A novel methodology for fast reservoir simulation of single-phase gas reservoirs using machine learning.

Reservoir simulation is needed for forecasting hydrocarbon production, determining pressure and saturation, well planning, and field development, among other things. The primary objective is to estimate reservoir performance over a period of time and use that data to enhance hydrocarbon recovery under existing operating conditions. In commercial reservoir simulators, a large number of grid blocks are employed to capture the comprehensive information about a reservoir model, such as porosity and permeability, when the reservoir becomes heterogeneous and complicated. This large number of grid blocks is associated with a large number of mass balance equations that need to be solved simultaneously thereby increasing the amount of computational time it takes to solve them. During reservoir simulation, while moving from one-time level to the next requires a large number of iterations if the properties of reservoir fluids are pressure-sensitive. These further increases the computational cost needed during simulation. The primary objective of this paper is to present a novel approach for reservoir simulation that uses Random Forest (RF) which is one of the widely used Machine learning (ML) algorithm to reduce the number of iterations at each time step and speed up the process. This study investigated the benefits of employing the novel approach created using RF with an application to a conventional single-phase gas reservoir. The study's novelty is in developing a new ML-based reservoir simulator that will make reservoir simulation much faster and computationally more efficient. The standard physics-based system of equations has been included while the traditional reservoir simulation algorithm is modified.