30-5 Estimation of Pitzer Parameters using Neural Network, Random Forest, and exTreme Gradient Boosting: A comparative analysis of machine learning applications in critical mineral extraction and nuclear waste management

Session: Geological and Geochemical Investigations of Critical Minerals in New Mexico and Beyond, and Technological Advances in Extraction of Critical Minerals

Presenting Author:

Authors:

Abstract:

Thermodynamic models valid for concentrated electrolytes play an important role in many fields including nuclear waste management, extraction of critical minerals (CM), and treatment of produced water, etc., as concentrated electrolytes are usually present in those processes.  For instance, tank waste at the Hanford site, Washington, is characterized with multiple components with high ionic strength.  Aluminium (Al) is one of the critical minerals listed by Department of Energy (DOE), and is primarily extracted from bauxite ores using concentrated alkaline solutions such as NaOH.  The Pitzer model is a thermodynamic model used to determine the activity coefficients of aqueous species with a wide range of ionic strengths up to saturation of salts. The utility of the Pitzer model, however, is constrained by experimental limitations that impose challenges on the data collection process, thereby restricting its application.

Machine Learning can potentially facilitate the improvement of parameter estimation in the Pitzer model where experimental data is limited.  We developed three machine learning techniques including Artificial Neural Network (ANN), Random Forest (RF), and eXtreme Gradient Boosting (XGB) to estimate beta⁽⁰⁾, beta⁽¹⁾, and phi parameters of the Pitzer model and compared their performances.  We observed that RF and XGB models exhibited significantly better performance in estimating Pitzer parameters.  We further applied these models to predicted the Pitzer parameters of FrCl, RaCl₂, and PoCl₄ and subsequently extracted their mean estimates using bootstrap analysis. By this, we demonstrate an effective application of machine learning in predictive modeling of geochemical processes and its potential to advance nuclear waste disposal management and CM extractions.

Sandia National Laboratories is a multi-mission laboratory operated by National Technology and Engineering Solutions of Sandia, LLC., a wholly owned subsidiary of Honeywell International, Inc., for the U.S. Department of Energy’s National Nuclear Security Administration under contract DE-NA-0003525.  This work was supported by the U.S. Department of Energy (DOE) Office of Environmental Management (EM) through the Laboratory Policy Office Hanford Tank Waste R&D Program.  SAND2026-16735A.

© Copyright 2026 The Geological Society of America (GSA), all rights reserved.