Prediction of tunnel face rock mass classification using an ensemble model enhanced by feature cross based on TBM boring data

In tunneling engineering, accurate prediction of rock mass classification at the tunnel face is crucial to ensure construction safety and efficiency. In this study, we propose a novel approach utilizing Tunnel Boring Machine (TBM) excavation data collected from the Yin-Chao project in China, to develop a rock mass classification model. The dataset comprises 401 sensors installed on the TBM, providing data on various mechanical control parameters during the excavation process. We conduct thorough data preprocessing, including feature selection and valid data extraction, to prepare the dataset for model training. Additionally, we implement an ensemble learning framework, integrating a feature cross enhanced network (E-DeepFM) with base models such as Logistic Regression (LR), Support Vector Machine (SVM), K-Nearest Neighbors (KNN), and Decision Tree (DT), to enhance predictive performance. The results demonstrate that the integration of E-DeepFM significantly improves the predictive precision of the base models, particularly for SVM and KNN, as the F1 score from 0.827 to 0.877 for the SVM model and from 0.923 to 0.944 for the KNN model. Furthermore, interpretability analysis using SHAP values highlights the importance of cross features in enhancing model performance. This study contributes to the field of tunneling engineering by providing a reliable method for real-time rock mass classification during TBM excavation, thereby enhancing construction safety and efficiency.