Usage
  • 172 views
  • 374 downloads

Geometallurgical Modeling with Data Imputation and Response Surface Methodology

  • Author / Creator
    Kumara, Paolo Christmasdato
  • Geostatistical modeling used to focus on the grade of the main commodity or metal being mined and sold for profit. As mining has developed, the metallurgical characteristics of the rock have become important. Geometallurgy tests are developed to understand the processing characteristics of the mined rock. Yet, geometallurgy tests are expensive and takes longer than geology grade assays. Multivariate geostatistical techniques are used to model the geometallurgy variables. Yet, many of the geometallurgy variables do not average linearly and are compositional, that is, they sum to unity. These complexities make modeling geometallurgy variables challenging.
    Missing geometallurgical data may degrade the quality of prediction. There are two evident modeling frameworks that can be used. The first framework is an imputation framework which calculates the spatial continuity and relationships to grade variables to predict the missing data. The second framework is a response surface methodology (RSM) framework that accounts for the relationship to grade variables. There are different RSM techniques including (1) linear regression, (2) Alternating Conditional Expectations (ACE), and (3) random forest; that are compared to understand their advantages and disadvantages. The two frameworks perform differently in different circumstances. Considerations for the best framework are developed in this thesis.
    Two new imputation techniques that account for data spatial continuity and complex multivariate relationships are developed. The first proposed technique, called RF-enhanced, alters the imputation likelihood mean calculation with random forest prediction without changing the variance while the second proposed technique, called RF-moment, alters both likelihood mean and variance. Both frameworks consider the prior spatial distribution in the same way as parametric imputation. The proposed techniques improve the imputation accuracy in certain circumstances. Numerous examples are presented to provide guidance on technique selection.
    Random forest regression does not always perform better in predicting missing values. Yet, both proposed imputation techniques still perform quite well and have a promising result for imputation development.

  • Subjects / Keywords
  • Graduation date
    Spring 2020
  • Type of Item
    Thesis
  • Degree
    Master of Science
  • DOI
    https://doi.org/10.7939/r3-2009-9a86
  • License
    Permission is hereby granted to the University of Alberta Libraries to reproduce single copies of this thesis and to lend or sell such copies for private, scholarly or scientific research purposes only. Where the thesis is converted to, or otherwise made available in digital form, the University of Alberta will advise potential users of the thesis of these terms. The author reserves all other publication and other rights in association with the copyright in the thesis and, except as herein before provided, neither the thesis nor any substantial portion thereof may be printed or otherwise reproduced in any material form whatsoever without the author's prior written permission.