<![CDATA[Tunnel Boring Machine (TBM) operations are governed by complex and nonlinear interactions among geological variability, machine control parameters, and energy consumption, posing significant challenges for reliable performance prediction and operational optimization. Conventional empirical and physics-based approaches often struggle to capture regime-dependent behavior and parameter coupling under heterogeneous excavation conditions. To address these limitations, this study proposes an integrated and interpretable data-driven framework that combines ensemble machine learning, time-series modeling, unsupervised regime identification, multi-objective optimization, and explainable artificial intelligence within a unified analytical architecture. A multisource dataset encompassing geotechnical, operational, environmental, energy, and economic parameters was analyzed using Extreme Gradient Boosting (XGBoost), Random Forest, Gradient Boosting Regression, and recurrent neural networks. Among these, XGBoost demonstrated superior predictive capability, achieving the highest coefficient of determination and consistently lower prediction errors compared with baseline models. Unsupervised clustering identified distinct operational regimes—efficient, intermediate, and aggressive—enabling a structured evaluation of energy–cost trade-offs. Regime-aware optimization further indicated substantial potential for reducing both energy consumption and operational costs relative to high-intensity operating conditions. Sensitivity analysis using SHAP, mutual information, ANOVA, and Sobol indices revealed strong interaction effects among thrust force, torque, and rock strength parameters, highlighting the coupled nature of TBM excavation mechanics. The proposed framework extends conventional predictive modeling approaches by translating data-driven insights into interpretable, regime-based operational strategies. It provides a scalable methodological foundation for the future development of digital twin applications in TBM systems and contributes to more energy-efficient, cost-effective, and sustainable tunneling operations in complex underground environments.]]>
