Estimation of Shadan gold grade using borehole coordinate data by machine learning technique

Abstract

Grade estimation is one of the key stages in technical and economical evaluation of a mine. The grade values have significant effect on planning, designing and managing mine. Therefore, it seems necessary to apply methods that estimate these values with high accuracy and correctness. One of the best and accurate methods to obtain a grade in a deposit is to dig exploratory boreholes which is not possible due to high costs. So we used artificial neural network to reduce the costs. Artificial neural network, back propagation error algorithm has the ability to estimate the gold grade using borehole coordinate exploration data. This plan is also consonant with operational and real requirements. Because in all wells, logs are taken and in a small number of wells, coring is done or we may not have a core in some parts of the well or it has been destroyed. Therefore, having a well-trained network, it is possible to simulate the gold grade of the well or the parts without a core. As a result, in each well, an estimate of the gold grade is obtained, which can be used to provide a better and more reliable model for the deposit for simulation. The results of test and validation data indicate the remarkable ability of the machine learning techniques system in estimating the gold grade in the data. Grade estimation using two criteria called squared normalized, and squares mean of network performance error (for train, average is 0.057), (for test, average is 0.094) , (for validation, average is 0.088) and also The regression between the predicted and the real values of Train, Test and Validation are obtained respectively, as 0.90, 0.75 and 0.80. The results show that the neural network gives a reasonable estimation for gold grade.