Prediction of Foam Half-Life Time Using Machine Learning Algorithms for Enhanced Oil Recovery and CO₂ Sequestration

Foam is widely utilized in upstream applications, including enhanced oil recovery (EOR), CO₂ sequestration, and well stimulation, with its stability being crucial for optimal performance. Foam stability, often assessed through its half-life, is traditionally determined via experimental methods, which are time-consuming and resource-intensive. Despite the increasing application of machine learning (ML) techniques, no study has specifically predicted foam stability using ML models. This study aims to develop ML-based predictive models for foam half-life using four algorithms: random forest (RF), gradient boosting regression (GBR), extra trees regression (ETR), and artificial neural networks (ANN). A data set of 1,392 experimental measurements, incorporating gas type, additives, surfactants, temperature, pressure, and salinity, was utilized. The models were trained, tested, and validated, with performance assessed using statistical metrics such as R² and AAPE. Additionally, a hybrid RF-ANN model was developed to enhance predictive accuracy, and sensitivity analysis was conducted to identify key influencing factors. Results showed that the GBR model achieved the best standalone performance, with R² values of 0.99, 0.98, and 0.97 for training, testing, and validation. The hybrid RF-ANN model outperformed all individual models, attaining R² values of 0.994, 0.993, and 0.994 for training, testing, and validation, respectively, with minimal AAPE values. Sensitivity analysis indicated that temperature significantly influenced foam stability, with the foam half-life decreasing sharply above 100 °C. This novel ML-based approach offers a fast, cost-effective alternative to traditional experimental methods, enabling the accurate prediction and optimization of foam stability in EOR and CO₂ sequestration applications.