Data-driven modeling to predict adsorption of hydrogen on shale kerogen: Implication for underground hydrogen storage

Created by W.Langdon from gp-bibliography.bib Revision:1.8051

@Article{KALAM:2023:coal,

• author = "Shams Kalam and Muhammad Arif and Arshad Raza and Najeebullah Lashari and Mohamed Mahmoud",
• title = "Data-driven modeling to predict adsorption of hydrogen on shale kerogen: Implication for underground hydrogen storage",
• journal = "International Journal of Coal Geology",
• volume = "280",
• pages = "104386",
• year = "2023",
• ISSN = "0166-5162",
• DOI = "doi:10.1016/j.coal.2023.104386",
• URL = "https://www.sciencedirect.com/science/article/pii/S0166516223002045",
• keywords = "genetic algorithms, genetic programming, Hydrogen adsorption, Kerogen, Shale gas reservoirs, Machine learning, Data-driven modeling, Underground hydrogen storage",
• abstract = "The interaction of hydrogen in shale gas formations holds significant interest for long-term subsurface hydrogen storage. Accurately and rapidly predicting hydrogen adsorption in these formations is crucial for assessing underground hydrogen storage potential. Many laboratory experiments and molecular simulations have been conducted to determine hydrogen adsorption. However, laboratory experiments and molecular simulations require complex setups and extensive calculations, which can be time-consuming. Consequently, end-users may prefer quick and accurate prediction of hydrogen adsorption to reduce the experimental and computational burden. This study introduces a novel model for predicting hydrogen adsorption using gradient boosting regression and available molecular simulation data from the literature. The data-driven model predicts hydrogen adsorption on kerogen structures based on pressure, temperature, adsorbed methane, hydrogen-to-carbon ratio, oxygen-to-carbon ratio, and kerogen density. We compared gradient-boosting regression with other machine learning tools, including artificial neural networks, symbolic regression assisted with genetic programming, decision trees, and random forests in terms of their capability to predict H2 adsorption on shale kerogen. A simple mathematical equation based on symbolic regression via genetic programming has also been provided, with training and testing coefficients of determination of 88.4percent and 85.8percent, respectively. However, the digital model created using gradient boosting regression outperformed all other machine learning tools, achieving a coefficient of determination of 99.6percent for training data and 94.6percent for testing data. A sensitivity analysis was also conducted that demonstrates the robustness of the developed model. In the case of kerogen type A, the order of increasing hydrogen adsorption is KIA < KIIA
  }

Genetic Programming entries for Shams Kalam Muhammad Arif Syed Hamid Arshad Raza Najeebullah Lashari Mohamed Mahmoud

Citations