Comparative evaluation of radial basis functions and kriging for ore grade estimation

摘要

In this study, radial basis function network (RBF) was investigated as a tool for grade estimation in a bauxite deposit, and its performance was evaluated with respect to the well-established geostatistical technique. Two critical variables namely, Alumina (Al_2O_3) and Silica (SiO_2) were chosen for the grade estimation purpose. Several important aspects related to RBF network modeling are discussed. For the neural network (NN) modeling the entire dataset was divided into three statistically similar subsets viz., the training dataset, the calibration dataset and the validation dataset using genetic algorithm (GA). The validation dataset remained same for both the cases to make a legitimate comparison. A comparative evaluation of the two techniques was done on the basis of the mean square error (MSE) and the coefficient of determination (Rsq) calculated from their outputs on the validation dataset. Though the Rsq values of both the above mentioned methodologies were not encouraging which (might be due to the presence of noise in the data), the results indicated that the RBF did a better job in the prediction of the Silica grade. However, the ordinary kriging technique performed slightly better while predicting the alumina grade.