Advanced ML and AI Approaches for Proxy-based Optimization of CO2-Enhanced Oil Recovery in Heterogeneous Clastic Reservoirs

This is a Preprint and has not been peer reviewed.

Downloads

Download Preprint

Authors

Watheq J. Al-Mudhafar , Dandina N Rao, Sanjay Srinivasan

Abstract

The EOS-compositional reservoir simulation, Design of Experiments, and Proxy Modeling were integrated to obtain the optimal future performance scenario and to construct the most accurate simplified model alternative to the complex reservoir flow simulation. This integrated workflow was adopted on a sector of the main pay/upper sandstone member in the South Rumaila oil field, located in Iraq. After conducting the acceptable history matching, 6 operational decision parameters, which constrain the production and injection activities, were optimized for their optimal level to achieve optimal flow response factor. Given these decision parameters, the Latin Hypercube Sampling was employed as a low-discrepancy and uniform approach to create hundreds of simulation runs (experiments) to construct a proxy-based optimization approach. The optimal cumulative oil production, by the end of the prediction period, led to obtaining 4.6039 MMMSTB of oil production, while the base case of the GAGD process evaluation of default parameters’ setting resulted to obtain 4.3887 MMMSTB of oil production.
Finally, four proxy metamodels were constructed to provide simplified models alternative to the complex compositional reservoir simulation: Second-Degree Polynomial Equation (QM), Multivariate Additive Regression Splines (MARS), Fuzzy Logic-Genetic Algorithm (FUzzy-GEnetic), and Generalized Boosted Modeling (GBM). The cross-validation with Adjusted $R^{2}_{adj}$ and Root Mean Square Error were employed to find the optimal proxy model that has the least mismatch between the proxy- and simulator-based cumulative oil production response through CO2-GAGD process. It was concluded that both GBM and FUzzy-GEnetic are the most accurate simplified alternative metamodels for the GAGD Process evaluation and prediction.

DOI

https://doi.org/10.31223/osf.io/wsu6g

Subjects

Artificial Intelligence and Robotics, Computer Sciences, Design of Experiments and Sample Surveys, Earth Sciences, Multivariate Analysis, Other Earth Sciences, Physical Sciences and Mathematics, Programming Languages and Compilers, Statistics and Probability

Keywords

Clastic Oil Reservoirs, Cross-validation, Design of Experiments, Enhanced Oil Recovery, Gravity Drainage, Proxy Metamodeling

Dates

Published: 2019-12-04 16:07

Older Versions
License

GNU Lesser General Public License (LGPL) 2.1

Add a Comment

You must log in to post a comment.


Comments

There are no comments or no comments have been made public for this article.