Improving Wellbore Placement Accuracy Using Stratigraphic Misfit Heatmaps

Geosteering of wells requires fast interpretation of geophysical logs which is a non‐unique inverse problem. Current work presents a proof‐of‐concept approach to multi‐modal probabilistic inversion of logs using a single evaluation of an artificial deep neural network (DNN). A mixture density DNN (MDN) is trained using the ”multiple‐trajectory‐prediction” loss functions, which avoids mode collapse typical for traditional MDNs, and allows multi‐modal prediction ahead of data. The proposed approach is verified on the real‐time stratigraphic inversion of gamma‐ray logs. The multi‐modal predictor outputs several likely inverse solutions/predictions, providing more accurate and realistic solutions compared to a deterministic regression using a DNN. For these likely stratigraphic curves, the model simultaneously predicts their probabilities, which are implicitly learned from the training geological data. The stratigraphy predictions and their probabilities obtained in milliseconds from the MDN can enable better real‐time decisions under geological uncertainties.

show abstract

“…Gee et al. (2020) and Maus et al. (2020) developed a method for tracking several solutions using stratigraphic misfit heatmaps.…”

Section: Introductionmentioning

confidence: 99%

“…The look-up takes several minutes, and cloud solutions are necessary for real-time performance. Gee et al (2020) and Maus et al (2020) developed a method for tracking several solutions using stratigraphic misfit heatmaps. Their multi-modal inversion is proprietary and takes about a minute.…”

mentioning

confidence: 99%

Direct Multi‐Modal Inversion of Geophysical Logs Using Deep Learning

Alyaev

Elsheikh

2022

Earth and Space Science

View full text Add to dashboard Cite

show abstract

“…The look-up takes several minutes, and cloud solutions are required for real-time performance. Gee et al (2020) and Maus et al (2020) developed a method for tracking several solutions using stratigraphic misfit heatmaps. Their multi-modal inversion is proprietary and takes about a minute.…”

Section: Introductionmentioning

confidence: 99%

Direct multi-modal inversion of geophysical logs using deep learning

Alyaev¹,

Elsheikh²

2022

Preprint

View full text Add to dashboard Cite

Geosteering of wells requires fast interpretation of geophysical logs which is a nonunique inverse problem. Current work presents a proof-of-concept approach to multimodal probabilistic inversion of logs using a single evaluation of an artificial deep neural network (DNN). A mixture density DNN (MDN) is trained using the "multiple-trajectoryprediction" (MTP) loss functions, which avoids mode collapse typical for traditional MDNs, and allows multi-modal prediction ahead of data. The proposed approach is verified on the real-time stratigraphic inversion of gamma-ray logs. The multi-modal predictor outputs several likely inverse solutions/predictions, providing more accurate and realistic solutions compared to a deterministic regression using a DNN. For these likely stratigraphic curves, the model simultaneously predicts their probabilities, which are implicitly learned from the training geological data. The stratigraphy predictions and their probabilities obtained in milliseconds from the MDN can enable better real-time decisions under geological uncertainties.

show abstract

Automated Geosteering with Fault Detection and Multi-Solution Tracking

Maus

Gee

Mitkus

et al. 2020

IADC/SPE International Drilling Conference and Exhibition

View full text Add to dashboard Cite

Development of autonomous drilling technologies requires the automated analysis and interpretation of Logging While Drilling (LWD) data to optimally land the well in the target formation and keep it in the pay zone. This paper presents a fully automated geosteering algorithm, which includes advanced LWD filtering, fault detection, correlation, tracking of multiple interpretations with associated probabilities and visualization using novel stratigraphic misfit heatmaps. Traditional geosteering uses manual stretch, compress and match techniques to correlate measurements along the subject wellbore against corresponding reference type logs. This results in a crude representation of strata by linear sections with offsets at fault locations. Instead of automating this manual process, we instead determine the possible interpretations as solutions of a geophysical inverse problem in which the total misfit between the subject and reference data is minimized. Interpretations are parameterized as discontinuous splines to accurately follow curved strata interjected by fault offsets. To account for ambiguities, multiple possible interpretations are continuously tracked in real time and assigned probabilities based on the misfit between the latest measurements and the reference data. Unrealistic solutions are suppressed by penalizing strong curvature and large fault offsets. Viable interpretations are simultaneously visualized in real time as paths on a novel stratigraphic misfit heat map, where they may be corroborated against valleys of minimal misfit between the subject and reference data. The user can guide the interpretation by setting control points on the heat map which the automated solutions must respect. The algorithm has been validated using wells from different regions across North America for which previous manual geosteering interpretations are available. The automated spline interpretations represent the actual curved strata more accurately than manual interpretations. Operationally, the automated interpretations can be provided within minutes compared to typical manual turn-around times of hours. Automation leads to more consistent and repeatable results, removing the subjectivity of manual interpretations.

show abstract

Improving Wellbore Placement Accuracy Using Stratigraphic Misfit Heatmaps

Cited by 7 publications

References 8 publications

Direct Multi‐Modal Inversion of Geophysical Logs Using Deep Learning

Direct Multi‐Modal Inversion of Geophysical Logs Using Deep Learning

Direct multi-modal inversion of geophysical logs using deep learning

Automated Geosteering with Fault Detection and Multi-Solution Tracking

Contact Info

Product

Resources

About