Integrated Subtle Fault Prediction Technique Based on Seismic Conditioning and Deep Learning

Abstract With the development of oilfield, the role of subtle faults becomes more and more important. Due to the limitation resolution of seismic data, it is hard for conventional methods to identify and predict subtle faults to meet the needs of reservoir development. In order to improve the accuracy of subtle fault prediction, a new fault prediction technique based on seismic conditioning and deep learning was proposed in this study. The workflow mainly includes the following 6 steps: 1) Seismic conditioning based on regional stress and well data; 2) Forward modeling of fault models using different frequencies of wavelet, and get the seismic response sample for different throw fault; 3) Seismic attributes optimization based on the frequency and azimuth optimization for different throw of fault; 4) Fast fault identification and prediction for all different frequency and azimuth seismic data by using deep learning method; 5) Integrated the FMI and well data to optimize the different scales faults prediction results, and fusion of multi-scale faults. 6) Output subtle faults predict results after the subtle faults modeling and 3D QC. The application of this new method in M oilfield in the Middle East shows that, the seismic data with higher frequency and NE-SW azimuth can better identify the NW-SE subtle faults. When the main frequency of seismic data is 30Hz, faults with a fault displacement >10m can be clearly identified, and when the main frequency of seismic data is 45Hz, faults with a fault displacement >5m can be clearly identified. Deep learning technology can quickly and effectively extract faults from seismic data. Fault optimization and fusion based on different frequency and azimuth seismic data can predict multi-scale fault system more reasonably. A series of small NW-SW faults are identified in the western part of the long axis anticline in the study area. These subtle faults cut the long axis anticline and formed a fault block traps with lower OWC in the southwest slope, which explains why the OWC and formation pressure in the west are different from east. 7 new wells drilled through the fault, also confirms the presence of NW-SE subtle fault. The subtle fault prediction techniques based on deep learning and model forward modeling can better integrate the geological faults information into the deep learning model samples to improve the identification ability and prediction accuracy of multi-scale faults. The optimization of seismic frequency and azimuth can fundamentally improve the effect of deep learning and prediction of faults. High accuracy subtle fault prediction results can deepen reservoir understanding and improve for efficient oilfield development.