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

Constructing a simpler model to represent a complex reservoir simulation that will be employed to define the optimum future development plans have been achieved through the use of different simulation techniques that include EOS-compositional reservoir simulation, Proxy Modeling as well as Design of Experiments. Once reliable history matching was achieved, the key five operational decision parameters were manipulated to find the optimum value to attain maximum objective function. A low-discrepancy and consistent procedure was used to generate several hundred simulation jobs or experiments to build a proxy metamodeling optimization by adopting the Latin Hypercube Sampling with the five decision parameters. At the end of the forecast case, the optimum cumulative produced oil resulted in achieving 4.6 MMMSTB of oil production compared with 4.4 MMMSTB of oil production that was produced from the base scenario of the GAGD technique assessment of original decision parameters' conditions.
Lastly, four machine learning (ML) and Artificial Intelligence (AI) algorithms were considered as proxy meta-models as substitute to the large numerical simulation model: Quadratic Modeling (QM), Fuzzy Logic-Genetic Algorithm (FUzzy-GEnetic), Multivariate Additive Regression Splines (MARS), and Generalized Boosted Modeling (GBM). The cross validation of the Adjusted R2-adj along with the RMSE were the base to conclude the optimum proxy metamodel which provides the lowest variance between the predicted and calculated model considering the produced oil as response by CO2-GAGD technique. Consequently, GBM was determined to be the best shorten alternative metamodel for the performance of GAGD process.