Integrating Electrofacies and Well Logging Data Into Regression and Machine Learning Approaches for Improved Permeability Estimation in a Carbonate Reservoir In a Giant Southern Iraqi Oil Field

Abstract

Abstract Integrating discrete facies classification into the estimation of formation permeability is a crucial step to improve reservoir characterization and to preserve heterogeneity quantification. Therefore, it is essential to obtain the most accurate estimation of permeability in non-cored intervals in order to attain realistic reservoir characterization and modeling. In our most recent paper [OTC-30906-MS], the electrofacies classifications have been conduced for a well from a carbonate reservoir in a Giant Southern Iraqi oil field. The same predicted discrete electrofacies distribution was included in this paper along with well logging interpretations to model and predict the reservoir core permeability for all wells. The well logging interpretations that were included in permeability modelling are neutron porosity, shale volume, and water saturation as a function of depth. The regression and machine learning approaches adopted for permeability modelling are multiple linear regression (MLR), smooth generalized additive Modeling (SGAM) and Random Forest (RF) Algorithm. The classified electrofacies were considered as a discrete independent variable in the core permeability modelling to provide different model fits given each electrofacies type in order to capture the different permeability variances. The matching visualization between the observed and predicted core permeability, the computed root mean square prediction error and adjusted squared R were considered as validation and accuracy tools to compare between the three modelling approaches. Since there are too many intervals with missing core permeability measurements, the modelling was first adopted on the intervals that have permeability readings (known subset). The prediction was then conducted given the same known permeability intervals in addition to the entire dataset (full dataset). The root mean square prediction error and adjusted squared R for the Random Forest were significantly better than in both MLR and SGAM for the known subset and full dataset. It can be concluded that combining the electrofacies in one permeability model has accurate, fast and an automation procedure of prediction for other wells. The two machine learning algorithms were implemented by R software, the most powerful statistical programming language.