Estimating oil recovery efficiency of carbonated water injection with supervised machine learning paradigms and implications for uncertainty analysis

Abstract

Limited efforts have been made to develop a time-efficient and cost-effective predictive model capable of estimating the oil recovery efficiency of carbonated water injection (CWI). Therefore, in this study, we utilized supervised machine learning (ML) techniques: decision tree, support vector regression, and random forest (RF) to predict the recovery efficiency of CWI, with experimental conditions, rock properties, and fluid properties as predictors. The influence of various parameters on oil recovery efficiency was assessed using correlation technique, permutation importance, and Shapley Additive Explanations (SHAP), which sets our study apart from existing studies. Generally, the ML models yielded remarkable recovery efficiency prediction results, achieving coefficients of determination, mean absolute errors, and root mean square errors of 0.81–0.87, 4.30–4.96 %, and 4.82–5.89 %, respectively. The RF model outperformed its counterparts. Most importantly, the RF model successfully predicted the recovery efficiency on entirely new data with an error and absolute relative error of less than 15 % and 19 % respectively According to the SHAP analysis, high injection rate, porosity, permeability, and pressure improve oil recovery, and vice versa. Similarly, low temperature, oil density and viscosity, and salinity enhance oil recovery while injection rate and temperature were the most and least influential parameters, respectively. The RF model was successfully deployed to predict the oil recovery efficiency for 1000 randomly generated sets of independent variables in conjunction with Monte Carlo simulation, demonstrating the applicability of the model in uncertainty analysis. The current modeling study not only bridges the knowledge gaps in predictive modeling of the oil recovery efficiency of CWI but also holds significant promise for rapid estimation and optimization of oil recovery efficiency.