A feature importance-based intelligent method for controlling overbreak in drill-and-blast tunnels via integration with rock mass quality

Optimizing blasting parameters is of paramount importance in minimizing overbreak during tunneling. Hence, this paper proposed an intelligent approach, which integrates separate parameter optimization based on varying rock mass qualities, with the objective of reducing overbreak. This novel intelligent method constructs a comprehensive model with three distinct functions, which can provide precise overbreak prediction, analyze the mechanisms by which input parameters influence overbreak, and integrate feature importance into the blasting parameters optimization process. First, the hyperparameters of seven tree-based algorithms were optimized using the Sparrow Search Algorithm (SSA), and the best predictive model was selected by comparing various performance metrics. Then, the importance and influencing mechanisms of input features on overbreak were revealed through the utilization of the Shapley Additive Explanation. Subsequently, interactions among crucial parameters were investigated, and their design values were recommended. Finally, a novel parameter optimization method was employed to reduce overbreak, which combines the conclusions drawn from the importance analysis of input features with SSA through three key steps: modification of the initial position, expanding the search scope, and neighborhood perturbation. Compared to the unimproved methods, the proposed approach can significantly reduce post-blast overbreak areas in different sections of the tunnel by 12.8 % and 16.4 %, respectively. •Predicted blast overbreak based on deep learning algorithms.•Stabilized importance analysis results of input parameters.•Revealed influencing mechanisms of overbreak via SHAP.•Incorporated feature importance with parameter optimization.