盐碱土壤Philip入渗模型参数的神经网络预报模型

Philip Infiltration Parameters of Neural Network Prediction Model for Saline-alkali Soil

以改良盐碱土壤、提供入渗参数为研究目的,在山西省北部的4种盐碱荒地进行了系列入渗试验和基本理化参数测定试验。基于误差反向传播算法(Back Propagation算法),建立了盐碱地土壤基本理化参数与Philip入渗模型参数之间的神经网络预报模型。预测所得Philip入渗模型参数的平均相对误差如下:稳渗率A为4.30%、吸渗率S为0.31%,预测值与实测值吻合程度高。研究结果表明,基于盐碱地土壤条件,选择土壤体积含水率、容重、质地、有机质含量、全盐量以及p H作为预报模型输入变量,Philip入渗模型参数为输出变量的BP神经网络的预报模型是可行的。

In order to improve saline-alkali soil and provide infiltration parameters, the series of infiltration experiments and physicochemical parameter determination experiments were carried out in four kinds of saline soil in northern Shanxi Province. Based on the Back Propagation algorithm, the neural network prediction model was established between the basic physicochemical parameters and Philip infiltration parameters of saline-alkali soil. The average relative errors of the parameters predicted by BP neural network were as follows： the steady infiltration rate was 4.30% and the sorptivity rate was 0.31%. The predictive values were well coincident with the actual values. The results showed that based on the saline-alkali soil, it was feasible that selecting volumetric moisture content, density, texture, organic matter, content of salt and pH as input variable and Philip infiltration parameters as output variable in the BP prediction model.