Research On Prediction Of Soil Heavy Metal Content Based On Neural Network Optimization Model

An accurate understanding of the detailed information of heavy metal content in soil can be more effective in the control of soil heavy metal pollution.Nevertheless,in the actual detection process,owing to the limitations of human and material resources and other factors,it is difficult to detect all areas to be inspected,and the prediction effect is not satisfactory.In this study,the heavy metal content of farmland soil in six new urban areas of Wuhan was taken as the research object,and an optimization model based on a radial basis neural network(RBFNN)was proposed for soil heavy metal content prediction,which can effectively improve the accuracy of prediction.The research content of this study is as follows.(1)The center point and width vector of the RBFNN hidden layer are generated by the traditional K-means clustering algorithm;however,the K-means clustering algorithm cannot ignore the sensitivity to outliers,which can easily to reduce the effect of the algorithm.Therefore,an efficient density peak clustering(EDPC)algorithm was proposed to determine the central point of the RBFNN hidden layer.An adaptive variance measure(AVM)is proposed to calculate the width vector of the hidden layer.At the same time,the clustering effect of several clustering algorithms was compared under the same experimental conditions,and the experimental results showed that the clustering effect of the EDPC algorithm was better.(2)Traditional genetic algorithms have the problems of prematurity and low local search accuracy.Based on the traditional genetic algorithm,the adaptive dynamic genetic optimization algorithm(ADGOA)was improved.It was used to generate the initial weight and threshold of the output layer of the radial basis neural network.Simultaneously,the convergence accuracy and speed of several intelligent algorithms for the test function were compared under the same experimental conditions.By comparing and analyzing the experimental results,it can be concluded that ADGOA algorithm has higher convergence accuracy and speed.(3)By combining the radial basis neural network optimized by the EDPC and AVM algorithms with the ADGOA algorithm,a dynamic neural network optimization model(DNNOM)was generated.Subsequently,under the same experimental conditions,the model was compared with several common soil heavy metal content prediction models.The radial basis neural network(RBFNN),genetic algorithm optimized radial basis neural network(GA-RBFNN),support vector machine(SVM),and light gradient boosting machine(Light GBM)were used for comparison experiments.The validity of the prediction model proposed in this study was verified.