Predicting oil sands viscosity from well logs using an industry provided dataset

Eric A. Rops, Laurence R. Lines

This study is an expansion of the work the author did in the previous CREWES report (Rops and Lines 2015), where it was demonstrated that heavy oil viscosity could be predicted directly from well logs within 25% error using a limited dataset. To further explore this idea, Donor Company has generously provided viscosity data from their Athabasca North and Athabasca South oil sands development projects, with multiple measurements per well.

Multi-attribute analysis enables a target attribute (viscosity) to be predicted using other known attributes (the well logs). In the Athabasca North area, P-wave sonic and Density porosity were used to predict viscosity and the average validation error was 147,000cP, or 19% of the total viscosity range. In the Athabasca South area, medium resistivity, gamma ray, and P-wave sonic were used to predict viscosity and the average validation error was only 70,000 cP, or 13% of the total viscosity range.