Classification of Multiphase Flow Regime in a Horizontal Pipe Using Supervised Machine Learning and Bayesian Optimization

Multiphase flow refers to a fluid flow consisting of more than one phase. Conventional/traditional methods rely on empirical correlations and flow maps of the flow regimes that show overlapping characteristics. Machine learning (ML) has gained traction for multiphase flow regime classification, offering improved accuracy and adaptability to complex flow conditions. This paper presents the supervised ML approach to classify the three-phase flow regimes in horizontal pipes. The superficial gas and liquid velocities and water cuts are utilized as input features for flow regime classification. ML classification models, such as Bayesian Classifier and K-Nearest Neighbors, and boosting models, such as XGBoost, CatBoost, and LightGBM, are employed and evaluated. Bayesian optimization is used to enhance model accuracy and robustness by hyperparameter tuning. The selected supervised ML models provided an accurate and efficient classification of multiphase flow regimes when enhanced using Bayesian optimization. The optimized ML models significantly improved the classification accuracy and the F1 score, making them practical for real-time monitoring and control applications.