An Explicit Data-Driven Model for Estimation of Free and Absorbed Gas Volumes in Shale Gas Reservoirs

Abstract

Shale gas reservoir fluid flow is characterized by complex interactions between a number of variables, including stress sensitivity, matrix shrinkage, and critical desorption pressure. The behavior and productivity of shale gas reservoirs are significantly influenced by these variables. Therefore, in this study, an in-house finite element poro-elastic simulator was integrated with an in-house Fortran-coded artificial neural network model for accurate estimation of free and absorbed gas volumes in shale gas reservoirs, taking into account the combined effects of stress sensitivity, matrix shrinkage, and critical desorption pressure. The in-house finite element simulator was used to create a database of free and absorbed gas volumes under different reservoir, fluid, and geo-mechanical characteristics of shale gas, while the in-house neural network code was utilized using the outcomes of the poro-elastic simulator in order to propose an explicit data-driven mathematical formula of free and absorbed volumes of shale gas. The study findings indicate that the neural network model achieved high accuracy with an R-squared value exceeding 99% during both the training and blind-testing phases. The newly developed ANN-based correlations can be used as a quick and easy-to-use tool to forecast the volume of shale gas in place both free and adsorbed without any prior experience with utilizing or implementing machine learning models.