基于灰色理论—BP神经网络方法的表层土壤容重预测

Prediction of Surface Soil Bulk Density Based on Grey Theory and BP Neural Network

以山西省黄土高原区15个试验点的年度跟踪监测样本为依据,利用灰色关联理论与BP神经网络相结合的方法,构建了表层土壤容重与土壤基本理化参数和累积接收水量之间的土壤传输函数预报模型。结果表明:影响表层土壤容重的7个土壤理化参数对于土壤容重的关联度均大于0.6;基于BP神经网络,以关联度较高的土壤粉粒含量、土壤砂粒含量、累积接收水量、体积含水率、有机碳含量和全盐量作为输入变量对表层土壤容重进行预测是可行的,预测值与实测值高度吻合,预测结果相对误差的平均值为0.41%,预测精度较高;检验样本预测结果相对误差的平均值为1.05%,误差完全在可接受范围内。研究结果可为黄土区土壤容重的获取提供新思路,为科学指导农田农事和灌溉管理提供理论支撑。

Based on the annual monitoring samples from the fifteen test points in Loess Plateau area of Shanxi Province,a soil transfer function prediction model between surface soil bulk density and soil physicochemical parameters and accumulated receiving water is established by combining grey correlation theory and BP neural network. The results show that the correlation between surface soil bulk density and 7 physicochemical parameters affecting surface soil bulk density is greater than 0.6; Based on BP neural network; it is feasible to select soil silt content,sand content,accumulated receiving water,organic carbon content,volumetric water content and total salt content as input variables to predict the surface soil density,and the predicted values are in good agreement with the measured values; the average relative error of the prediction is 0.41% and the prediction accuracy is high; the average relative error of the test sample prediction results is1.05%,the errors are acceptable totally. The research results can provide new ideas for obtaining soil bulk density in loess region,and provide theoretical support for scientific guidance of farmland farming and irrigation management.