Study On Prediction Model Of Loess Water Characteristic Curve Based On Soil Physical And Chemical Parameters

Based on the second phase project of water-saving irrigation with loan from the World Bank of Shanxi Province,this paper bases on 165 groups of soil physical and chemical parameters and water characteristic curves measured in 16 experimental sites in 6 counties(cities and districts)of Shanxi Province in the loess plateau region.The pseudo sample database of soil physical and chemical parameters and water characteristic curve Van Genuchten model parameters was established.The basic physical and chemical parameters(clay content,powder content,dry bulk density,organic matter content,total salt content)of loess were taken as the input variables of the prediction model,and the parameters ? and n of Van Genuechten model were taken as the output variables of the prediction model.The regression models of multivariate nonlinear,grey-BP artificial neural network,grid search and crossverification-support vector machine(SVM)are carried out,and the error analysis of the modeling samples and the measured verification samples of the prediction model is carried out.The main results and conclusions are as follows:(1)taking the basic physical and chemical parameters of loess(clay content,powder content,dry bulk density,organic matter content,total salt content)as input variables,the prediction model of Van Genuchten parameters is established.That is,the idea of soil transfer function is feasible.(2)the multivariate nonlinear of parameter ? and parameter n of Van Genuechten model based on loess texture(clay content,powder content),dry bulk density,organic matter content and total salt content.Grey-BP artificial neural network and grid search and cross-verification-support vector machine are feasible,and the relative error of prediction accuracy is 0.89% ~ 10.01%,all of which are in the acceptable range.(3)Grey-BP artificial neural network,grid search and crossverification support vector machine,multivariate nonlinear prediction model can be used to predict the parameters of loess water characteristic curve model.According to the accuracy of various types of models built by parameter ? and parameter n of Van Genuechten model,the average relative error of grey-BP artificial neural network is the lowest and less than 5%.The average relative error of grid search and cross-verificationsupport vector machine is less than 6%,and that of multivariate nonlinear prediction model is less than 10%.In the 15 groups of test samples,the average relative error of grey-BP artificial neural network is less than 3%,and the average relative error of grid search and cross-verificationsupport vector machine is less than 5%.The average relative error of multivariate nonlinear is less than 6%.The results show that the three prediction models have high accuracy and can be used to predict the parameters of loess water characteristic curve model.(4)the grey-BP artificial neural network prediction model has the highest accuracy and the best prediction effect.According to the prediction model,the predicted value of the parameters is obtained,and the predicted value is brought back to the empirical formula of Van Genuechten model of soil moisture characteristic curve.Then the relative error between the predicted volume moisture content and the measured volume moisture content is calculated to determine the comprehensive accuracy of the model.The synthesis relative error of multivariate nonlinear is 2.384%,the synthesis relative error of grey-BP artificial neural network is 0.514%,and the relative error of grid search and crossverification-support vector machine is 1.207%.The results show that the grey-BP artificial neural network prediction model has the highest accuracy and the best prediction effect.To sum up,through the comparison and analysis of the parameter prediction models of various types of soil moisture characteristic curve Van Genuechten model,it can be seen that the grey-BP artificial neural network prediction model has the highest accuracy and the best prediction effect.However,the grey-BP artificial neural network prediction model is more complex,relatively inconvenient to apply,but also can choose a simple form of easy to understand and master the multivariate nonlinear prediction model.Therefore,the prediction model of loess water characteristic curve based on basic physical and chemical parameters of loess in this paper can provide a method for different levels of scientific and technological personnel and managers to obtain loess water characteristic curve model.It also promotes and enriches the development of soil parameter transfer function theory.At the same time,there are still some shortcomings in the selection of input parameters in this paper,which should be further improved in the future research.In addition,the relevant guiding theories of the model should be further studied and deepened in order to achieve more stable and excellent prediction results.