Research Info

Accurately estimating asphaltene stability in oil and gas operations is critical for ensuring uninterrupted production and preventing flow assurance challenges caused by asphaltene deposition and precipitation. The accumulation of asphaltene can lead to severe operational issues, including reduced well productivity, increased maintenance costs, and potential pipeline blockages. Conventional laboratory methods for evaluating asphaltene stability, while effective, are often time-intensive and expensive, necessitating the development of alternative predictive approaches. In this study, a reliable and efficient machine learning model was developed to predict the asphaltene weight percentage (wt.%) in crude oil mixtures using SARA (saturates, aromatics, resins, and asphaltenes) data. The initial dataset comprised 1948 samples, from which 21.34% of the data were removed through outlier detection using the Gaussian method, resulting in a final dataset of 1532 samples for model development. To achieve this, 9 ML algorithms were assessed and compared, including artificial neural network (ANN), decision tree (DT), Gaussian process regression (GPR), K-nearest neighbor (KNN), linear regression (LR), multiple linear regression (MLR), random forest (RF), support vector machine (SVM), and support vector regression (SVR). The performance of these models was evaluated based on their predictive accuracy and error metrics. Among the examined algorithms, KNNs demonstrated the highest predictive capability, achieving an R2 value of 0.9649 on the test dataset and 0.9636 on the training dataset. Furthermore, the model exhibited a root mean square error (RMSE) of 2.481 for the test data and 2.329 for the training data, indicating its robustness and reliability. The superior performance of KNN suggests that ML can serve as a powerful and cost-effective alternative to conventional laboratory techniques for asphaltene stability assessment.