Dispersion and hydrate formation analysis for CCS/CCUS in depleted gas reservoirs

Depleted gas reservoirs are attractive targets for CO₂ storage, and CO₂ injection in such reservoirs can bring additional gas production. Technical risks and economic potential associated with CCS/CCUS in depleted gas reservoirs need to be investigated thoroughly. Dispersion is one of the critical phenomena for CO₂ Enhanced Gas Recovery (CO₂-EGR). Since CO₂-EGR is miscible displacement, dispersion takes place to mix injected CO₂ and in-situ methane, significantly impacting gas recovery. However, the dispersion itself and its modeling in a field scale are still uncertain. This study aims to develop a fundamental understanding of dispersion and propose a consistent simulation approach for dispersion during CO₂-EGR. We perform reservoir simulation studies using stochastic heterogeneous permeability fields to provide insights as to how dispersivity is physically generated inside heterogeneous porous media at different scales. Our study shows that the variance of permeability and convective spreading are the primary causes of dispersivity at any scale. Although geoscientists often assume numerical dispersion can represent physical dispersion, our study indicates this is an oversimplification and could cause significant errors in calculated gas recovery estimation. Heterogeneity is always essential for the dispersion growth process and the final displacement behavior. It must be modeled correctly in reservoir simulations. Hydrate formation is another issue during CO₂ storage. CO₂ injection into depleted formations could lead to hydrate formation near the wellbore due to Joule-Thomson cooling, which might cause injectivity issues. Numerical modeling often requires excessive computational time and effort for the analysis. This study aims to propose a novel approach for hydrate risk assessment using physics-based Machine Learning (ML). Input parameters for the ML models are selected considering their importance in hydrate forming, and the ML models are tuned and tested using datasets from numerical reservoir simulation results. To the best of our knowledge, this is the first ML application for hydrate risk assessment during CO₂ storage in depleted gas reservoirs. The ML models are capable of predicting hydrate-forming events with high predictability for test datasets with 95% recall and 84% precision. These results imply that ML models can be further utilized as risk assessment tools for screening future CCS projects.