The presentation ”Predicting likely stratigraphy realizations from shallow logs with AI” will be given by Sergey Alyaev from NORCE Norwegian Research Centre.
Geosteering of wells requires fast petrophysical and geological interpretation of logs, often forming non-unique inverse problems. We developed a multi-modal probabilistic interpretation method using a single evaluation of an artificial deep neural network (DNN) in milliseconds. A mixture density DNN is trained using the “multiple-trajectory-prediction” loss functions. It outputs several stratigraphic solutions and their probabilities and predictions of stratigraphy ahead of data. Thus, it yields an AI that learns the ability to match the logs and select likely geological scenarios from the training data. The proposed approach is verified on a stratigraphic interpretation of gamma-ray and rate-of-penetration logs from drilling in an unconventional field. For a single chunk of data, the multi-modal predictor outputs several viable inverse solutions/predictions, providing more accurate and realistic solutions than a deterministic regression using a DNN. Applying this method sequentially (using the precious interpretations as starting points) allows tracking several likely stratigraphic realizations throughout a geosteering operation or until new data disconfirm them. Multi-solution tracking is vital for drilling operations where only a little, noisy, or low-quality data is available. It enables informed decisions from quantified geological uncertainties.
This presentation is based on the work produced in the Center for Research-based Innovation DigiWells (NFR SFI project no. 309589, https://DigiWells.no).
Alyaev, S., & Elsheikh, A. H. (2022). Direct Multi‐Modal Inversion of Geophysical Logs Using Deep Learning. Earth and Space Science, 9(9), e2021EA002186
Alyaev, S., Ambrus, A., Jahani, N., & Elsheikh, A. H. (2022). Sequential Multi-Realization Probabilistic Interpretation of Well Logs and Geological Prediction by a Deep-Learning Method. In SPWLA 63rd Annual Logging Symposium. OnePetro
Sergey Alyaev holds MSc and PhD degrees in applied and computational mathematics from the University of Bergen. He works as a senior researcher at NORCE, where he develops mathematical methods for problems in flow-modeling drilling and geosteering. His research interests include forward and inverse modeling, uncertainty quantification, data-driven methods, and decision theory; and their real-time applications.