A physical-informed and data-driven intelligent decision-making model for drill-and-blast tunnel excavation

The selection of the excavation scheme and the footage is crucial to ensuring the safety of the drill-and-blast tunnel construction. However, the complex geological environment and construction-induced disturbances present significant challenges for accurately predicting and estimating these factors. This study presents a hybrid physical–data-driven framework for ensemble prediction to support decision-making in rock tunnel excavation. An analytical solution is employed to assess the stability of the tunnel face, and its results are incorporated as physical information into the ensemble model. The proposed ensemble framework integrates three base models (RF (random forest), SVM (support vector machine), and MLP (multilayer perceptron)), a meta-model (XGBoost), and two optimization algorithms (Bayesian optimization (BO) and Sparrow Search Algorithm (SSA)). BO is employed to automate the hyperparameter tuning of the base model, while SSA is utilized both for weight allocation in the “soft voting” method and for hyperparameter optimization of the meta-model in the stacking ensemble model. Additionally, SHAP values are utilized to assess the relative importance of input variables. The comparison between the base and ensemble models, as well as the incorporation of physical information, indicates that the ensemble model significantly enhances predictive performance. Incorporating physical information further improves accuracy and generalization capability.