基于土壤基本理化参数的土壤水分特征曲线Van-Genuechten模型预报研究

发布时间：2018-04-09 09:42

本文选题：土壤水分特征曲线　切入点：Van-Genuechten模型参数　出处：《太原理工大学》2017年硕士论文

【摘要】：土壤水分特征曲线是表征土壤水吸力(基质势)与土壤含水率关系的曲线,表示土壤水能量和数量之间的关系。作为反映基本土壤水力特性的指标,土壤水分特征曲线对研究土壤水分的运移与滞留具有重要作用,尤其对农业灌溉管理具有重要意义。但人们,尤其是基层水利工作者直接测定土壤水分特征曲线存在费时费力,技术不达,土壤类型空间变异性大等问题,故本研究从土壤传输函数的理论出发,通过测定较易的土壤基本理化参数来预测得到较难获取的土壤水分特征曲线模型参数,实现对土壤水分特征曲线的预测。因此实现对土壤水分特征曲线Van-Genuechten模型参数的预测为本研究的核心内容。本研究基于黄土高原区土壤水分特征曲线系列室内试验,选取精度较高、适用广泛的Van-Genuechten模型作为拟合模型,运用RETC软件和MATLAB程序进行拟合,拟合Van-Genuechten模型参数α与n,并同步测试与Van-Genuechten模型参数α与n相对应的土壤理化参数,构建由Van-Genuechten模型参数和土壤理化参数组成的数据样本。基于历时一年以上的试验过程所建立的108组数据样本,在分别分析每个基本理化参数与Van-Genuechten模型参数α与n之间的定性定量关系的基础上,建立了以土壤基本理化参数作为输入变量,Van-Genuechten模型参数α与n作为输出变量的非线性模型预报、BP神经网络模型预报和支持向量机模型预报研究。主要结果与结论如下:(1)影响土壤水分特征曲线的主要基本理化因素有土壤质地、干容重、有机质含量和盐分含量。通过单因素分析,分别确定了Van-Genuechten模型参数α与n与各个土壤基本理化参数的函数关系。参数α:与质地呈线性关系,与干容重呈对数关系,与有机质含量呈指数关系,与盐分含量呈对数关系;参数n:与质地呈线性关系,与干容重呈线性关系,与有机质含量呈对数关系,与盐分含量呈对数关系;(2)通过常规土壤理化参数对Van-Genuechten模型参数进行预测是可行的,所建立的Van-Genuechten模型参数α与n的非线性、BP和支持向量机预报模型误差都在可接受范围,都是可行的。以Van-Genuechten模型参数α与n为输出变量的非线性、BP和支持向量机预报都以土壤质地、干容重、有机质含量和盐分含量为输入变量,都基于统一的数据样本,包括100组建模样本和8组验证样本。非线性预报模型:参数α与n建模样本的平均相对误差分别是10.23%和7.20%,验证样本的平均相对误差分别是7.65%和5.77%;BP神经网络预报模型:参数α与n建模样本的平均相对误差分别是1.52%和0.67%,验证样本的平均相对误差分别是1.01%和0.28%;支持向量机预报模型:参数α与n建模样本的平均相对误差分别是8.32%和7.48%,验证样本的平均相对误差分别是6.99%和4.54%。这三种预报模型都具有本身特点与规律,本文研究中均取得较为理想的预报效果,较好的建立了土壤水分特征曲线模型参数的土壤传输函数。(3)非线性预报模型或许是首选模型。通过对比非线性、BP和支持向量机预报模型结果,发现非线性预报模型形式简单,物理意义十分明确,但精度稍有逊色;BP预报模型精度很高,可以精准地实现Van-Genuechten模型参数预测,但模型形式复杂,其程序对使用人员要求高;作为首次引入土壤水分特征曲线模型参数预测的支持向量机模型,其精度符合参数预报的要求,且具有理想的预报效果,但支持向量机模型理论繁复,深度较高,在实践应用上可能要逊色于非线性和BP预报模型。因此,综合推荐使用非线性和BP预报模型,非线性预报模型或许是首选模型,具体使用何种预报模型,应根据不同的使用境况合理选取。本文实现了土壤水分特征曲线模型参数的非线性、BP、支持向量机三种模型的预测,但在输入变量的选取方面可能存在不全面性。在今后的研究中还应进行进一步的探讨,以进一步提高预测精度。此外,对于支持向量机模型的研究应继续深入,以进一步提高预报精度。
[Abstract]:The soil water characteristic curve is the characterization of soil water suction (matrix potential) and the relationship between soil moisture curve, the relationship between soil water energy and quantity. As a reflection of the hydraulic characteristics of basic soil index, migration and hysteresis of soil water characteristic curve of soil moisture retention plays an important role, especially has important significance to the agricultural irrigation management. But people, especially the grassroots water conservancy workers direct determination of soil water characteristic curve are time-consuming, the technology of soil spatial variability and other issues, so this study from the soil transfer function theory, through the determination of easily soil basic physical and chemical parameters to predict the parameters of soil water characteristic curve model difficult to obtain, to predict the soil water characteristic curve. It is used to predict the parameters of soil water characteristic curve of Van-Genuechten model for the research The core content of the research. This research is based on the Loess Plateau soil water characteristic curves of a series of indoor tests, selection of high precision, wide application of the Van-Genuechten model as the fitting model were fitted using the RETC software and the MATLAB program, n alpha and Van-Genuechten model parameters, physicochemical parameters a and n and synchronous test and Van-Genuechten model parameters corresponding to the soil sample data constructed by Van-Genuechten model parameters and soil physicochemical parameters. 108 samples of test process lasted more than one year on, respectively based on qualitative and quantitative analysis of relationship between the alpha and n parameters and Van-Genuechten model parameters of each basic physicochemical on the established soil physicochemical parameters as input variables, the parameters of Van-Genuechten model and N model as the alpha prediction of nonlinear output variables, BP neural network prediction model Support vector machine model and prediction research. The main results and conclusions are as follows: (1) the main factors affecting the soil water characteristic curve of the basic physical and chemical soil texture, bulk density, organic matter content and salt content. Through single factor analysis, the function relationship between Van-Genuechten model basic physicochemical parameters parameters alpha and N with various soil the parameters are determined. Alpha: a linear relationship with the logarithm relationship with the texture, dry bulk density, and exponential relationship between the content of organic matter, the logarithm relationship with salt content; there is a linear relationship between the parameters of n: and texture, showed a linear relationship with the logarithm relationship with dry bulk density, organic matter content, the logarithm relationship with salt content; (2) by conventional soil physicochemical parameters to predict the parameters of the Van-Genuechten model is feasible, nonlinear alpha and n parameters of Van-Genuechten model established by BP, and the support vector machine model prediction errors are acceptable The range is feasible. The alpha and n parameters of Van-Genuechten model for nonlinear output variables, BP and support vector machine prediction with soil texture, dry bulk density, organic matter content and salt content as the input variables are based on the unified data samples, including 100 form appearance and 8 groups of nonlinear test samples. The average forecast model parameters a and N modeling sample relative error are 10.23% and 7.20%, the average relative error of testing samples were 7.65% and 5.77%; BP neural network prediction model, the average relative error of parameter modeling and N samples respectively is 1.52% and 0.67%, the average relative error of testing samples were 1.01% and 0.28%; support vector machine prediction model: the average parameter alpha and N modeling samples with the relative error are 8.32% and 7.48%, the average relative error of testing samples were 6.99% and 4.54%. of the three prediction model has the Personal characteristics and rules, prediction of ideal effect were obtained in this study, the soil transfer function well established parameter model of soil water characteristic curve. (3) the nonlinear prediction model may be the preferred model. By comparing the nonlinear BP and SVM prediction model results, found that the form of nonlinear prediction model is simple, physical the meaning is clear, but the precision is slightly inferior; the accuracy of BP prediction model is very high, can accurately realize the Van-Genuechten model parameter prediction model, but its complex forms, procedures on the use of personnel requirements; as for the first time into the model of soil water characteristic curve parameter prediction model of support vector machine, its accuracy meets the requirement of prediction parameters. The forecast effect and has the ideal, but the model of support vector machine theory complicated, higher depth may be inferior to BP and nonlinear prediction model in practice. This comprehensive, recommended the use of nonlinear and BP prediction model, nonlinear prediction model may be the preferred model, what the specific use of prediction model should be reasonably selected according to the different use condition. This paper realizes nonlinear, parameter model of soil water characteristic curve of BP prediction, support vector machine model three, but there may be not comprehensive in the selection of input variables. In the future research should be conducted to further explore, to further improve the prediction accuracy. In addition, for the study of support vector machine model should be further, to further improve the prediction accuracy.

【学位授予单位】：太原理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：S152.7

【参考文献】