Recurrent neural network approach to mineral deposit modelling

Abstract

Estimation of grade distribution of iron ore has been attempted using a neural network based model for a banded iron formation (BIF) type iron ore deposit of East India. The model provides an alternative to traditional geostatistical approach. Geostatistical methods assume linear spatial correlation among sample values within a mineral deposit and characterizes various parameters of a mineral deposit. A Recurrent Neural Network (RNN) model, on the other hand, owing to its capability of capturing non-linearity of distribution of a sample space, is suggested to provide estimatesof the iron ore grade. Due to its dynamic training mechanism, RNN model additionally provides a fast and robust solution. Cross-validation of the RNN model has been carried out by comparing its resulting Fe grade outputs with the kriged estimated Fe values. The RNN model provides an improved technique forcogent grade modelling.