Abstract
The residual oil saturation (Sor) evaluation is relevant for developing oil fields, standing out as an input to flow simulation models for production forecasting. Also, Sor understanding is crucial to guide enhanced oil recovery techniques. Moreover, Sor laboratory measurement tends to be time-consuming and expensive. This work aims to understand and predict Sor from X-ray mu Ct and RCAL data employing Ensemble Learning techniques (AdaBoost, Gradient Boost, and XGBoost) in Pre-salt carbonates of the Barra Velha Formation, Santos Basin. Morphological attributes related to pore size, shape, and orientation were extracted from X-ray mu Ct scans. Hence, these attri-butes, together with routine core analysis (RCAL) data, were used to build machine learning (ML) models for the prediction of Sor. The results indicated strong faciological control in Sor, where the genesis of the rock implies different characteristics of the porous framework, impacting Sor and other petrophysical features. Rocks with larger pores usually lead to larger heterogeneity, which tends to trap more oil. Furthermore, the shape and orientation of the pores have substantial faciological control, given the textural organization of the different rock facies. These attributes showed weak control over Sor, impacting each type of facies differently, depending on the rock fabric. Even though the ML algorithms have similar results, the Gradient Boosting showed the best results. Furthermore, the inclusion of RCAL data does not increase the accuracy of the models. So, it is possible to predict the Sor only with morphological pore attributes reasonably. The most important features are mainly related to pore size and subordinately to orientation, confirming their impact on Sor. Finally, this methodology, in addition to predicting and bringing understanding to Sor in Pre-Salt rocks, can be adapted for use in other reservoirs.
NSTL
Elsevier