Application of Gaussian Process Regression for Bench Blasting Rock Fragmentation Prediction and Optimization at Wolongan Open-Pit Mine

Abstract

This study developed a Gaussian process regression (GPR) model to predict and optimize blast fragmentation at Wolongan Mine by using the primary data from the mine and secondary data from other mines. The blast data comprised 125 datasets, each containing seven blast design parameters as inputs and the muckpile mean fragment size as the model output. Additionally, the study developed artificial neural networks (ANNs), support vector regression (SVR), and multiple linear regression (MLR) models, and compared their prediction performances with the GPR model. The models’ accuracies were evaluated using five statistical metrics, including coefficient of determination (\({R}^{2}\)), root mean square error (RMSE), variance accounted for (VAF), mean absolute bias error (MABE), and mean absolute percentage error (MAPE). The GPR model outperformed the other models, with \({R}^{2}\), RMSE, VAF, MABE, and MAPE values of 0.9302, 0.0487, 93.2670, 0.0383, and 13.9405, respectively, for the test data. Based on the top-down correlation and Kendall’s coefficient of concordance analyses on the four sensitivity analysis methods used, the study found that the in situ block size and Young’s modulus of the rock were the most important parameters affecting fragmentation. Using the GPR model, the study showed that reducing the blast burden by 13–23% could decrease the mean fragment size of the muckpile at Wolongan Mine by 6–12%, leading to a significant reduction in the percentage of boulders.