考虑黄土结构变形的Philip入渗模型参数预报

摘要

基于黄土高原区区域尺度大田耕作土壤的水分入渗试验，考虑黄土备耕头水地土壤结构的变形特性，建立了Philip半理论半经验入渗模型参数的BP神经网络预报模型，实现了以土壤基本理化参数为输入变量、Philip模型参数为输出变量的BP预报。对Philip模型中吸湿率S、稳渗率A以及90 min累计入渗量的预测值与实测值进行比较，结果显示：吸湿率 S的平均相对误差为1．41％，稳渗率 A的平均相对误差为2．81％，90 min累计入渗量的平均相对误差为1．75％，三者的平均相对误差值均在3％以下，预测精度很高。这表明以土壤基本理化参数为输入变量的BP神经网络预测是可行的，考虑备耕头水地土壤结构变形使得BP预报结果更符合土壤水分入渗实际状况。%Based on water infiltration test in field farming soil of Loess Plateau region scale and considering the soil structure deformation characteristics of head plowing soil in loess, this paper established BP neural network prediction model based on Philip half experience and half theory infiltration model which the soil physicochemical parameters were as input variables and the parameters of Philip infiltration model were as output variable. The accuracy comparison was carried out among moisture absorption rate S, steady infiltration rate A and 90 min cumulative infiltration I. The results show that the average relative error of S in BP prediction model is 1.41%, the average relative error of A is 2.81% and the average relative error of 90 min cumulative infiltration I is 1.75%. The three average relative errors are all under 3%, so the BP neural network prediction model has a relatively high prediction accuracy and effect. It′s feasible to established BP neural network prediction model based on soil physicochemical parameters;the BP prediction model is more consistent the actual state of water infiltration by considering soil structure deformation.