We propose a volume-to-volume network to predict a 3D relative geologic time (RGT) volume directly from the input seismic volume. The RGT volume stores all the geological structural information from the seismic volume, leading to a full 3D seismic structural 3D interpretation. We train the network by using automatically generated numerous 3D synthetic seismic volumes and the corresponding RGT targets. Although trained with only synthetic datasets, the network learns to automatically capture the complex geological structures from the input amplitude volume. Multiple field examples show that the network is significantly superior to the conventional methods in RGT estimation. Seismic horizons, even those across faults, can be accurately tracked by simply extracting iso-surfaces of the predicted RGT volumes. In addition to horizons, faults and fault throws are also indicated in the RGT volumes.

Presentation Date: Monday, October 12, 2020

Session Start Time: 1:50 PM

Presentation Time: 4:20 PM

Location: 351F

Presentation Type: Oral

Keywords:
reservoir characterization, deep learning, machine learning, neural network, data augmentation, artificial intelligence, upstream oil & gas, training dataset, pattern recognition, us government
Subjects:
Reservoir Characterization, Information Management and Systems, Seismic processing and interpretation, Artificial intelligence, Neural networks
This content is only available via PDF.
2020. Society of Exploration Geophysicists
You can access this article if you purchase or spend a download.