Predictive modeling of oil and water saturation during secondary recovery with supervised learning

In the petroleum reservoir, the secondary oil recovery (SOR) process is employed by injecting water into wells to enhance the moment of oil toward the production wells. The SOR process gives rise to the instability (fingering) phenomena due to the injecting force and the difference in the wettability and viscosity of the oil and water at the common interface. Since the late 1800s, mathematical models of petroleum reservoirs have been extensively used in the oil and gas industry. In this paper, we investigated the saturation of two immiscible fluid (oil and water) flows through homogeneous porous media during the SOR process by solving the modeled partial differential equation using the supervised machine learning algorithm based on feedforward back-propagated neural networks (FFBNNs) and Levenberg-Marquardt (LM) optimization algorithm. The designed scientific computing technique (FFBNN-LMA) is further employed to study the detailed sensitivity analysis of the approximate solutions. Performance measures like average absolute deviations, Theils' inequality measure, regression, and Nash-Sutcliffe model efficiency coefficient.