Kostiakov入渗模型参数的预测精度对比分析研究

A Comparative Analysis and Study on Forecast Precision of Parameters of Kostiakov Infiltration Models

基于黄土高原系列大田入渗试验数据,以土壤体积含水率、干容重、黏粒含量、粉粒含量、有机质含量为输入因子,采用支持向量机和BP神经网络两种算法,对Kostiakov二参数入渗模型参数进行预测,并对两种算法下预测结果的相对误差值进行分析,结果表明:采用支持向量机算法对入渗系数和入渗指数进行预测的结果相对误差最大值和平均值都比BP算法的预测结果要小,相对误差最小值比BP算法的预测结果要大;支持向量机算法比BP算法所得预测结果的稳定性好,精确度高。研究结果丰富了采用土壤传输函数获取入渗参数这一研究方向,同时为获取更高精度的入渗参数在方法的选取上提供一定的理论依据。

Based on the land infiltration test data of the Loess Plateau, the parameters of the Kostiakov infiltration model were predicted by using two algorithms of support vector machine and BP neural network with the input parameters of the soil water content, bulk density, clay content, silt particle content and organic matter. And the relative errors of the prediction results were analyzed. The results showed that the maximum relative error value and the average relative error value of the prediction results of infiltration coefficient and infiltration index by using support vector machine were both smaller than those by using BP neural network while the least relative error value was bigger. The stability of the prediction results by support vector machine was better than that by BP neural network and the precision was higher. The results enriched the research trend of determining infiltration parameters with pedo-transfer functions and provided a theoretical basis for the selection of the method to obtain more precise infiltration parameter.