Prediction of phase equilibria for multicomponent hydrogen hydrates with thermodynamic promoters using Bayesian-optimized interpretable XGBoost

Abstract

The accelerating exploration of natural hydrogen and the development of hydrate-based gas storage technologies have ignited significant interest in the phase behavior of hydrogen-containing hydrates. However, accurately determining the phase equilibrium conditions of multicomponent hydrogen hydrates, particularly when thermodynamic promoters are introduced, remains a formidable challenge. From an experimental perspective, accurately measuring the phase equilibrium conditions of these complex high-pressure systems is notoriously difficult and hazardous. From a theoretical standpoint, traditional thermodynamic models struggle to describe the severe non-ideality introduced by multicomponent gas mixtures and liquid promoters, frequently suffering from parameter explosion and iterative non-convergence. To bridge these gaps, this study proposes a data-driven model based on a Bayesian-optimized interpretable XGBoost algorithm. A comprehensive dataset was established, starting from 921 raw experimental data points and yielding 724 high-quality samples after rigorous physical screening. To enhance practical applicability, three categories of key parameters—system temperature, initial feed gas composition, and promoter concentration—were utilized as input features to predict the phase equilibrium pressure. The optimized model exhibited exceptional predictive performance, achieving a high R² of 0.9421 and MAE of 0.9323 MPa on the independent testing set. Furthermore, the SHAP method was integrated to unbox the model and reveal complex non-linear feature interactions, quantifying the compounding destabilizing penalties of high temperature and hydrogen, uncovering the distinct thermodynamic limitations of conventional co-guests (CH₄ and CO₂), and highlighting that the absolute pressure-reducing efficacy of liquid promoters (THF and TBAB) is markedly augmented under high-temperature and hydrogen-rich environments. This work not only provides a highly accurate, non-iterative prediction tool but also offers valuable physical insights into the phase behavior of complex hydrogen hydrates, providing reliable theoretical support for natural hydrogen exploration and the optimal design of solid-state hydrogen storage systems.