The study of water cut in the AB reservoir unit of Zubair formation at South Rumaila oilfield, Southern Iraq using petrophysics, geostatistics, and machine learning techniques

This study investigated the spatiotemporal variation of water cut in the AB reservoir unit of the Zubair Formation at the South Rumaila oilfield in Iraq using petrophysics, geostatistics, and machine learning techniques. The study found that the spatial distribution of petrophysical properties such as porosity, permeability, volume of shale, and unit thickness had little impact on the distribution of water cut. The most important factor was the rates of water injection and oil production. The study also found that the AB unit is homogeneous rather than heterogeneous, and this heterogeneity does not play a crucial role in the evolving water cut across the oilfield. The study of historical water cut data showed that the northern part of the oilfield had a higher water cut than the central and southern areas in 2012. However, as production and injection rates increased, the entire oilfield saw significant increases in water cut. Modeling of water cut using four machine learning algorithms (random forest, cubist, support vector machine, and linear regression) and a multi-layer perceptron deep learning technique showed that the random forest and cubist algorithms were the best in both training and testing stages. The stand-alone models of these algorithms for each well location can be used to quickly and easily predict water cut values throughout the oilfield, providing a way to efficiently manage the AB reservoir unit.