A Machine-learning-assisted 3D Geologic Model for the Late Devonian Duvernay Formation, East Shale Basin, Western Canada
The Late Devonian Duvernay Formation, previously evaluated as the primary source rock for conventional reservoirs within the Western Canada Basin (comprising West and East Basins), has been recently explored and exploited as unconventional shale reservoirs with estimated reserves of 483–820 TCF gas and 67–208 billion barrels of oil. Although several successful development projects have been conducted in the Duverney Formation within the West Basin, the East Basin has attracted less investor attention because of lower kerogen maturity and organic matter dilution by higher sedimentation rates. However, these indicators of a lower quality reservoir have been challenged by horizontal drilling data suggesting otherwise. A 3-D geologic facies model based on petrophysical data is a vital component of the reservoir evaluation process because it provides the spatial distribution of facies and related petrophysical parameters for reserve calculations and reservoir simulations. However, due to technical and financial constraints, core logging, which is the most accurate input for a facies model, was not conducted at all wells in the research area. Instead of using core logs, geologic facies interpreted from wireline logging can be used as substitutes. In this study, a machine-learning bi-directional long short-term memory network was trained on 20 wells and used to predict geologic facies in 14 wells without core logs. Ultimately, facies data from the 34 wells were input to a geostatistical sequential Gaussian simulator. By comparing modeled cross-sections with and without machine-learning estimates, the 3-D model with all the data yielded improved spatial representations of the geologically interpreted stratigraphic sequence with better vertical resolution resulting in more accurate reserve calculations.
AAPG Datapages/Search and Discovery Article #90351 © 2019 AAPG Foundation 2019 Grants-in-Aid Projects