基于灰色理论-BP神经网络方法的土壤水分特征曲线预测模型  

Prediction of the soil water characteristic curve based on Gray-BP neural network

作　　者：李彬楠 樊贵盛[1] LI Binnan;FAN Guisheng(Taiyuan University of Technology, Taiyuan 030024, Chin)

机构地区：[1]太原理工大学水利科学与工程学院,太原030024

出　　处：《干旱区资源与环境》2018年第7期166-171,共6页Journal of Arid Land Resources and Environment

基　　金：国家自然科学基金项目(40671081);山西省农田节水技术开发服务推广站项目资助

摘　　要：以黄土高原区土壤为研究对象,通过土壤基本理化参数与土壤水分特征曲线的系列试验,获得了Van-Genuchten模型参数的数据样本。运用灰色理论对土壤基本理化参数进行了灰色关联度分析,建立了以土壤基本理化参数为输入变量,土壤水分特征曲线Van-Genuchten模型参数为输出变量的BP神经网络预测模型。研究结果表明:以土壤黏粒含量、粉粒含量、容重、有机质含量、全盐量为输入变量,运用BP神经网络方法对土壤水分特征曲线Van-Genuchten模型参数进行预测是可行的。所建立的灰色BP神经网络预测模型下,Van-Genuchten模型参数α与参数n的预测值与检验值平均相对误差都小于5%,建模样本和检验样本都具有较高的精确度。研究成果一方面有助于丰富黄土水力参数的理论研究,另一方面为土壤水分特征曲线的获取提供技术支撑。Taking soil of Loess Plateau for research objects,the samples of the parameters of Van-Genuchten were obtained from soil physical and chemical parameters and measurement of the soil water characteristic curves. The soil physical and chemical parameters were analyzed by the correlation method of Grey system theory,and the BP neural network prediction model was established by taking basic soil physicochemical parameters as input variables and the parameters of Van-Genuchten as output variables. The results showed that taking the content of soil clay,silt,organic matter,total salt and bulk density as input variables to predict the parameters of Van-Genuchten was effective for BP neural network. The average relative error of the parameter of Van-Genuchten was less than 5% for the Gray-BP neural network model. The results of research not only redound to extend theoretical research of loess hydraulic parameters,but also provide technical support for obtainment the soil water characteristic curve.

关 键 词：土壤水分特征曲线 土壤理化参数 Van-Genuchten模型参数 灰色理论 BP神经网络 

分 类 号：S152.7[农业科学—土壤学]