摘要
以胶凝砂砾石(CSG)为例,通过119组试验试块进行渗透试验测定,形成了CSG的28天渗透系数的数据集。数据集中有119组渗透系数数据,其中最小值为3.41×10^(-5)cm/s,最大值为27.812×10^(-5)cm/s。数据主要集中在3×10^(-5)~22×10^(-5)cm/s范围内,约占总样本数的97%。根据箱线图确定去除异常值并通过偏态峰度检验、K-S检验和分布图结果,可认为CSG材料的渗透系数数据服从正态分布规律。在此基础上,使用BP和GABP神经网络模型进行渗透系数预测,并对两种模型的预测精度进行比较。结果表明,GABP估计模型的精度略优于BP模型。CSG渗透系数实际值与预测值吻合较好,说明预测效果较好。
In this paper,the 28-day permeability coefficient data set of CSG was formed by the experimental determination of permeability through 119 sets of experimental test blocks with cemented sand gravel(CSG)as the research object.There are 119 sets of permeability coefficient data in the dataset,of which the minimum value is 3.41×10^(-5)cm/s and the maximum value is 27.812×10^(-5)cm/s.The data are mainly concentrated in the range of 3×10^(-5) to 22×10^(-5) cm/s,accounting for about 97%of the total number of samples.Based on the box line plot to remove the outliers and by the results of the skew kurtosis test,K-S test,and distribution diagram,it can be concluded that the permeability coefficient data of CSG materials obey the law of normal distribution.Based on this,the BP and GABP neural network models were used to predict the permeability coefficients,and the prediction accuracy of the two models was compared.The results show that the accuracy of the GABP model was slightly better than that of the BP model.The actual values of CSG permeability coefficients agreed better with the predicted values,which indicated that the prediction was better.
作者
韩立炜
陈明
HAN Li-wei;CHEN Ming(College of Water Resources,North China University of Water Resources and Hydropower,Zhengzhou 450046,China)
出处
《水电能源科学》
北大核心
2024年第3期110-114,共5页
Water Resources and Power
基金
国家重点研发计划(2018YFC0406803)
国家自然科学基金项目(51509091)。
关键词
胶凝砂砾石
渗透系数
箱线图
正态分布
神经网络
colluvial gravel
permeability coefficient
box-line plot
normal distribution
neural networks
