%0 Journal Article
%T 基于GA-BP神经网络的粗粒土渗透系数预测
%T Prediction of permeability coefficients of coarse－grained soil based on GA－BP neural network
%A 饶云康
%A 丁瑜
%A 倪强
%A 许文年
%A 刘大翔
%A 张恒
%A RAO,Yunkang
%A DING,Yu
%A NI,Qiang
%A XU,Wennian
%A LIU,Daxiang
%A ZHANG,Heng
%J 水利水运工程学报
%J Hrdro-Science and Engineering
%@ 1009-640X
%V
%N 6
%D 2018
%P 92-97
%K 粗粒土； 渗透系数； BP神经网络； 遗传算法； 孔隙比； 级配
%K coarsegrained soil; permeability coefficient; BP neural network; genetic algorithm; void ratio; gradation
%X 针对粗粒土渗透性能受颗粒级配、密实程度等因素影响而呈现明显差异，提出一种粗粒土渗透系数预测方法。收集并整理得到93组粗粒土数据，以全级配(d10~d100)和孔隙比作为BP神经网络的输入变量，利用遗传算法优化BP神经网络的初始权值与阀值，构建基于BP神经网络和遗传算法的粗粒土渗透系数预测模型。结果表明：该GA-BP神经网络经过55次迭代之后精度满足要求；87组训练样本预测结果的平均相对误差为5.10%，其中有75%的样本相对误差小于平均相对误差；6组检测样本预测结果的平均相对误差为6.39%，该网络模型泛化性能良好。采用GA-BP神经网络，由全级配和孔隙比能较好地预测粗粒土的渗透系数，且收敛速度、预测精度及泛化性能均优于标准的BP神经网络模型。
%X In view of the obvious difference in the permeability of the coarse grained soil, effected by factors such as gradation of grain and compaction degree, a prediction method for the permeability of the coarse－grained soil is proposed in this study. 93 groups of data of the coarse－grained soil are collected and obtained. Taking the full gradation(d10~ d100) and the porosity ratio as the input variables of the BP neural network, a prediction model for the permeability coefficients of the coarse－grained soil is developed on the basis of the BP neural network and genetic algorithm, by using the genetic algorithm to optimize the BP neural network′s initial weights and thresholds. The research results show that the accuracy of the GA-BP neural network meets the requirements after 55 iterations. And the mean relative error of the predicted results of 87 groups of the training samples is 5.10%. Moreover, a relative error of 75% of the training samples is less than the mean relative error. In addition, the mean relative error of 6 groups of the testing samples is 6.39%, which indicates that the generalization performance of the network model is high. It is concluded that the permeability coefficients of the coarse-grained soil can be well predicted by applying the GA-BP neural network considering the full gradation and void ratio. Moreover, The convergence rate, the prediction accuracy and the generalization performance of the GA-BP neural network are better than those of the standard BP neural network model. And the model based on the GA-BP neural network can provide technical references and support for the selection and improvement of the coarse-grained soil in practical engineering.
%R
%U http://slsygcxb.cnjournals.org/ch/reader/view_abstract.aspx
%1 JIS Version 3.0.0