International Petroleum Technology Conference, Год журнала: 2025, Номер unknown
Опубликована: Фев. 17, 2025
Abstract The role of hydrogen geo-storage and production in addressing global warming energy demand concurrently cannot be understated. Diverse factors such as interfacial tension (IFT) wettability influence safe effective production. IFT controls the maximum H2 storage column height, capacity, capillary entry pressure. Current laboratory experimental techniques for determination H2/cushion gas systems are resource-intensive. Nonetheless, extensive data supports machine learning (ML) deployment to determine time-efficiently cost-effectively. Hence, this work evaluated predictive capabilities supervised ML paradigms including random forest, extra trees regression, gradient boosting regression (GBR), light machine, wherein novelty study lies. An comprehensive dataset comprising 2564 instances was gathered from literature, encompassing independent variables: pressure 0.10–45 MPa), temperature (20–176 °C), brine salinity (0–20 mol/kg), hydrogen, methane, carbon dioxide, nitrogen mole fractions (0-100 mol.%). pre-processed split into 70% model training 30% testing. Statistical metrics visual representations were utilized quantitative qualitative assessments models. Leverage approach subsequently applied classify different categories verify statistical validity database reliability constructed paradigms. impact variables on prediction using Spearman correlation, permutation importance, Shapley Additive Explanations (SHAP). Nitrogen CO2 demonstrated least greatest gas/brine based correlation analysis, SHAP. Generally, developed successfully captured underlying relationships between IFT, recording an overall R2 > 0.97, MAE < 1.30 mN/m, RMSE 2 AARD 2.3% GBR superior performance, yielding highest lowest MAE, RMSE, 0.987, 0.507 0.901 0.906%, respectively. also provided more accurate results pure H2/water than empirical molecular dynamics-based correlations by other scholars. Only 0.43–2.11% outside range, underscoring beneficial tools toolbox domain experts, which could fast-track workflows minimize uncertainties surrounding conventional aqueous systems. This progress is promising mitigating loss optimizing strategies
Язык: Английский
