Deterministic tools to predict gas assisted gravity drainage recovery factor

Abstract

Naturally fractured rocks contain most of the world's petroleum reserves. This significant amount of oil can be recovered efficiently by gas assisted gravity drainage (GAGD). Although, GAGD is known as one of the most effective recovery methods in reservoir engineering, the lack of available simulation and mathematical models is considerable in these kinds of reservoirs. The main goal of this study is to provide efficient and accurate methods for predicting the GAGD recovery factor using data driven techniques. The proposed models are developed to relate GAGD recovery factor to the various parameters including model height, matrix porosity and permeability, fracture porosity and permeability, dip angle, viscosity and density of wet and non-wet phases, injection rate, and production time. In this investigation, by considering the effective parameters on GAGD recovery factor, three different efficient, smart, and fast models including artificial neural network (ANN), least square support vector machine (LSSVM), and multi-gene genetic programming (MGGP) are developed and compared in both fractured and homogenous porous media. Buckingham π theorem is also used to generate dimensionless numbers to reduce the number of input and output parameters. The efficiency of the proposed models is examined through statistical analysis of R-squared, RMSE, MSE, ARE, and AARE. Moreover, the performance of the generated MGGP correlation is compared to the traditional models. Results demonstrate that the ANN model predicts the GAGD recovery factor more accurately than the LSSVM and MGGP models. The maximum R2 of 0.9677 and minimum RMSE of 0.0520 values are obtained by the ANN model. Although the MGGP model has the lowest performance among the other used models (the R2 of 0.896 and the RMSE of 0.0846), the proposed MGGP correlation can predict the GAGD recovery factor in fractured and homogenous reservoirs with high accuracy and reliability compared to the traditional models. Results reveal that the employed models can easily predict GAGD recovery factor without requiring complicate governing equations or running complex and time-consuming simulation models. The approach of this research work improves our understanding about the most significant parameters on GAGD recovery and helps to optimize the stages of the process, and make appropriate economic decisions.