Viscosity is a critical parameter in selecting the best recovery method to exploit a heavy oil reservoir. While heavy oil viscosities can be measured in the lab from well samples, it would be very useful to have a method to reliably estimate heavy oil viscosity from well logs. In this study, data from thirteen wells were obtained from the Athabasca region of northern Alberta. Each well has laboratory oil viscosity measurements, as well as dipole sonic logs, and a full suite of the standard well log curves.
Multi-attribute analysis enables a target attribute to be predicted using other known attributes. In this study, the available well log curves were used to predict viscosity. Five wells were used to train the relation to blindly predict the viscosity of the remaining wells. Four out of the seven remaining wells successfully predicted the viscosity comfortably below an error bound of 25%. The remaining three wells predicted the viscosity above the error bound of 25%. It was found that the shear sonic is the most important viscosity predictor. Further observations suggested that viscosity predictions are most accurate when there is separation between the deep, medium, and shallow resistivity curves.
View full article as PDF (2.55 Mb)