Geological identification using shield tunneling parameters based on four unsupervised clustering methods

Abstract

This study develops a comprehensive framework for identifying geological characteristics (GC) during shield tunneling. Four unsupervised clustering methods are employed: K − means++, Fuzzy C − means (FCM), Gaussian Mixture Model (GMM), and Hierarchical Clustering (HIC). These clustering methods are applied to both the original and dimensionally reduced datasets, which are obtained through principal component analysis (PCA). The performance of each clustering method is emphasized by capturing the inherent variability of GC. K − means++ and FCM exhibit high recognition rates (89.54 % and 90.37 % respectively) and excellent stability. The mean Rand index values for both the original and PCA − processed data are close to 1. PCA significantly enhances the performance of GMM. The identification rate of GMM increases from 67.28 % to 87.68 %, and its mean Rand index improves towards 1. Conversely, HIC has low recognition rates (18.63 % for the original data and 20.60 % after PCA) and low mean Rand index values (0.93 for the original data and 0.91 after PCA), indicating poor stability. The proposed framework, integrating data preprocessing, comprehensive index calculation, and the application of clustering algorithms, provides a robust and effective approach to improve the identification and understanding of GC during the tunneling process. This, in turn, can enhance decision − making processes in similar engineering projects.