Deep-learning-based coupled flow-geomechanics surrogate model for CO2 sequestration

A deep-learning-based surrogate model capable of predicting flow and geomechanical responses in CO2 storage operations is presented applied. The 3D recurrent R-U-Net combines deep convolutional neural networks to capture the spatial distribution temporal evolution saturation pressure fields 2D surface displacement maps. method trained using high-fidelity simulation results for 2000 storage-aquifer realizations characterized by multi-Gaussian porosity log-permeability fields. Detailed comparisons between full-order new are presented. saturation, provided display a high degree accuracy, both individual ensemble statistics. applied with rejection sampling procedure data assimilation. Although (synthetic) observations consist only small number measurements, significant uncertainty reduction buildup at caprock achieved.