Research On The Application Of Group Teaching Model In The Prediction Of Soil Heavy Metal Content

With the industrialization of human society and the rapid development of modern agriculture,the soil will be polluted to a certain extent in the process of crop planting,especially in the accumulation of heavy metals,causing irreparable harm to farmland soil.Due to the characteristics of non degradability and easy absorption by plants,heavy metals pose a long-term threat to the earth’s ecology.After being absorbed by other organisms,agricultural products contaminated by heavy metals may eventually come into contact with human beings at the top of the food chain,or even flow into the human body,causing great harm to human health.Therefore,focusing on the prediction of soil heavy metal content can bring great significance for the future development of the earth’s ecology and human beings.In this thesis,based on 1161 data collected from farmland soil in six new urban areas of Wuhan City,after data processing,the first 500 data were used as training set,and a total of30 data from 501 to 530 were used as test set.Four common neural network models were established to predict and analyze the content of heavy metal As.The experimental results show that(1)The average value of four evaluation indexes of BP neural network model is higher than that of WNN model,and the distribution of prediction value is also better than WNN model.(2)The generalized regression neural network model(GRNN)has the best prediction effect when the expansion coefficient is 1.5,and compared with the radial basis function neural network model(RBF),GRNN model is more suitable for the prediction of soil heavy metal content in terms of prediction accuracy and model stability than RBF.Aiming at the problems of BP network in the process of training,such as easily falling into local optimal value and slow convergence speed,this thesis puts forward a predictive model in group teraching mode(GTOA-BP),optimizes the weight of BP network,and establishes the corresponding model for comparative analysis,the results show that the predictive point ratio distribution of the GTOA-BP model and the four evaluation indexes are better than the optimal BP model and the optimal GRNN model,and the prediction point ratio distribution can be maintained at 12.Compared with the other two models,the mean absolute error is 0.176 and 0.108 less,the root mean square error is 0.258 and 0.076 less,the mean absolute percentage error is 1.46% and 0.97% less,and the symmetric mean percentage absolute error is 0.97% and 0.83% less,and the fitting effect of the GTOA-BP prediction points is better than the other two models,so the GTOA-BP model has more advantages in the prediction of soil heavy metal content,and is more practical than common network models.It can provide effective help for the prevention and treatment of soil pollution in the future,and promote the development of modern agriculture.