Enhancing Mine Blasting Safety: Developing Intelligent Systems for Accurate Flyrock Prediction through Optimized Group Method of Data Handling Methods

Abstract

Flyrock, the unintended projection of rocks during mining blasts, poses significant safety risks and potential damage. Predicting flyrock is essential for implementing safety measures, minimizing injuries, preventing equipment and structural damage, optimizing blast plans, reducing downtime, and saving costs. Accurate predictions mitigate hazards, improve operational efficiency, and ensure the safety of workers and surrounding infrastructure. This study explored and developed hybrid methods for predicting flyrock using the group method of data handling (GMDH). Four swarm-based algorithms—particle swarm optimization (PSO), artificial bee colony (ABC), ant colony optimization (ACO), and whale optimization algorithm (WOA)—were combined with GMDH to enhance prediction accuracy. Additionally, a k-fold cross-validation method was applied to the datasets to improve reliability. The accuracy of these methods was evaluated using various statistical functions, such as Nash–Sutcliffe coefficient and Willmott's index, along with R-squared correlation (R²) graphs, half-violin plots, and quantile–quantile plots. The R² values for the WOA–GMDH, ACO–GMDH, ABC–GMDH, and PSO–GMDH models were 0.99, 0.97, 0.96, and 0.96, respectively. The WOA–GMDH method yielded the most accurate results, demonstrating superior performance when combined with GMDH. Furthermore, the performance of the WOA–GMDH model was compared with models developed in the literature using the same database, confirming its effectiveness. Sensitivity analysis identified that, in WOA–GMDH modeling, the powder factor as the most significant parameter while the spacing parameter was the least significant. The ACO–GMDH method exhibited the narrowest uncertainty band; whereas, the PSO–GMDH method had the widest, indicating the highest level of uncertainty.