Prediction of the severe slugging period in gas-liquid two-phase pipeline-riser systems using an artificial neural network

Severe slugging flow (SS) is a very harmful flow regime found during oil and gas transportation. An accurate prediction of the SS period would help to obtain the needed conditions for regime transition models, which are essential when designing and operating an oil and gas field. However, previous empirical correlations are limited to specific flow and structural conditions, resulting in significant errors whenever these parameters deviate. Gas-liquid two-phase flow experiments in a pipeline-riser system were carried out and used as additional training data. The analysis of scatter matrix and correlation coefficient was used to determine the most relevant features impacting SS period. Then, a prediction model based on an artificial neural network (ANN) was established with one single output (the SS period) and seven input features, including five pipeline structural parameters and two flow parameters. The prediction errors coming with previous empirical models are much larger than those of our ANN model. While the relative errors of the empirical models range from −95 % to 3000 %, the ANN has a prediction error within a range of 25 % and a mean absolute error of only 11.3 %, showing a significant improvement over previous predictions.