Soft Computing Approach for the Prediction of Surfactant Adsorption

Abstract

Abstract Surfactants are commonly used in chemical enhanced oil recovery (cEOR). The quantity of surfactant loss due to adsorption on a rock directly influences a cEOR project economics. Therefore, surfactant adsorption quantification is an important area of interest. Surfactant adsorption is greatly influenced by the mineral composition present in the rock. This paper presents a novel machine learning (ML) intelligent model to predict surfactant adsorption as a function of mineral composition, maximum adsorption capacity, and surfactant concentration. Several pure minerals were used to determine the static adsorption of a novel cationic Gemini surfactant. The novel surfactant is compatible with high salinity and high-temperature environment. XRD was utilized to show the percentage of the rock-forming minerals. The solid-liquid ratio used in this study was 1 gm in 15 ml, and the time given for rock fluid interaction was 24 hours. The supernatants obtained after 24 hours of shaking and 20 minutes of centrifuging were analyzed using high-performance liquid chromatography to determine the remaining surfactant concentration. ML algorithms were applied to the dataset to predict surfactant adsorption. Hyperparameters tuning was performed using K-fold cross-validation integrated with an exhaustive grid search technique. Surfactant adsorption isotherms were constructed from the real experimental data for each pure mineral. The dataset was divided into an 80:20 ratio for training and testing, respectively. Extreme Gradient Boosting (XGBoost) and random forest (RF) techniques were applied to the training dataset to predict the surfactant adsorption as a function of mineral composition, maximum adsorption capacity, and surfactant concentration. The remaining 20% of the dataset was used to test the models. The evaluation error metrics comprising R2 and RMSE showed good agreement of predictions with the unseen data. Also, it was found that XGBoost outperformed RF in surfactant adsorption predictions with R2 of 0.9914 and 0.8990, respectively. The developed model can be used to predict surfactant adsorption by using mineral composition and surfactant concentration. The developed model saves a significant amount of time in running the tedious and time-consuming experiments and helps to provide a good quick estimate of surfactant adsorption. This model will add a great value in the practical application of a chemical EOR project.