Improved 3D neural network architecture for fault interpretation on field data
Enning Wang, Maisha Amaru, S. Jayr, Barton Payne · 2021
Fault interpretation is a critical step for building reservoir models, mitigating drilling and production hazards such as reservoir compartmentalization. Estimating fault probability from 3D seismic using Deep Learning based image segmentation techniques has been widely studied and applied throughout the industry to speed up manual picking and improve interpretation quality and consistency. Despite the success in shallow sediments with high signal to noise ratio and sharp faults, applying the technology to field seismic data at reservoir depth or even subsalt poses generalization challenges due to divergences in scale, resolution and noise levels between training and prediction data. Existence of coherent noise, illumination shadow and spatially varying scale further complicate the problem. We deal with these issues by modifying the architecture of a classic 3D U-Net to utilize a lightweight encoder and multi-scale feature extractor and increase the 3D subcube size to broaden the field of view for superior fault continuity.