Optimization of haulage-truck system performance for ore production in open-pit mines using big data and machine learning-based methods

Abstract

Ore haulage systems are considered critical when evaluating the efficiency of the investment and design of open-pit mines. Smart mines are also adopted to increase mine production, and the, optimization of the production of equipment is necessary. Therefore, this study proposes an unsupervised intelligent system for predicting the performance of a truck-haulage system in the ore transportation process in open-pit mines using a combination of Harris hawks optimization (HHO) and support vector machine (SVM), named the HHO-SVM model. Different kernel functions were investigated with the proposed HHO–SVM model, including radial basis, polynomial, and linear functions. Random forest (RF) and back-propagation neural network (BPNN) models were also developed and compared with the proposed model. To demonstrate the performance in practice, 16005 datasets of a truck-haulage system was collected, and the downscaling method was applied to downscale the size of the dataset into 3000 observations, aiming to improve the computing cost of the models. The results revealed that the BPNN, RF, SVM (without optimization), and HHO–SVM models are potential intelligent models for predicting ore production. The comparisons between the models indicated that the radial basis function was the best fit of the HHO-SVM model in predicting ore production with a root-mean-squared error (RMSE) of 197.213, determination coefficient (R2) of 0.991, and mean absolute error (MAE) of 154.256. Meanwhile, the polynomial function achieved lower performance with an RMSE of 275.427, R2 of 0.982, and MAE of 205.460; the linear function achieved the lowest performance overall with an RMSE of 844.111, R2 of 0.841, and MAE of 595.173. Similar results were also obtained in practice through the validation datasets, with an accuracy in the range of 98–99% for the proposed HHO-SVM model with the radial basis function. However, accuracies in the range of only 84–85% for the linear function and 97–98% for the polynomial function were achieved.