Porosity and permeability represent the main parameters for an accurate petrophysical evaluation. These parameters are often evaluated either from well logs interpretation or core data. The wireline logs provide continuous measurements of physical rock properties and can be interpreted to provide porosity and permeability indirectly. Thus, it has to be inferred through relationships with core data from the same field or well or from empirically derived equations. Another approach is to model the relationship between porosity and permeability from core data which provide more accurate estimations but are expansive and cannot be acquired at every depth on every well. In addition, machine learning technics gained a lot of importance in solving similar problems. To produce a continuous permeability from a computed porosity in any well, we use statistical analysis on core data to obtain a correlation between porosity and permeability for a particular formation. This paper aims to generate permeability-porosity data-driven models for the Bakken formation, representing an unconventional reservoir within the Williston Basin in the US, using 426 core data with a wide range of porosity and permeability. Different machine learning algorithms have been developed including Linear Regression (LR), Artificial Neural Network (ANN), Random Forest Regressor (RFR), Extreme Gradient Boosting (XGBoost), Adaptive Booster Regressor (AdaBoost), and Support Vector Regression (SVR), to predict the permeability from porosity. Evaluating the obtained correlation and the machine learning algorithms was based on the R2 score, the Minimum Squared Error (MSE), and the Mean Absolute Error (MAE) as evaluation metrics. The developed models yielded an R2 score ranging from 0.61 to 0.74, with the ANN model outperforming the other algorithms resulting in the highest R2 score and lowest error. The models were evaluated on unseen data from other wells drilled in the same formation, and a good match of permeability was obtained.
Permeability is one of the important petrophysical parameters in reservoir engineering (Zhao et al., 2022). an accurate determination is vital for hydrocarbons recovery, carbon storage, geothermal energy capacity, and well placement optimization (Byrnes, 1994). It has a strong relationship with reservoir quality evaluation and the fluid production capacity (Baas et al., 2007), which is based on reservoir characterization, the flow unit identification, and the location of the perforation intervals (Doyen, 1988; Pittman, 1992). It represents the ability of the rock to transmit fluid (Ghafoori et al., 2008). The accurate permeability detection affects the economic value of hydrocarbons accumulation and petroleum reservoir management before and during production.
