TY - GEN
T1 - Fault mapping of the Gassum Formation reservoir and the Fjerritslev Formation caprock interval at the Stenlille gas storage site using a pre-trained convolutional neural network
AU - Lorentzen, Mads
AU - Bredesen, Kenneth
AU - Gregersen, Ulrik
AU - Smit, Florian
AU - Laghari, Shahjahan
N1 - Conference code: 16
PY - 2022/11/15
Y1 - 2022/11/15
N2 - Fault mapping provides important information for defining compartments in reservoirs and for investigating caprock integrity. However, due to complex fault geometries, manual interpretations based on seismic data and seismic attributes can be timeconsuming and ambiguous. In this study, a convolutional neural network (CNN) trained on synthetic data is applied to 3D post-stack seismic data from a Danish onshore aquifer gas storage facility in the town of Stenlille, which is currently being considered as a demonstration site for geological storage of CO2. Comparison with a manual fault interpretation based on traditional seismic attributes shows that the neural network predicts faults with more details and faults that were overlooked in the manual interpretation. The neural network predictions are, however, in some cases patchy and lack coherence, which may lead to erroneous fault predictions. Therefore, the CNN model should be treated as an additional fault interpretation tool for the interpreter to quality check in a critical manner. Nonetheless, the method represents a novel fault mapping tool that can be useful for de-risking future geothermal and carbon capture storage and utilization prospects.
AB - Fault mapping provides important information for defining compartments in reservoirs and for investigating caprock integrity. However, due to complex fault geometries, manual interpretations based on seismic data and seismic attributes can be timeconsuming and ambiguous. In this study, a convolutional neural network (CNN) trained on synthetic data is applied to 3D post-stack seismic data from a Danish onshore aquifer gas storage facility in the town of Stenlille, which is currently being considered as a demonstration site for geological storage of CO2. Comparison with a manual fault interpretation based on traditional seismic attributes shows that the neural network predicts faults with more details and faults that were overlooked in the manual interpretation. The neural network predictions are, however, in some cases patchy and lack coherence, which may lead to erroneous fault predictions. Therefore, the CNN model should be treated as an additional fault interpretation tool for the interpreter to quality check in a critical manner. Nonetheless, the method represents a novel fault mapping tool that can be useful for de-risking future geothermal and carbon capture storage and utilization prospects.
KW - CCS
KW - Carbon capture and storage
KW - Machine learning
KW - Neural network model
KW - Fault mapping
KW - Stenlille
U2 - 10.2139/ssrn.4277405
DO - 10.2139/ssrn.4277405
M3 - Conference article in proceedings
BT - Proceedings of the 16th Greenhouse Gas Control Technologies Conference (GHGT-16) 23-24 Oct 2022
PB - Elsevier
T2 - 16th International Conference on Greenhouse Gas Control Technologies
Y2 - 23 October 2022 through 27 October 2022
ER -