The DNNA method is trained using facies logs from wells simultaneously with seismic data. The principle of DNNA is to combine several neural networks, all with different learning strategies, to segment the seismic data space, according to the facies interpretation provided by geologists or petrophysicists. DNNA thereby provides a convenient bridge between finer scale geology / petrophysics interpretations and coarser scale seismic data which also does not inherently possess geological information. The DNNA algorithm searched through fifteen 3D seismic volumes simultaneously, and was able to build a model which reconstructed the nine lithofacies. The oil-filled packstones facies with no false positives or false negatives seen at the wells. The seismic volumes were: Hi-Res time migrated, Instantaneous Frequency, Instantaneous Q Factor, Inversion (P-Impedance), Semblance, Dominant Frequency, Most negative curvature, and eight angle stacks: 0-5, 5-10, 10-15, 15-20, 20-25, 25-30, 30-35, 35-40 degree ranges.
The neural network learnings were applied through the 3D survey, and results were delivered with up to a 0.5 ms two-way time vertical resolution, or about 5 ft, a significant uplift from conventional seismic resolution. Lateral resolution was also improved. Additional drilling opportunities can identified from the seismic facies thickness map or the facies probability voxel clouds. The bootstrap classification rate, an alternative to total well replacement, was 80%, indicating good prediction quality.
Authors: PETER WANG*, BRUNO DE RIBET*, MONTE MEERS(1), HOWARD “PETE” RENICK(2), RUSS CREATH(3), RYAN MCKEE(4)
* Emerson E&P Software1 Independent Geologist2 Independent Geophysicist3 Reservoir Geophysicist, Hardin International, 5300 Democracy Dr, Suite 100, Plano, TX 750244 Geophysical Technician, RAM Imaging Technology
The Webinar is at 12:00 PM Central USA time on July 19, 2018
Please register using this link: http://info.pdgm.com/permian_wolfcamp_july_19_webinar