基于主成分与聚类分析滇池湖相粉土物理力学参数

Investigation of Physical and Mechanical Parameters of Lacustrine Facies Silty Soils in Dianchi Lake Based on Principal Component and Cluster

昆明地铁5号线大部分站点及区间都位于滇池湖相软土沉积层,其中粉土层分布范围最广,工程性质差,物理力学参数空间变异性较大,难以获得可靠的物理力学参数.现选取会展中心站场地粉土的9个物理力学参数,保留原样本主要信息,先利用系统聚类及主成分分析法(PCA)实现将多个变异系数不同的物理力学参数的分类和降维;再采用最优k值的k-means聚类分析查明各类参数的空间分布规律,最大程度排除误差数据;最后采用多种截尾法进一步优化参数取值.将优化结果与原参数取值、综合设计值对比,检验本次提出参数取值方法的可行性.同时,采用PCA法从昆明地铁5号线17个站点选取粉土参数离散性较强的两个站点的9个物理力学参数进行验证,获得的优化参数取值与参考值相一致.表明把主成分与聚类分析相结合的方法对参数取值进行研究是可行的,方法的适用性强,具有较高的推广应用价值.

The majority of stations and sections along Kunming Metro Line 5 are situated in lacustrine soft soil deposits of Dianchi Lake.Among them,silty clay layers have the widest distribution,poor engineering properties,and significant spatial variation in physical and mechanical parameters,making it difficult to obtain reliable parameter values.This study focuses on nine physical and mechanical parameters of silty clay at the Exhibition Center station while preserving the key information from the original samples.Firstly,systematic clustering and principal component analysis(PCA)are employed to classify and reduce the dimensions of these parameters,considering their varying coefficients of variation.Subsequently,using the optimal value of k determined by k-means cluster analysis,the spatial distribution patterns of the parameters are determined to eliminate erroneous data to the greatest extent possible.Finally,various truncation methods are applied to further optimize the parameter values.The feasibility of the proposed parameter selection method is tested by comparing the optimized results with the original parameter values and comprehensive design values.Simultaneously,the PCA method is used to verify the selected parameters for two stations with high dispersion of silty clay parameters among the 17 stations along Kunming Metro Line 5.The optimized parameter values obtained are consistent with the reference values,indicating the feasibility of studying parameter values through the combination of principal component and cluster analysis.The method demonstrates strong applicability and holds significant value for practical application.