Gas hydrate formation and dissociation: predictive, thermodynamic, and dynamic models

Abstract

Gas hydrates are a type of crystalline compound consisting of water and small gas molecules. A wide range of applications of gas hydrates in storing natural gas in the form of artificially created solid hydrates, known as solidified natural gas technology, gas separation processes, and seawater desalination technology, has attracted great interest in scientific and practical studies. Gas hydrate formation may also cause deleterious effects, such as blockage of gas pipelines. Therefore, accurate prediction of equilibrium conditions for gas hydrates is of great interest. In this regard, machine learning-based models were proposed to predict methane-hydrate formation temperature for a wide range of brines. A comprehensive database including 987 data samples covering 15 different brines was gathered from the literature. After data cleaning and preparation, three different models, namely multilayer perceptron (MLP), decision tree (DT), and extremely randomized trees (ET), were trained and tested. The ET model achieved the best performance with a root mean squared error (RMSE) of 0.6248 K for the testing dataset. Moreover, in an additional independent testing with MgBr2 samples, ET achieved an RMSE of 0.3520 K, confirming its strong generalization ability. The order of model accuracy was ET greater than MLP greater than DT. Compared to previous studies, the developed models achieved similar or better accuracy while covering a wider range of brine types. The findings of this study can be used as a reliable tool to predict methane-hydrate formation PT curves for pure water, single-salt brines, and multi-salt brines. The research further focuses on improving the prediction of equilibrium conditions in methane hydrate systems by incorporating diverse water-soluble hydrate formers and applying advanced machine learning techniques. Methane hydrates, which naturally form under high pressure and low temperature, can be more efficiently formed or dissociated by altering thermodynamic conditions using these hydrate formers. Accurate prediction of these conditions is crucial for optimizing gas storage and energy applications. Molecular descriptors and operational parameters, such as mole fraction and pressure, were used as input variables to predict equilibrium temperature. Machine learning methods, including Decision Trees (DT), Random Forests (RF), Support Vector Machines (SVM), and Multi-Layer Perceptron (MLP), were employed, using a novel former-based data-splitting approach rather than traditional sample-based methods. The RF model achieved the best results, with R2 = 0.930, RMSE = 1.71, and AARD = 0.48%. Feature selection, preprocessing, and Shapley Additive Explanations (SHAP) provided valuable insights into variable importance. Additional findings from the reduced model revealed that even less influential features significantly impacted distance-based models such as SVM and MLP. Interaction analysis through SHAP dependency plots highlighted the critical interplay between polar surface area and rotatable bonds in hydrate formation conditions. This work advances methane hydrate research by offering a more accurate and interpretable framework for predicting hydrate equilibrium, addressing key gaps in previous studies, and extending its applicability to a broader range of systems. Moreover, the introduction of a former-based data-splitting method improves generalization across different hydrate formers, while the use of SHAP values for model interpretability offers deeper insights into the relationships between molecular descriptors and hydrate equilibrium conditions. This study paves the way for improved selection of hydrate formers in hydrate systems. In addition to the phase equilibrium studies, this research also addresses the behavior of gas hydrates under confinement, focusing on hydrate dissociation in porous media. Understanding the dissociation behavior of gas hydrates in confined porous media is crucial for evaluating their stability and potential applications in energy storage, carbon capture, and climate modeling. Two distinct approaches were developed, namely a thermodynamic activity model and machine learning (ML) models, to predict equilibrium dissociation temperatures of gas hydrates in porous media of varying pore sizes. The activity model accounted for capillary effects and surface interactions and was validated against an unfiltered experimental dataset. For CH4 hydrates, the model achieved an AAD% of 0.17%, and for C3H8 hydrates, an AAD% of 0.62%. Complementary machine learning models (DT, RF, SVM, MLP) were trained using pore diameter, pressure, and gas critical properties as features. Group-based data splitting, with propane data reserved for testing, ensured robust evaluation. Among ML models, the SVM achieved the best predictive performance with an AAD% of 0.52%. SHAP analysis revealed that critical temperature, system pressure, and pore size were dominant predictors. The study also noted that experimental scatter was linked to pore structure variability and procedural differences, with larger pores showing convergence to bulk hydrate behavior. The combined modeling framework effectively captures hydrate behavior across a wide range of confined conditions, offering valuable predictive capability for both industrial and geological hydrate systems. In conclusion, the integration of physics-based and data-driven modeling enables accurate prediction of hydrate dissociation temperatures across a range of porous media. These findings support the development of predictive tools for hydrate systems in both geological and industrial applications. Finally, to complement the thermodynamic and equilibrium predictions, the dynamic transport behavior of hydrate particles in pipelines was investigated through CFD–DEM simulations. The dynamic behavior of hydrate particles suspended in water-dominated horizontal pipe flow using a two-way coupled CFD–DEM framework based on OpenFOAM and LIGGGHTS via CFDEM® coupling was explored. Multiphase flow simulations were conducted across inlet velocities of 0.2, 0.5, and 0.8 m/s and hydrate volume fractions of 2%, 5%, 8%, 15%, and 20%. Pressure drop behavior was quantified by extracting pressure gradients between two axial positions (z = 0.10 m and z = 0.49 m) early in the simulation. Results indicated that pressure drop increases with hydrate volume fraction at all flow velocities, with clustering phenomena becoming more prominent at higher solid loadings. Cross-sectional velocity profiles visualized the early evolution of particle clustering, wall interactions, and domain depletion. Increased flow velocity enhanced particle suspension but reduced domain uniformity over time. Time-resolved analyses of pressure drop, drag force, particle velocity, interparticle forces, and radial migration were conducted to explore flow regime transitions and mechanical resistance. Early clustering near the pipe walls was observed under dense flow conditions, driven by cohesive and frictional forces, leading to partial stratification and localized energy dissipation. The study highlights the importance of considering early-time flow dynamics, where suspension quality and transport resistance are most sensitive to hydrate loading. These findings contribute to a deeper understanding of hydrate slurry transport in multiphase pipeline systems and offer practical guidance for improving flow assurance models and mitigation strategies in subsea energy operations.