| Soil is the carrier of all life activities and carries the continuation of all kinds of life.However,a series of industrial production such as ore mining and chemical processing directly aggravate the environmental damage and lead to the accumulation of pollutants with the development of science and technology.It is the premise of soil remediation to master the distribution characteristics of site pollution.In view of the shortcomings of the previous method of assessing site pollution by using single index value,this paper proposes a model of field pollution analysis based on machine learning algorithm,which takes into account the environmental factors,in order to explore the relationship between environment and pollution characteristics,and analyze the spatial distribution characteristics of pollution from macro level.At the same time,based on the geophysical data,the pollution status of the horizontal plane and vertical plane of the site is analyzed to verify the effectiveness of the algorithm.Finally,considering the environmental factors and combining the multi-source data coupling analysis of the pollution characteristics of the chromium salt production chemical plant,it provides a reference for the relevant departments to carry out the site pollution remediation according to local conditions.The main achievements of this paper are as follows:(1)Comparative analysis of decision tree algorithms.Use random forests,GBDT and XGBoost three type decision tree algorithm to environment correlation research,the results reflected by the three calculation methods are quite different,random forests and GBDT algorithm was not able to dig out the relationship between pollution information,only highlights the importance of polluting multiples,and ignore the environmental factors for pollution state regulation.The XGBoost algorithm based on CART decision tree realizes parallel operation in the feature dimension and analyzes its structural features,and uses regularization rules to reduce the risk of model overfitting,thereby effectively mining the correlation between environmental information.(2)A pollution characteristic analysis model based on XGBoost algorithm considering environmental factors is constructed.In order to realize the comprehensive analysis at the macro level,this paper takes a chromium salt factory in East China as an example,and designs a pollution characteristic analysis model based on XGBoost algorithm according to the correlation relationship between site pollution data and the related research of machine learning.The model extracts the environmental characteristic values by calculating the correlation between the site environmental information,and integrates the improved Nemero Pollution Index Method to calculate the site pollution characteristic values that take into account environmental factors,so as to map the chemical plant soil and groundwater Cr(VI)Spatial distribution map,extract its pollution spatial distribution characteristics.(3)Cr(VI)pollution characteristics of soil and groundwater.The Cr(VI)pollution of the chemical plant showed obvious spatial heterogeneity,and the pollution decreased with increasing depth.Due to the direct or indirect contact with each link of chromite production in the shallow northeast of the chemical plant,the pollution diffuses through landfill and seepage,resulting in serious pollution in the northeast.In addition,due to the limited leakage of pollutants,there is no obvious pollution in the southwestern part of the plant and the deep-seated soil of the entire plant.(4)Verification of site pollution feature machine learning algorithm based on local geophysical data.The results show that the resistivity of the contaminated soil is obviously different from that of the normal original soil,and the resistivity range of the contaminated soil is below 3 Ohm·m.Cr(VI)pollution is mainly concentrated in the soil depth of 0-10 meters.The verification results show that the machine learning algorithm can effectively mine the correlation between pollution and environment,and consider the restrictive factors of pollution diffusion from the multi-dimensional level,which has an important application value for the analysis of site pollution characteristics. |