Investigation on flow characteristics of highly viscous oil-water core-annular flow in horizontal pipes based on machine learning

Abstract

Water-annulus technology is regarded as a promising and efficient method for reducing drag and saving power in the transportation of highly viscous oil. Most existing theoretical prediction models for high-viscosity oil-water core-annular flow in horizontal pipes are based on a concentric core-annular flow configuration with a circular oil core and a smooth oil-water interface. In reality, however, this ideal configuration may not be achieved due to the buoyant force resulting from the density difference between the oil and water, and the instability of the interface. This discrepancy leads to a significant deviation between predicted results and experimental values for flow characteristics, particularly the pressure gradient and water holdup. Therefore, this study proposes a back propagation neural network model enhanced by particle swarm optimization to predict these flow characteristics for highly viscous oil-water core-annular flow in horizontal pipes. The model is trained and tested using experimental data obtained in experimental studies reported in the literature. The results indicate that the new model achieves high prediction accuracy for both pressure gradients and water holdups, significantly surpassing the accuracy of existing phenomenological models for core-annular flow. This proposed modeling approach has the potential to be a powerful tool for design and flow optimization in the petroleum industry.