Data-driven modeling of heavy oil viscosity in the reservoir from geophysical well logs

Viscosity is the most crucial fluid property on recovery and productivity of hydrocarbon reservoirs, more particularly heavy oil reservoirs. In heavy and extra heavy oil reservoirs e.g. bitumen and tar sands more energy is required to be injected into the system in order to decrease the viscosity to make the flow easier. Therefore, attempt to develop a reliable and rapid method for accurate estimation of heavy oil viscosity is inevitable. In this study, a predictive model for estimating of heavy oil viscosity is proposed, utilizing geophysical well logs data including gamma ray, neutron porosity, density porosity, resistivity logs, spontaneous potential as well as P-wave velocity and S-wave velocity and their ratio (Vp/Vs). To this end, a supervised machine learning algorithm, namely least square support vector machine (LSSVM), has been employed for modeling, and a dataset was provided from well logs data in a Canadian heavy oil reservoir, the Athabasca North area. The results indicate that the predicted viscosity values are in agreement with the actual data with correlation coefficient (R²) of 0.84. Furthermore, the outlier detection analysis conducted shows that only one data point is out of the applicability of domain of the develop model.