Gas Hydrate Dissociation Temperature Prediction in Porous Media: Effects of Pore Size and Modeling Approach

Understanding the dissociation behavior of gas hydrates
in confined
porous media is crucial for assessing their stability and potential
applications in energy storage, carbon capture, and climate modeling.
In this study, we develop two distinct approaches to predict the equilibrium
dissociation temperature of gas hydrates in porous materials with
varying pore sizes: a thermodynamic model based on the activity approach
and a suite of machine learning (ML) models. The thermodynamic model
explicitly accounts for the effects of confinement on hydrate phase
stability and was validated using an unfiltered data set for methane
(CH4) and propane (C3H8) hydrates,
achieving low average absolute deviations (AAD%) of 0.17 and 0.62%,
respectively. To complement and generalize these predictions, we trained
four ML models: Decision Tree, Random Forest, Support Vector Machine
(SVM), and Multi-Layer Perceptron. These models used input features
such as pore diameter, system pressure, and critical gas properties.
A group-based data splitting strategy was applied, with propane data
exclusively reserved for testing to assess true generalization. The
SVM model exhibited the highest predictive performance on unseen data,
with an AAD% of 0.52%. To enhance interpretability, SHapley Additive
exPlanations (SHAP) analysis was employed. The results confirmed alignment
between the ML model’s decision logic and known physical principles
and identified critical temperature, pressure, and pore size as the
most influential features. While group-based splitting improved robustness,
discrete SHAP patterns suggest that a broader variety of gases in
the training data could further enhance generalizability. Overall,
this integration of physics-based and data-driven modeling provides
accurate and interpretable predictions of hydrate dissociation behavior
in porous systems, supporting future developments in both geological
and industrial applications.