Abstract:

To address the challenges of high uncertainty and complex influencing factors in predicting blast fragmentation in mining operations, this study proposed a LightGBM prediction model GA-QLightGBM that integrated GA optimisation with quantile regression. First, GA was employed to optimise the hyperparameters of LightGBM by simulating the natural selection process (selection, crossover, and mutation), thereby improving the model's predictive accuracy and stability. Then, different quantiles were set to construct prediction intervals for blast fragmentation, enabling the quantification of prediction uncertainty. Finally, the proposed model was applied to mine field datasets to verify its predictive performance and generalisation ability, providing an effective approach for blast fragmentation prediction and uncertainty analysis. The results show that the model achieves a coefficient of determination (R²) of 0.880 and a mean squared error (MSE) of 0.004 in point prediction, outperforming traditional point prediction models. In interval prediction, the prediction interval coverage probability (PICP), prediction interval normalized average width (PINAW), and corrected prediction interval accuracy (CPIA) are 0.947, 0.228, and 0.762, respectively, confirming the accuracy and reliability of GA-QLightGBM model. These findings offer a practical framework for quantifying the uncertainty of blast fragmentation, supporting refined blast design and risk control in mining engineering.

Key words: genetic algorithm (GA), light gradient boosting machine (LightGBM), blasting fragmentation, uncertainty, quantile regression, prediction model