Evaluation of saturation changes during gas hydrate dissociation core experiment using deep learning with data augmentation

• DOI: 10.1016/j.petrol.2021.109820
• Publication Date: 2021-11-13T21:45:27Z
• Authors: Sungil Kim; Kyungbook Lee; Minhui Lee; Jaehyoung Lee; Taewoong Ahn; Jung-Tek Lim
• Keywords: Saturation (graph theory); Clathrate hydrate

This study proposes a reliable evaluation method for three-phase saturation (water, gas hydrate (GH), and gas) during the GH dissociation core experiment using deep learning. A convolutional neural network (CNN) takes computed tomography (CT) images obtained as an input provides output. Although machine/deep learning methods have been applied to from CT in previous research, they were not due lack of adequate amount training data where model could find appropriate parameters. Besides, non-zero showed it was supposed be zero. improved solved problem by acquirement extra application augmentation with CNN. The results CNN presented 34% 29% error compared those random forest. brought 85% 44% its variance without augmentation, respectively. Consequently, based on domain knowledge GH, when comes robustness composition consistency performance, can boosted augmentation.