The use of machine learning models to estimate the viscosity of Brazilian water-in-crude oil emulsions

Abstract

The Brazilian oil and gas sector has observed rapid growth over the last decade, principally due to substantial pre-salt reserves and advancements in deepwater technologies. This expansion has, however, introduced complex challenges related to flow assurance in offshore production systems, particularly with high-viscosity water-in-oil (W/O) emulsions that significantly impact flow efficiency. In this context, the current study aims to address these challenges by applying machine learning (ML) models to predict the viscosity behavior of emulsions derived from 23 Brazilian crude oils, utilizing a large dataset, encompassing 6,930 datapoints and evaluating different model configurations as random forest (RF), gradient boosting (GB), and artificial neural networks (ANN). The performance of the models was evaluated through R² score, mean squared error (MSE) and average relative deviation (ARD) metrics, demonstrating that RF was the most balanced model, achieving an R² of 0.9962 and the lowest ARD of 10.67%, indicating its high accuracy and precision. While the GB model achieved the highest R² of 0.9980, its higher ARD (36.3%) indicated limitations in point-specific accuracy. Similarly, the ANN model demonstrated robust general trend prediction (R² of 0.9971) but struggled with localized precision (ARD of 36.67%). Furthermore, comparisons were performed with well-stablished viscosity correlations, including those proposed by Rønningsen and Farah et al.. These comparisons demonstrated the superior performance of the RF model, evidenced by lower ARD and higher R² values. Moreover, the RF model also outperformed other machine learning models reported in the literature, further validating its robustness. The obtained results highlighted the improved capacity of the developed models in predicting viscosity behaviors of W/O emulsions and the reliability of ML techniques for accurately predicting emulsion viscosity, providing significant support for engineering applications where precise flow assurance is critical.