AI-driven hybrid modeling and GA-based optimization of polymer solution viscosity for enhanced oil recovery applications

Abstract

Predicting the zero-shear viscosity ( η0) of Partially Hydrolyzed Polyacrylamide (HPAM)based polymer solutions is a key challenge in enhanced oil recovery due to the complex, nonlinear interactions among formulation parameters. A previously validated polynomial model was used to generate a synthetic dataset. Six artificial intelligence models were trained, including five machine learning algorithms (Decision Tree Regression (DTR), Random Forest Regression (RFR), K-Nearest Neighbors (KNN), Support Vector Regression (SVR), Artificial Neural Network (ANN)) and one deep learning model Convolutional Neural Network (CNN). Optimization using genetic algorithm significantly improved the performance of the models. Hyperparameter optimization notably enhanced the predictions, with Support Vector Regression improving from R2 of 0.6462 to 0.9976, and KNN improving from 0.9940 to 0.9999. The best performance was achieved by the DTR, reaching a perfect R2 of 0.9998, and KNN maintaining a high performance with R2 = 0.9999. Moreover, the optimization of the operating conditions, including particle size, concentration, and temperature, led to similar optimal conditions across all models, with a particle size of 1000 nm, a concentration of 50 ppm, and a temperature of 20 ∘C. These conditions resulted in a predicted maximum viscosity ranging from 0.556 Pa ⋅s for the ANN model to 0.5612 Pa ⋅s for the DTR model, very close to the experimentally measured value of 0.56 Pa ⋅s. This study demonstrates the potential of AI-driven optimization for guiding intelligent formulation design in polymer applications.