APPLICATION OF MACHINE LEARNING ON PREDICTING OIL RECOVERY DURING SPONTANEOUS IMBIBITION BY LOW SALINITY WATER

Abstract

This thesis explores how machine learning can be used to predict spontaneous imbibition recovery, the process where oil is naturally displaced by water in porous rock, such as in fractured reservoirs. Traditionally, running these lab experiments can take a lot of time, sometimes even months. Instead of waiting that long, this research gathers data from real laboratory results and applies different machine learning models to predict how much oil can be recovered. The focus is on using input parameters like core size, porosity, salinity, temperature, and more to estimate recovery performance without having to physically run the test every time. Six different models were tested: Artificial Neural Networks, Decision Tree, Gradient Boosting, Random Forest, Support Vector Machine, and Extreme Gradient Boosting. The performance of each model was evaluated based on how accurately it could predict real experimental outcomes. The Gradient Boosting model stood out as the most accurate, especially when trained on a combination of secondary and tertiary imbibition data. While the predictions were promising, the study also discusses the limitations of the current dataset and suggests that including more physical parameters like interfacial tension and contact angle could make future predictions even better. Overall, this work shows that machine learning has real potential to speed up and improve decision-making in enhanced oil recovery research.