基于BP神经网络方法的黄土水分特征曲线预测模型比选  

Comparison of Loess Water Characteristic Curve Prediction Model Based on BP Neural Network Method

作　　者：李彬楠 樊贵盛[1] 申丽霞[1] LI Bin-nan;FAN Gui-sheng;SHEN Li-xia(Taiyuan University of Technology,Taiyuan 030024,Shanxi Province,China)

机构地区：[1]太原理工大学水利科学与工程学院,山西太原030024

出　　处：《中国农村水利水电》2023年第1期171-175,共5页China Rural Water and Hydropower

基　　金：国家自然科学基金资助项目(40671081)。

摘　　要：黄土高原区水资源严重匮乏,研究土壤水分特征曲线对于提高水分利用率、节约水资源有着重要的现实意义,但直接试验测量土壤水分特征曲线面临操作技术难度大、耗时费力等诸多问题,因此对土壤水分特征曲线进行科学合理预测十分必要。为提高黄土高原区土壤水分特征曲线预测模型精度,以山西省5个县市的试验点黄土为研究对象进行模型比选。基于BP神经网络算法,以土壤基本指标黏粒含量、粉粒含量、干容重、有机质和全盐量共5个影响因素作为预测模型的输入变量,以经验模型的参数作为预测模型的输出变量,分别建立了Gardner经验模型参数和Van Genuchten经验模型参数的预测模型,并根据实测数据库的预测结果进行对比和分析。结果表明:建立的经验模型参数的BP神经网络预测模型,Gardner经验模型建模和验证后的两个参数相对误差的平均值都小于4%,Van Genuchten经验模型建模和验证后的两个参数相对误差的平均值都小于5%;不论是建模的训练数据库还是验证数据库,Gardner经验模型参数的预测模型精度均高于Van Genuchten经验模型参数的预测模型精度。因此,建议针对黄土高原区的黄土水分特征曲线预测模型的建立,选用Gardner经验模型更加合适,且此经验模型的表达式简单易懂,更利于农田水利相关的基层工作人员的学习与利用。The Loess Plateau region has a serious shortage of water resources, and the study of soil moisture characteristic curves has important practical significance for improving water efficiency and saving water resources, but the direct test of measuring soil moisture characteristic curves faces many problems such as difficult operation technology, time-consuming and laborious, so it is necessary to make scientific and reasonable predictions of soil moisture characteristic curves.In order to improve the accuracy of the prediction model of soil moisture characteristic curve in the Loess Plateau area, the loess of the test site in five counties and cities in Shanxi Province was compared for the model.Based on the BP neural network algorithm, the prediction models of Gardner empirical model parameters and Van Genuchten empirical model parameters were established respectively, and the prediction models of Gardner empirical model parameters and Van Genuchten empirical model parameters were compared and analyzed according to the prediction results of the measured database.The results show that the average value of the relative errors of the two parameters after modeling and verification of the Gardner empirical model is less than 4% in the BP neural network prediction model of the empirical model parameters established, and the average of the relative errors of the two parameters after the modeling and verification of the Van Genuchten empirical model is less than 5%. It can be seen that whether it is the modeling training database or the validation database, the prediction model accuracy of Gardner empirical model parameters is slightly higher than that of Van Genuchten empirical model parameters.At the same time, compared with the Van Genuchten experience model, the expression of the Gardner experience model is simpler, and the meaning of the parameters is easier to understand, which is also more conducive to the learning and application of grass-roots workers related to farmland water conservancy.Therefore, this paper pr

关 键 词：黄土 土壤水分特征曲线 Gardner经验模型 Van Genuchten经验模型 BP神经网络 

分 类 号：TV93[水利工程—水利水电工程] S27[农业科学—农业水土工程]