Prediction of excavation damaged zone depth of chamber groups by damage initiation and spalling limit: Numerical simulation and machine learning algorithms

As underground chamber group structures increase, predicting excavation damage zones (EDZ) depth is crucial for maintaining chamber stability. Traditional prediction methods exhibit limitations in terms of accuracy and efficiency. To address this, a predictive method integrating numerical simulations using the damage initiation and spalling limit (DISL) approach with Bayesian-optimized XGBoost (BO-XGBoost) was developed. Initially, a FLAC3D numerical model was established, and 500 data groups were generated by adjusting the model parameters. BO-XGBoost was applied to predict the depths of the highly damaged zone (HDZ), inner EDZ (EDZ i ), and outer EDZ (EDZ o ). Subsequently, the evaluation was conducted using the root mean square error (RMSE), mean absolute error (MAE), variance accounted for (VAF), and coefficient of determination ( R 2 ). The results were compared with those of the light gradient boosting machine (LightGBM), back propagation neural network (BPNN), random forest (RF), and support vector machine (SVM), and a feature importance analysis was conducted. Finally, the method was applied to predict the EDZ o depth of chambers in the Fankou lead-zinc mine. The results showed that BO-XGBoost demonstrates high predictive accuracy, outperforming the other models, with RMSE, MAE, VAF, and R 2 values for the depths of HDZ, EDZ i , and EDZ o of 0.053, 0.041, 89.8%, 0.905; 0.045, 0.036, 94.8%, 0.958; and 0.064, 0.051, 95.4%, 0.962, respectively. The chamber size and spacing significantly influence the EDZ i and EDZ o depths, with the excavation sequence causing variations in HDZ depth. In engineering applications, the EDZ o depth prediction error for the chambers is within 5%, offering valuable insights for predicting EDZ depth in chamber groups.