(66b) Prediction of Surfactant Retention Using Intelligent Algorithms | AIChE

(66b) Prediction of Surfactant Retention Using Intelligent Algorithms


Abu-Khamsin, S., King Fahd University of Petroleum & Minerals, 31261 Dhahran, Saudi Arabia
Patil, S., KFUPM
Kamal, M. S., King Fahd University of Petroleum & Minerals
Al Shalabi, E. W., Khalifa University
Hussain, S., King Fahd University of Petroleum & Minerals
Surfactants play an essential role in chemical enhanced oil recovery (cEOR). The quantity of surfactant loss caused by retention in reservoir rock during flooding directly impacts a cEOR project's economics. Adsorption of surfactants causes mass loss of surfactants and reduction of surfactant concentration during the flooding process. Therefore, the degree of interfacial tension reduction between oil and the flood water decreases, which unlocks a smaller amount of oil than expected resulting in poor recovery efficiency. Surfactant retention is influenced by numerous parameters, which makes its prediction difficult through analytical modeling. In this paper, artificial neural networks (ANN) and adaptive neuro-fuzzy inference system (ANFIS) algorithms were applied to develop a new empirical correlation to predict surfactant retention based on real core-flood data. Real experimental data on dynamic surfactant retention available in the literature was used in the development of the new empirical correlation. The data was divided into a 70:30 ratio for training and testing, respectively. Feedforward neural networks and subtractive clustering were applied to correlate surfactant retention with several dependent parameters. The coefficient of determination (R2), root mean squared error (RMSE), and average absolute percentage error (AAPE) were used in the error metrics to find the optimal results. The trained model showed good agreement with the unseen data.

Results show that surfactant retention can be correlated with several inputs, such as the surfactant's molecular weight, the solution pH, co-solvent concentration, the total acid number of the oil, temperature, the salinity of the polymer drive, and mobility ratio. Surfactant retention ranges between 0.01 and 0.37 mg/g-rock in the digitalized model. Graphical analysis using scatter plots illustrates that the ANN model produces the most accurate predictions for surfactant retention with R2 in excess of 0.95. A sensitivity analysis of ANN and ANFIS parameters is also provided. This research reports a new correlation to predict surfactant retention using ANN. The data set comprises a large amount of dynamic surfactant retention experiments taken from the literature. The newly developed ANN model gives a quick estimate of surfactant retention and saves a lot of time running the dynamic surfactant retention experiments.