An Intelligent Approach to Predicting Dilution, Overbreak and Costs in Underground Mining Using Kolmogorov-Arnold Networks

Vera-Encarnación J.K.; Cotrina-Teatino M.A.; Arango-Retamozo S.M. · Mathematical Modelling of Engineering Problems (2025) · DOI: 10.18280/mmep.120308

Resumen

The main objective of this research was to predict dilution, overbreak, and costs in underground mining using Kolmogorov-Arnold Networks (KAN). The KAN model was rigorously evaluated against other machine learning approaches, including Random Forest (RF), Extreme Gradient Boosting (XGBoost), and Multilayer Perceptron (MLP). The methodology included the collection of 732 records, which was used for training and testing the models using the k-fold technique. The results indicated that the KAN model exhibited the highest performance, particularly in dilution prediction, with a Mean Squared Error (MSE) of 0.001, a Root Mean Squared Error (RMSE) of 0.032, a Mean Absolute Error (MAE) of 0.055, and an R2of 0.999. Regarding overbreak, KAN achieved an MSE of 0.002, an RMSE of 0.045, an MAE of 0.004, and an R2of 0.984. For cost, it achieved an MSE of 0.001, an RMSE of 0.032, a MAE of 0.027, and an R2 of 0.999, reaffirming its position as the most efficient model. Additionally, the integration of a multi-objective particle swarm optimization algorithm enabled the optimization of drilling and blasting parameters, resulting in notable reductions: 11.3% in operational dilution, 25% in overbreak, and 7.6% in operational costs. This study demonstrates the effectiveness of KANs in predicting key mining indicators, improving efficiency, safety, and profitability.

Instituciones: Universidad Nacional de Trujillo