Spatiotemporal interpretation of three-phase saturation behaviors in gas hydrate formation and dissociation through deep learning modeling

This study provides an interpretation of the three-phase saturation (water, gas, gas hydrate (GH); SW, SG, SGH) in the GH cores during GH formation and depressurization experiments. The saturations are predicted in real-time based on X-ray computed tomography (CT) images using the verified deep learning model reported earlier (Kim et al., 2022), using the convolutional neural network (CNN) with data augmentation. The interpretation explains the saturation behaviors spatiotemporally: depressurization schedule from 20 MPa (gauge pressure) to 0 MPa (ambient pressure), in terms of time and 64 CT images in the longitudinal direction of the core, in terms of space. Time-wise, there is a difference in pressure propagation from the inlet (1st slice) to the outlet (64th slice) within the core during the depressurization for GH dissociation. This transient status leads to increasing variations in SW and SGH near the dissociation pressure (16 MPa) and declining variations at the end of depressurization (0 MPa). Space-wise, before GH dissociation, SGH is highly saturated near the outlet, while it is low near the inlet. This gradational trend in SGH within the core results from the spatial distribution of the initial SW in the preliminary stage; GH was formed in high amounts with high SW as long as sufficient amounts of methane were provided. This spatiotemporal analysis of the saturation behaviors is in accordance with the domain knowledge. Therefore, a spatiotemporally integrated analysis including not only the proposed deep learning model but also a reasonable explanation for the saturation behaviors will help future research on GH-related.