摘要
目前,常规土体试验及预测方法都难以准确获得压缩模量E_(s)。现基于机器学习理论建立一种非参数集成优化法计算E_(s),并与传统回归模型对比分析。从昆明地铁5号线会展中心场地选取203组泥炭质土物理力学指标样本,结合工程经验,选取其中8个重要的物理指标作为输入集,利用遗传算法优化BP神经网络输入层、隐含层及输出层之间的权值及阈值,采用相关系数R、正确率ACC及均方根误差RMSE多个评估指标优化确定算法的重要参数,将建立好的模型应用于多种土体,并与目前应用较多的方法对比分析,最后比较经验公式与本文方法预测地基沉降的性能。结果显示,GA-BP神经网络方法对分析样本适应性强、算法收敛快、所得结果精准可靠,具有较大优越性。该方法对软土场地多参数预测具有一定的指导意义。
Conventional methods were tested to predict the properties of soil struggle to accurately obtain its compression modulus E_(s).In this study we developed a nonparametric integrated method of optimization based on machine learning to calculate E s and compare it with the traditional model of regression to this end.We chose 203 groups of physical and mechanical indices of samples of peaty soil from the Kunming Metro Line 5 Exhibition Center for this purpose.Eight important physical indices were used as the input set,and the weights and thresholds of the input layer,hidden layer,and output layer of the BP neural network were optimized by using a genetic algorithm.The correlation coefficient,accuracy,and root mean-squared error were used to optimize the parameters of the algorithm,and the resulting model was applied to a variety of soils and compared with prevalent methods in the area.Finally,the predictive performance of the proposed method was compared with the results of the relevant empirical formula.The results showed that the GA-BP neural network-based method was highly adaptable to the analysis of samples,converged quickly,and generated accurate and reliable results.This method could be used to predict multiple parameters of soft soil.
作者
阮永芬
李鹏辉
施虹
吴龙
李飞鹏
肖潇
RUAN Yongfen;LI Penghui;SHI Hong;WU long;LI Feipeng;XIAO Xiao(Faculty of Civil Engineering and Mechanics,Kunming University of Science and Technology,Kunming 650500,China;Survey Institute Co.,Ltd.,Yunnan Design Institute Group,Kunming 650228,China;China Railway 11th Bureau Group Urban Rail Engineering Co.,Ltd.,Wuhan 430074,China)
出处
《成都理工大学学报（自然科学版）》
CAS
CSCD
北大核心
2024年第2期258-268,280,共12页
Journal of Chengdu University of Technology: Science & Technology Edition
基金
国家自然科学基金重点项目(41931294)。
关键词
压缩模量
机器学习
遗传算法
BP神经网络
规范法
compression modulus
machine learning
genetic algorithm
BP neural network
normative method
