摘要
冻融循环作用下盐碱土的渗透规律十分复杂,在含水率、干密度和孔隙度的影响下难以评价盐碱土的渗透系数与众影响因素间的关系,而传统的渗透系数试验存在些许不足和不可确定因素。基于此,以山东省西南部黄河泛滥故区盐碱土为材料,试验土壤按照13%,16%,19%三组含水率梯度,1.48 g·cm^(-3),1.53 g·cm^(-3),1.58 g·cm^(-3)三组干密度梯度重塑,经0,1,5,10,15,20次冻融循环后得到累计54组土壤样品,利用饱和导水率试验获得土壤渗透系数,利用压汞试验获得土壤孔隙度和区间孔隙体积分布,后通过BP神经网络预测土壤渗透系数,探究试验地区不同初始条件对盐碱土渗透系数的影响。通过神经网络的误差分析和拟合优度对比看出,利用神经网络进行试验区盐碱土渗透系数预测是可行的,该模型能充分预测不同初始条件盐碱土的渗透系数。
The permeability law of saline-alkali soil under freeze-thaw cycle is very complex, and it is difficult to evaluate the relationship between the permeability coefficient of saline-alkali soil and these factors under the influence of water content, dry density and porosity, while the traditional permeability coefficient test has some deficiencies and uncertain factors. Based on these, three groups of remolded soils with water content gradient of 13%,16% and 19% and 1.48 g·cm-3,1.53 g·cm-3 and 1.58 g·cm-3 with dry density gradient were obtained from the saline alkaline soil in the southwest floodplain of Shandong Province as the material. After different freeze-thaw cycles, the soil permeability coefficient was obtained by saturated water conductivity test, and the soil porosity and interval pore volume distribution were obtained by MIP. After that,the soil permeability coefficient was predicted by BP neural network. Therefore, the influence of permeability coefficient of saline-alkali soil under different initial conditions in the test area was considered. Through the error analysis and goodness of fit comparison of neural network, it is feasible to use neural network to predict the permeability coefficient of saline-alkali soil in the test area, and the model can fully predict the permeability coefficient of saline-alkali soil under different initial conditions.
作者
徐文硕
李克升
陈龙霄
宋贵磊
耿雨晗
刘传孝
XU Wen-shuo;LI Ke-sheng;CHEN Long-xiao;SONG Gui-lei;GENG Yu-han;LIU Chuan-xiao(College of Water Conservancy and Civil Engineering/Shandong Agriculture University,Tai’an 271018,China)
出处
《山东农业大学学报（自然科学版）》
北大核心
2021年第5期805-812,共8页
Journal of Shandong Agricultural University：Natural Science Edition
基金
山东省重点研发计划资助项目(2018GNC110023)
国家自然科学基金项目(51574156)。
关键词
BP神经网络
冻融盐碱土
渗透性
Neural network
saline-alkali soil after freeze-thaw
permeability
