A machine-learning approach for optimal ionic concentration determination in smart-water EOR applications

The smart water-enhanced oil recovery (EOR) process is a pioneering tertiary recovery method in the petroleum industry. Meanwhile, more than half of oil reserves in the world are carbonate. Accordingly, considering the technical and financial aspects, the determination of the accurate concentration of presented ions in smart water is very important. Although several experimental studies considered this issue, no appropriate statistical method has been suggested to deal with this problem during smart water injection in carbonate rocks. In the present article, five different multi-target regression machine learning (ML) algorithms (i.e., Random Forest, Decision Tree, K-Nearest Neighbours, Lasso and Linear), were used to predict the ionic concentration in imbibition tests. A completely reliable dataset of imbibition test results, which were gathered from the literature, was employed in the algorithm learning process to examine their accuracy. After data processing, feature extraction, splitting data and building candidate ML models, an exact hyperparameter tuning was carried out to evaluate the ML models and select the best model. It was found that the Random Forest algorithm is the best-acting approach, with the lowest total root mean squared error (RMSE) of ۱.۲۳۱ and the highest score of ۰.۹۸۱ for predicting ionic concentration in smart water EOR applications. In conclusion, the proposed model is the most efficient approach as compared with commonly used costly laboratory tests, which can be a good candidate for predicting the concentration of ions in smart water injection processes.

کلیدواژه ها:

Wettability Alteration ، Enhanced Oil Recovery ، machine learning ، Carbonate rocks ، Smart water