Underground hydrogen storage (UHS), considered a viable solution for large-scale storage, raises concerns about the integrity and performance of reservoir and caprock formations under hydrogen exposure. This study investigates the volumetric and hydraulic properties alterations of different type of rocks under the influence of hydrogen, through data-driven analysis by employing the random forest (RF) algorithm, a machine learning (ML) technique. Data have been collected from the existing literature which relate to porosity and permeability changes and calculation of hydrogen diffusion coefficients after the rock formations have been exposed to hydrogen. Variables such as the initial rock properties, type of rocks and environmental conditions are included as features in the ML models. For porosity and permeability, the most influential factors found, are the type of rock and its initial porosity and permeability values, with low-porosity rocks like shales showing higher sensitivity to hydrogen exposure, especially under high pressure (>10 MPa) and high temperature (>100°C). Based on the measurements, a unified Kozeny-Carman type equation across lithologies is derived, which can be used in reservoir mathematical models. In predicting hydrogen diffusion, initial porosity, pressure, and hydrogen concentration were the most important variables, with strong interactions observed between porosity and insitu conditions such as pressure, temperature and hydrogen exposure duration. Based on the feature importance results, the Chapman-Enskog equation was also fitted to the data to predict diffusivity, primarily for sandstone formations, and could also be used for modelling. The findings highlight clear gaps in the existing experimental literature and indicate the need for additional laboratory studies targeting under-represented combinations of operating conditions.
扫码关注我们
求助内容:
应助结果提醒方式:
