Estimation of fluid saturation and pressure distribution throughout a reservoir using machine learning techniques

Water saturation is one of the most critical yet often underappreciated petrophysical parameters in reservoir characterization. A wide range of petrophysical and reservoir engineering computations that lead to crucial field development decisions, including reserve estimation, waterflooding efficiency calculation, and capillary pressure deduction, rely on its accurate determination. This study demonstrates how machine learning techniques can forecast reservoirs’ fluid saturation and pressure distribution. This study describes a deep learning–based proxy modeling technique for accurately predicting reservoir pressure distribution and fluid (oil, water, and gas) saturation during water flooding in single-layer heterogeneous reservoirs. This study used recurrent neural networks (RNNs) and convolutional neural networks (CNNs) to build a proxy model. This work indicates that compared to simulation outcomes, computer-based machine learning algorithms can accurately predict fluid (oil, water, and gas) saturation and pressure distribution. The stated accuracy was evaluated numerically and graphically, and error analysis between various machine learning approaches and simulated results was utilized.