Deep learning-assisted Bayesian framework for real-time CO2 leakage locating at geologic sequestration sites

• DOI: 10.1016/j.jclepro.2024.141484
• Publication Date: 2024-03-01T10:45:22Z
• Authors: Xupeng He; Weiwei Zhu; Hyung Kwak; Ali Yousef; Hussein Hoteit

Accurate and efficient localization of CO2 leakage if occurred in subsurface formations, is significant importance achieving secure geological carbon sequestration (GCS) projects. However, this task inherently challenging due to the considerable uncertainties subsurface. In work, we develop a novel deep learning-assisted Bayesian framework for identifying potential sites based on reservoir pressure transient behavior measured at wellbores injection or observation wells. The method consists two essential steps: 1) Deep learning surrogate: This step aims effectively replace intensive high-fidelity simulation with an surrogate. 2) inversion: step, posterior distributions locations are inverted, which surrogate serves as forward model. above processes automated using optimization instead labor-intensive trial-and-error approach. proposed verified 3D model simulating into brine-filled reservoir. results demonstrate Bayesian-optimized could successfully capture underlying process CO2-brine flow. inversion algorithm enables localizing high accuracy. To our knowledge, implemented first time locate multiple field scale. workflow provides accurate approach detecting possible real-time manner has promising field-scale GCS applications.