Kuwaiti Carbonate Reservoir Oil Recovery Prediction Through Static Wettability Contact Angle Using Machine Learning Modeling

Abstract

Abstract The objective of this study is to predict EOR efficiencies through static wettability contact angle measurement by Machine Learning (ML) modeling. Unlike conventional methods of measuring static wettability contact angle, the unconventional digital static wettability contact angle is captured and measured, then (ML) modeled in order to forecast the recovery based on wettability distribution phenomenon. Due to success in big data collection from reservoir imaging samples, this study applies data science lifecycle logic and utilizes Machine Learning (ML) models that can predict the recovery through wettability contact angles and thus identify the treatment of oil recovery for a candidate reservoir. Using developed morphological driven pixel-data and transformed numerical wettability contact angle data are acquired from Scanning Electron Microscope Backscattered Electron (SEM-BSE) for 27 fresh core samples from top to bottom of the reservoir. These samples are properly sequenced and then images are selected. Big data from imaging technology have been processed in a manner to train, and test the model accuracy. Applied Data Science Lifecycle technique, such as data mining, is utilized. Data Exploration Analysis (DEA) is implemented to understand and review data distribution as well as relationships among input features. Different supervised ML models to predict recovery are utilized and an optimal model is identified with an acceptable accuracy. The selected prediction model is applied to model the optimal recovery practice. Extreme Gradient Boosting (XGBoost) algorithm is utilized and found as a best-fit model for this Kuwaiti reservoir case practice. Moreover, decision tree and Artificial Neural Network (ANN) models could provide acceptable accuracy. Other supervised learning models were attempted and were not promising to provide feasible accuracy for this carbonate reservoir. The novel of this unique solution of the data-driven ML model is to predict recovery based on static wettability contact angles (θ°). The static wettability contact angles (θ°) and pore morphological features introduce an insights method to support reservoir engineers in making value-added decisions on production mechanisms and hydrocarbon recovery for their reservoirs. Hence, it improves the field development strategy.