Interpretable Machine Learning for Predicting Cutting Force of Tunnel Boring Machine: A Study on SVR With Meta‐Heuristic Optimization

Abstract

Tunnel Boring Machines (TBMs) have been widely adopted in the construction of large‐scale and long‐distance rock tunnels. However, accurately predicting the cutting forces of disc cutters remains a significant challenge. To address this issue, a Support Vector Regression (SVR) model was introduced in this study to predict the normal force (FN) and rolling force (FR) based on an established dataset. To improve prediction accuracy, the SVR model's hyperparameters were optimized using three nature‐inspired algorithms: Whale Optimization Algorithm (WOA), Particle Swarm Optimization (PSO), and Crayfish Optimization Algorithm (COA). Performance evaluations using R2, VAF, NES, RMSE, and MAE demonstrated that the proposed hybrid models (WOA‐SVR, COA‐SVR, and PSO‐SVR) significantly outperformed traditional empirical and semi‐theoretical models, such as the Colorado School of Mines (CSM) model. Among these, the PSO‐SVR model yielded the best results (FN prediction: 0.9657, 96.5241%, 0.9652, 15.0258, 11.6712; FR prediction: 0.9691, 95.3219%, 0.9517, 2.3272, 1.6531). Furthermore, SHapley Additive exPlanations (SHAP) and Local Interpretable Model‐agnostic Explanations (LIME) were employed to enhance model interpretability. The analysis revealed that uniaxial compressive strength (UCS), cutter penetration depth (P), and disc cutter spacing (S) play a pivotal role in the accurate prediction of FN and FR. Additionally, a MATLAB‐based visual interface was developed to provide practical technical support for optimizing real‐world TBM tunneling performance.