Scholarlink

<p>The transition to renewable energy requires innovative hydrogen storage solutions. Underground hydrogen storage (UHS) in geological formations is gaining attention for its scalability and long-term potential. Accurately predicting the wettability of hydrogen-brine-rock systems is essential for hydrogen containment, capillary sealing, and storage security in UHS, but remains a research challenge. Previous studies have been limited by small datasets, narrow substrate diversity, and insufficient analysis of variables such as pressure, temperature, and salinity. This study addresses these limitations by using an expanded dataset of 599 measurements across 28 substrate types and subsurface conditions. Advanced machine learning models, including gradient boosting, extremely randomized trees, extreme gradient boosting, and categorical boosting, were developed to predict contact angle values with high accuracy. Gradient boosting performed best, achieving an average mean absolute error of 3.117 degrees and a coefficient of determination of 0.915 in sevenfold cross-validation. SHapley Additive exPlanations analysis identified substrate type as the primary factor, followed by temperature, pressure, and salinity, offering interpretable insights for lithology screening and reservoir selection. Model uncertainty was assessed using bootstrap resampling and cross-validation, which highlighted pressure–temperature regimes with higher predictive uncertainty. These results provide a scalable, interpretable, and uncertainty-aware framework to support engineering decisions for site selection, capillary sealing assessment, and operational optimization in UHS systems.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Uncertainty-aware machine learning for predicting wettability in hydrogen storage: Informed by diverse geological substrates

  • 期刊论文

  • 发表日期：

    2026年3月10日

查看全文
  • Ahmad Azadivash,
  • Atrina Seifirad,
  • Hosseinali Soleymani

摘要

The transition to renewable energy requires innovative hydrogen storage solutions. Underground hydrogen storage (UHS) in geological formations is gaining attention for its scalability and long-term potential. Accurately predicting the wettability of hydrogen-brine-rock systems is essential for hydrogen containment, capillary sealing, and storage security in UHS, but remains a research challenge. Previous studies have been limited by small datasets, narrow substrate diversity, and insufficient analysis of variables such as pressure, temperature, and salinity. This study addresses these limitations by using an expanded dataset of 599 measurements across 28 substrate types and subsurface conditions. Advanced machine learning models, including gradient boosting, extremely randomized trees, extreme gradient boosting, and categorical boosting, were developed to predict contact angle values with high accuracy. Gradient boosting performed best, achieving an average mean absolute error of 3.117 degrees and a coefficient of determination of 0.915 in sevenfold cross-validation. SHapley Additive exPlanations analysis identified substrate type as the primary factor, followed by temperature, pressure, and salinity, offering interpretable insights for lithology screening and reservoir selection. Model uncertainty was assessed using bootstrap resampling and cross-validation, which highlighted pressure–temperature regimes with higher predictive uncertainty. These results provide a scalable, interpretable, and uncertainty-aware framework to support engineering decisions for site selection, capillary sealing assessment, and operational optimization in UHS systems.