Predicting oil sands viscosity from well logs, NMR logs, and calculated seismic properties

Eric A. Rops and Laurence R. Lines

ABSTRACT

This study is an expansion of the work from last year where it was demonstrated that oil sands viscosity could be predicted directly from standard well logs within 13% error (or 0.72 of one standard deviation) using a real viscosity dataset from Donor Company. This work has been expanded by: normalizing the well logs, using seismic properties calculated from well logs to predict viscosity, adding NMR logs as predictors, improving the viscosity training model, and including reservoir depth as a predictor.

Multi-attribute analysis enables a target attribute (viscosity) to be predicted using other known attributes (the well logs). The top well logs for predicting viscosity were: resistivity, gamma ray, SP, NMR Total Porosity, NMR Free Porosity, and S-wave sonic. They successfully predicted viscosity with an average validation error of 69,000cP (or 0.69 of one standard deviation). The top seismic properties for predicting viscosity were: P-wave velocity and P-Impedance. They predicted viscosity with an average validation error of 94,000cP (or 0.94 of one standard deviation). The well logs modeled more viscosity variations than the calculated seismic properties did, and in most cases including depth as a predictor improved the prediction.

View full article as PDF (5.12 Mb)