Estimating permeability impairment due to asphaltene deposition during the natural oil depletion process using machine learning techniques

Abstract

Asphaltene deposition causes numerous problems in the petroleum industry, including porous media blockage. To better understand the effect of rock and oil properties, as well as operational parameters, on reservoir permeability reduction due to asphaltene deposition during the natural depletion of oil, this study explored the capability of ensemble learning models, including Extra Trees, Gradient Boosting, Adaboost, XGBoost, LightGBM, and Random Forest, for estimating permeability decline. A set of potential parameters affecting permeability impairment were identified to develop the model, and a dataset of 1189 samples was created and divided into training and testing categories in a 4:1 ratio. The Optuna optimization framework was used to estimate the models hyperparameters. Our findings indicate that eight parameters, including pore volume injected, API, asphaltene content, porosity, injection velocity, permeability, core area, and lithology, are the most important predictors for estimating permeability reduction. Evaluation results using testing data (238 samples) demonstrated that the developed models accurately predicted permeability reduction, of which the Extra Trees method outperformed the others, with Root Mean Square Error (RMSE) and coefficient of determination (R2) values of 0.03 and 0.98, respectively. These findings have implications for the petroleum industry and offer a more affordable and efficient alternative to laboratory studies for evaluating formation damage caused by asphaltene deposition.