Applied random forest for parameter sensitivity of low salinity water Injection (LSWI) implementation on carbonate reservoir

This study applied a Machine Learning Algorithm based on Random Forest Regression for eliminating the insignificant parameter and evaluating the correlation between each parameter and response parameter on the LSWI process. 1000 experimental designs of LSWI parameters, Reservoir & Injection Temperature, Volume Injection, Formation Water Composition, and Injection Water Composition were build using Design of Experiment on CMOST from Computer Modeling Group with Recovery Factor as the response parameter. Finally, the sensitivity analysis is carried out on Random Forest Regressor based on the decrease in the mean squared error (MSE). The Random Forest Algorithm methods respectively recognized Injection SO42- Composition, Formation Water SO(4)(2-)Composition dan Volume Injection as the top three of most significant parameters. Five variations of the random state value are applied and the hyperparameters of Random Forest also optimized. Both training and test data, the R-2 score respectively are consistently over 0.9 for 5 variations of the random state used. The information about the significant operation parameter of the LSWI process presented in this article is potential bearing the novel to the industry. The insight into those parameters is predicted to be useful to encourage the LSWI implementation on Carbonate Reservoir. (C) 2021 THE AUTHORS. Published by Elsevier BV on behalf of Faculty of Engineering, Alexandria University.

Automated summary

This study applied a Machine Learning Algorithm based on Random Forest Regression to assess the parameters in the LSWI process and identified the most important parameters as Injection SO42- Composition, Formation Water SO(4)(2-)Composition, and Volume Injection.