DEVELOPMENT OF A RANDOM-FOREST-BASED MODEL FOR PREDICTING SLUG FLOW CHARACTERISTICS USING MACHINE LEARNING TECHNIQUE

Abstract

The development of heavy oil has attracted attention in recent times. With increasing fluid viscosity, slug flow has become the most common flow pattern in oil and gas pipeline flow which poses a challenge in flow assurance as the need for stability of system and production maximization. The accurate prediction of slug flow parameters is an urgent problem to be solved in heavy oil development for more efficiency in productivity. In this research paper, the analysis of experimental data for the Air-Silicon oil slug transition in a 67 mm diameter and 6 m long vertical pipe was carried out in this work. The superficial velocity ranges of gas and liquid obtained from the ECT were 0.047 – 4.727 m/s and 0.05 – 0.284 m/s respectively. This research makes use of Random-forest-based Machine learning technique to predict liquid hold up and slog flow regime characteristics at different time intervals due to as it uses random subspace method and bagging to prevent overfitting. From the investigated data, the liquid hold up, void fraction were obtained and other slug flow parameters obtained were; structural velocity, slug frequency, length of slug and film thickness. Comparison with the data from the proposed algorithm accurately predicts the liquid hold up, void fraction, and liquid film thickness. They were seen to have a good agreement with a Mean Square Error of 0.2 % with the Machine Learning based Random forest prediction however slug frequency, structural velocity, and length of slug unit all had varying disagreement with the prediction leading to limitations in the use of the model algorithm in prediction of these flow parameters. The model was also tested against varying viscosity and a good agreement of over 99 % was seen from 5 cP to 1000 cP excluding high liquid viscosity of 5000 cP. The random-forest based machine learning model can then be used in predicting liquid hold up, void fraction, and liquid film thickness in low viscosity fluids less 1000 cP. The model developed aids the flow assurance and design involving multiphase flow slug flow in Oil and gas operations