| Soil plays an important role in agricultural safety production and human life.As an important rice producing area in China,and also an important non-ferrous metal producing area,there are serious problems of comprehensive prevention and control of heavy metal pollution in soil rice system in Xiangjiang River Basin and Dongting Lake region of Hunan Province.Due to the dynamics and complexity of soil ecosystem,how to effectively analyze and predict the content,migration characteristics,spatial distribution and risk of heavy metals in soil and rice is an important issue in the field of soil ecology.In recent years,with the rapid development of artificial intelligence theory and technology,using artificial intelligence and machine learning methods and tools to study heavy metal pollution in soil rice system has become an important development direction.Therefore,in view of the difficulties and challenges existing in the existing analysis and assessment methods of heavy metal pollution in soil rice system,based on the advanced machine learning methods and technologies such as over limit machine learning,support vector machine and unsupervised learning,this study collects soil and rice samples from Xiangjiang River Basin and Dongting Lake area in Hunan Province,and analyzes the ecological risks and scores of soil heavy metal pollution Research on intelligent analysis,prediction and evaluation of cloth features.The goal is to overcome the shortcomings of the existing research work in the prediction and transfer characteristics of heavy metal content in soil rice system,ecological risk assessment,spatial distribution characteristics analysis,etc.,such as low modeling accuracy,weak prediction and promotion ability,and provide efficient methods and technical means for the soil rice system pollution protection in Xiangjiang River Basin and Dongting Lake region of Hunan Province.The main results of this study are as follows:(1)Based on the comprehensive statistical analysis of Cu,Zn,Cd,Cr,Pb,As and other heavy metals in rice plants and Rhizosphere paddy soils from 80 sampling sites in Dongting Lake area,a health risk assessment model based on the target hazard coefficient(THQs)was established.It was found that the average THQ of arsenic and cadmium in adults and children were significantly higher than 1.0,5.17 and 3.61 for adults and 4.49 and 3.14 for children,respectively.The results show that there are health risks for adults and children in Dongting Lake.It is proposed that machine learning methods such as neural network and support vector machine can be used to analyze the heavy metal migration characteristics of soil rice system.By using machine learning method to predict and model the heavy metal content in soil and rice,we can clarify the heavy metal migration characteristics between soil and rice.The experimental results show that the neural network and support vector machine method can effectively predict the Cd amount in rice roots through the amount of Cd in the soil.The mean square error of the test samples is 0.1025 and 0.0625 respectively,which verifies that the metal Cd has strong transfering ability from the soil to rice roots.(2)The risk assessment method of heavy metal pollution using extreme learning machine(ELM)is proposed,and the whole situation and risk of heavy metal pollution in Xiangjiang River Basin are modeled and analyzed intelligently.By collecting the content data of Cu,Zn,Cd,Cr,Pb,As and other heavy metals in the farmland soil of Changzhutan(Changsha,Zhuzhou,Xiangtan),a heavy metal pollution risk assessment model based on the over limit learning machine was established,in which the pH value range of the soil was between 4.58 and 8.09,and the content range of Pb in the soil was 9.7mg·kg-1-194.8 mg·kg-1,with a large range of change;the content of Cd was relatively large The content range of Cr is 14.9 mg·kg-1-126 mg·kg-1,the content range of Hg is 0.03 mg·kg-1-0.433 mg·kg-1,and the content range of As is 2.42 mg·kg-1-57.83 mg·kg-1.Through the experimental verification,the average prediction accuracy of BP neural network for the evaluation standard of soil heavy metal pollution in 1995 is 94.258%,and the average prediction accuracy for the evaluation standard of soil heavy metal pollution in 2018 is 94.807%;the average prediction accuracy of the ELM algorithm for the evaluation standard of soil heavy metal pollution in1995 is 96.645%,and the average prediction accuracy for the 2018 standard is 92.9%.Compared with other traditional methods of soil environmental quality assessment,BP neural network and ELM method can simplify the problems and better solve the soil environmental quality assessment problems of multi input and multi output classification,so as to provide new intelligent decision support means for heavy metal pollution control,risk management and pollution management of rice in this area(3)A fuzzy support vector machines(FSVM)method with the ability of semantic rule extraction is proposed.This method has both the ability of accurate modeling of SVM and the ability of semantic modeling of fuzzy reasoning system.It can effectively establish the analysis and prediction model of effective heavy metal content in soil rice system.The experimental results of the soil heavy metal pollution data in Zhuzhou County,Hunan Province show that the method proposed in this chapter has a high-precision prediction ability of soil available heavy metal content,and the RMS prediction error in the test sample data set is 0.0078,which is superior to the multilayer neural network prediction method(the RMS prediction error in the test sample data set is 0.0164)and traditional support Vector machine method(RMS prediction error on the test sample data set is 0.1255),and the established prediction model overcomes the semantic interpretability and semantic reasoning ability that the traditional support vector machine model does not have.(4)An improved unsupervised clustering learning algorithm is proposed and applied to the spatial distribution modeling of heavy metal pollution in soil rice system in Dongting Lake area.The data of heavy metal content in soil rice system of Dongting Lake area were sampled,analyzed and tested.For 35 samples of early rice grain,the clustering algorithm proposed in this paper can quickly give two kinds of spatial distribution characteristics,which are more polluted and less polluted,including 30 samples with less pollution and 5samples with more pollution.For 37 samples of middle rice grain,the clustering algorithm proposed in this paper can quickly give two kinds of spatial distribution characteristics,including 34 samples with less pollution and 3 samples with more pollution.The unsupervised clustering learning algorithm proposed in this chapter overcomes the shortcomings of the existing spatial density analysis method and spatial difference method,which rely on prior information,have too many artificial parameters,and have unstable performance.It can realize the rapid spatial distribution modeling of heavy metal pollution in local soil rice system,and is important for the spatial situation analysis and scientific prevention and control of heavy metal pollution in soil.In summary,the research work of this paper will promote the application of machine learning method in heavy metal pollution analysis and risk assessment of soil rice system,improve the accuracy,adaptability of heavy metal pollution analysis and the degree of intelligence of risk assessment.In addition,on the basis of the research results of this paper,it is necessary to further study and solve the machine learning method of heavy metal pollution analysis with noise data,semi supervised learning when the number of labeled samples is insufficient,large-scale pollution analysis and risk assessment combined with geographic information system theory and key technologies,so as to provide comprehensive and efficient technology for agricultural soil ecological protection in China. |
PDF Full Text Request