Hybrid Machine Learning for Modeling the Relative Permeability Changes in Carbonate Reservoirs under Engineered Water Injection

Abstract

Advanced production methods utilize complex fluid iteration mechanisms to provide benefits in their implementation. However, modeling these effects with efficiency or accuracy is always a challenge. Machine Learning (ML) applications, which are fundamentally data-driven, can play a crucial role in this context. Therefore, in this study, we applied a Hybrid Machine Learning (HML) solution to predict petrophysical behaviors during Engineered Water Injection (EWI). This hybrid approach utilizes K-Means and Artificial Neural Network algorithms to predict petrophysical behaviors during EWI. In addition, we applied an optimization process to maximize the Net Present Value (NPV) of a case study, and the results demonstrate that the HML approach outperforms conventional methods by increasing oil production (7.3%) while decreasing the amount of water injected and produced (by 28% and 40%, respectively). Even when the injection price is higher, this method remains profitable. Therefore, our study highlights the potential benefits of utilizing HML solutions for predicting petrophysical behaviors during EWI. This approach can significantly improve the accuracy and efficiency of modeling advanced production methods, which may help the profitability of new and mature oil fields.