The potential for large-scale storage of carbon dioxide (CO2) through Geological Carbon Sequestration (GCS) in deep geological formations such as saline aquifers and depleted oil and gas reservoirs is significant. Effectively implementing GCS requires evaluating the risk of plume confinement and storage capacity at each site through a thorough assessment. To assess the stability of the caprock after CO2 injection, efficient tools are needed to evaluate the safe duration of CO2 injection. This study used Particle Swarm Optimization (PSO) evolutionary algorithm to optimize the maximum CO2 storage capacity in saline aquifers without risking the integrity of the caprock. A deep learning (DL) model, fully connected neural networks, was trained to predict the safe injection duration. The movement of CO2 was simulated for 170 years following a 30-year injection period into a deep saline aquifer using a physics-based numerical reservoir simulator. The simulation took into consideration uncertainty variables such as petrophysical properties and reservoir physical parameters, as well as operational decisions like injection rate and perforation depth. Sampling the reservoir model with the Latin-Hypercube approach accounted for a range of parameters. Over 720 reservoir simulations were performed to generate training, testing, and validation datasets, and the best DNN model was selected after multiple executions. The three-layer FCNN model with 30 neurons in each layer showed excellent prediction efficiency with a coefficient of determination factor over 0.98 and an average absolute Percentage Error (AAPE) less than 1%. The trained models showed a good match between simulated and predicted results and were 300 times more computationally efficient. PSO was utilized to optimize the operational parameters in the DL models to achieve maximum CO2 storage with minimum damage to the caprock. The results suggest that the DNN-based model can serve as a reliable alternative to numerical simulation for estimating CO2 performance in the subsurface and monitoring storage potential in GCS projects.
Bibliographical noteKAUST Repository Item: Exported on 2023-06-06
Acknowledged KAUST grant number(s): BAS/1/1351-01-01, BAS/1/1423-01-01, FCC/1/4491-22-01, URF/1/4074-01-01
Acknowledgements: B.Y. and Z.T. thanks KAUST for the Research Funding through the grant BAS/1/1423-01-01 and FCC/1/4491-22-01. S.S. and Z.T. thanks KAUST for the Research Funding through the grant BAS/1/1351-01-01 and URF/1/4074-01-01. The authors want to acknowledge the Computer Modeling Group for the academic license of CMG-GEM.