基于层次聚类算法的孔压静力触探土体分类方法及试验研究

划分土层、辨别土类是孔压静力触探（CPTU）成果应用的基础。常规的人工分层效果差强人意，而土体行为分类法尽管可靠性高，但无法起到分层效果。引用层次聚类算法，通过对变量的选择、数据的标准化、距离矩阵的生成和类数目的确定，得到了基于层次聚类算法的CPTU数据聚类流程图；结合Robertson和Campanella分类图，提出了基于CPTU测试数据的土层划分与命名规则。采用自主研制的静力触探-钻探一体机，在宁波市轨道交通4号线上展开试验，将土层划分与命名结果与钻孔柱状图展开对比，结果表明:以锥尖阻力qt、摩阻比Rf和孔隙水压力u2作为初始聚类参数的分层图对8个主层的划分与钻孔柱状图几乎一致。其中，以qt-Rf为初始聚类参数的分层图能够识别出单靠qt曲线无法识别的2个薄夹层，体现了Rf的作用；以qt-u2为初始聚类参数的分层图对砂类土划分得过于细致，对黏土划分得过于粗糙，表明u2对砂性土变化过于敏感，对黏性土变化不敏感；以qt-Rf-u2为初始聚类参数的分层图既保留了qt的主要特征，又适当地融入了Rf和u2对土层划分的影响，分层效果最佳。钻孔剖面黏性土的不排水抗剪强度曲线总体上符合土体性质与土层深度的变化规律，从侧面反映了聚类分层图的准确性。

Soil classification method and experimental research onCPTU based on the hierarchical clustering algorithm

Dividing the soil layer and identifying soil types are the basis for the application of CPTU. The conventional artificial stratification method is not satisfactory, but the soil behavior classification method cannot achieve the stratification effect despite its high reliability. By using the hierarchical clustering algorithm, the flow chart of CPTU soil classification based on hierarchical clustering algorithm is obtained. Through the selection of variables, CPTU test data are proposed. The test was carried out on Line 4 of the Ningbo Rail Transit with a self-developed static standardization of data, the generation of distance matrix and the determination of the number of classes. Combining the Robertson and Campanella classification maps, the soil layer classification and naming rules are based on cone penetration-drilling machine. The results of soil layer division and naming are compared with the borehole histogram. The results show that the division of 8 main layers by stratified graphs with the initial clustering parameters of cone resistance, qt, friction ratio, Rf and pore water pressure, u2, is almost identical with the drilling column. The stratified graph with qt- Rf as the initial clustering parameter can identify two thin layers which cannot be identified by qt curve alone, which reflects the role of Rf; the stratified graph with qt-u2 as the initial clustering parameter is too detailed for sandy soil classification and too rough for clay classification, indicating that u2 is insensitive to clay change and is not sensitive to sandy soil change. The stratified graph with qt- Rf-u2 as the initial clustering parameter not only retains the main characteristics of qt, but also incorporates the effects of Rf and u2 on soil stratification appropriately, and the stratification effect is the best. The undrained shear strength curve of cohesive soil in the borehole section generally conforms to the variation law of soil properties and soil depth, and reflects the accuracy of clustering stratification map from the side.