Prediction of Flyrock Distance in Surface Mining Using a Novel Hybrid Model of Harris Hawks Optimization with Multi-strategies-based Support Vector Regression

To weaken and control effectively the harm of flyrock in open-pit mines, this study aimed to develop a novel Harris hawks optimization with multi-strategies-based support vector regression (MSHHO-SVR) model for predicting the flyrock distance (FD). Several parameters such as hole diameter (H), hole depth, burden-to-spacing ratio, stemming, maximum charge per delay, and powder factor were recorded from 262 blasting operations to establish a FD database. The MSHHO-SVR model compared the predictive performance with several other models, including Harris hawks optimization-based support vector regression (HHO-SVR), back-propagation neural network, extreme learning machine, kernel extreme learning machine, and empirical methods. The root mean square error (RMSE), the mean absolute error (MAE), the determination coefficient (R2), and the variance accounted for (VAF) were employed to evaluate model performance. The results indicated that the MSHHO-SVR model not only performed better in the training phase but also obtained the most satisfactory performance indices in the testing phase, with RMSEs of 12.2822 and 9.6685, R2 values of 0.9662 and 0.9691, MAEs of 8.5034 and 7.4618, and VAF values of 96.6161% and 96.9178%, respectively. Furthermore, the calculation results of the SHAP values revealed that H was the most critical parameter for predicting FD. Based on these findings, the MSHHO-SVR model can be considered as a novel hybrid model that effectively addresses flyrock-like problems caused by blasting.

Automated summary

In order to mitigate the dangers of flyrock in open-pit mines, a new model called MSHHO-SVR, based on Harris hawks optimization and multi-strategies support vector regression, was developed to predict the flyrock distance (FD). The model was compared to other models such as HHO-SVR, back-propagation neural network, extreme learning machine, kernel extreme learning machine, and empirical methods. Evaluation metrics including RMSE, MAE, R2, and VAF were used to assess the model performance. The results showed that MSHHO-SVR outperformed the other models, achieving the lowest RMSE, highest R2, and MAE, and highest VAF values.