Orebody modeling with Uncertainty: a Bayesian Neural Network Approach

Abstract

A Bayesian Neural Network (BNN) based spatial modeling technique is proposed here for orebody modeling. The Bayesian method for posterior probability calculation of the output parameter (grades) helps to calculate the uncertainty associate with the estimate. The paramatrers of the BNN model is selected by grid search algorithm. The expected value and the variance of block support are calculated by Markov chain Monte Carlo (MCMC) sampling from the posterior distribution at discretize points within the block. The BNN model is validated by applying the method in Walker Lake data set and comparing with ordinary kriging results. The results revealed that the proposed BNN method perform marginally better than ordinary kriging results. The variance map is less smooth than ordinary kriging. The proportional effect is also less in BNN-based model than ordinary kriging model.