Prediction of Storage and Loss Modules of Preformed Particle Gels Using Machine Learning

Abstract Gel treatment is a useful technique to reduce water loss and block highly permeable zones along the reservoir. Thus, the rheological properties are crucial for the material to withstand different reservoir and fluid properties. By understanding the viscoelastic properties of preformed particle gels (PPGs) (storage and loss modules), the formulations can be tailored to specific reservoir conditions, optimize the injection strategies, and ensure long-term efficiency in water conformance control and enhanced oil recovery operations. A machine learning approach was developed to predict the storage modulus (G’) and loss modulus (G") of PPGs based on key parameters such as PPG type, frequency, temperature, salinity, particle size, and shear strain. Experimental data were collected from published studies on various PPG formulations. The Decision Tree (DT), Support Vector Regression (SVR), Gradient Boosting Regressor (GBR), and XGBoost models were employed. K-fold cross-validation and grid search optimization were applied using the mean squared error and coefficient of determination as performance metrics. The trained models were validated against independent datasets. The comprehensive error analysis, including R-squared values, mean squared errors, and scatter plots, enabled an objective comparison of the predictive models. XGBoost demonstrated superior performance among the individual models, achieving the highest R-squared (99.33% and 98.53%) and lowest mean squared error (0.05 and 0.09). Temperature and salinity had the most substantial impacts, with higher temperatures generally decreasing G’ while increasing G". The hyperparameter tuning further refined the model performance, and validation against experimental datasets confirmed the robustness of the predictions. The trend analysis reinforced the reliability of the XGBoost model in capturing the key factors affecting the PPG rheology under reservoir conditions. This predictive capability allows for more precise formulation adjustments, optimizing gel performance for specific field applications. The findings provide a novel tool for designing PPGs tailored to diverse reservoir conditions to predict viscoelastic behavior of PPGs, thereby improving water conformance control strategies.