RESERVE ESTIMATION USING NEURAL-NETWORK TECHNIQUES

Reserve estimation involves the modeling of spatial variation and distribution of ore grade in the region of exploration. Current approaches are based essentially on either geometrical reasoning or statistical techniques, and generally assume that the spatial distribution of ore grade is a function of distance. Recent advances in neural networks have provided a decidedly new approach to solving this problem. In this paper, we describe our research in using a multilayer feedforward neural network to capture the spatial distribution of ore grade by directly training the network with field assay data at borehole locations. The trained neural network then is used to predict the distribution of ore grade in the drilling region. Results predicted from the neural network model are reasonably accurate compared with other conventional models. The main advantage of this approach is that it requires no complicated mathematical modeling and makes no assumptions about the spatial distribution of ore grade. This research indicates that neural networks are a promising tool in solving the generic reserve estimation problem in mining engineering.