results showed that the gauss model was the best theoretical variation model,followed by the exponential model and the spherical model,and the linear model was the worst.(2)A prediction model of spatial distribution of heavy metal pollution in soil based on fuzzy neural network is proposed.On the basis of Radial Basis Function(RBF)neural network and with the purpose of improving the network generalization ability,kernel fuzzy c-means clustering(KFCM)algorithm was used to obtain the value of radial basis center vector in the hidden layer of the network.The network width value was determined by the fixed method,and the weight value was optimized by the gradient descent method.Different radial basis functions was used to train the network and the optimal radial basis function was selected through precision comparison analysis,then the heavy metal content of unknown soil was predicted.The experimental results showed that the optimal radial basis function was the gauss function,and the error accuracy of the three functions was: gauss function > reflected sigmoidal function > inverse multi-quadrics function.(3)Verified the prediction ability of KFCM-RBF neural network for the spatial distribution of soil heavy metal pollution.The prediction results of KFCM-RBF neural network were compared with those of 3D kriging and RBF neural network.By establishing the regression equation between the predicted value and the measured value,the fitting ability of the three interpolation methods was compared.In addition,cross validation(CV)was adopted to calculate the precision evaluation index parameters which contained Root Mean Square Error(RMSE),Mean Absolute Error(MAE)and Mean Relative Error(MRE).The comparison results showed that the KFCM-RBF neural network not only had the best fitting ability,but also had the smallest value of the three evaluation indexes and the highest accuracy.(4)A 3D visualization system of soil heavy metal pollution based on interpolation was developed.Based on the above research,the 3D visualization system of soil heavy metal pollution based on interpolation was developed by using MVC model and AE.The functions included basic map function,pollution concentration interpolation,data setting and pollution profile display.And the software had obtained the software copyright. |
PDF Full Text Request