Machine learning classification of flow regimes in a long pipeline-riser system with differential pressure signal

Abstract

Accurate classification of flow regimes in pipeline-riser system is a promising approach for flow assurance in offshore oil fields. Experiments on two-phase flow regimes in a 1657 m long-distance pipeline-riser system were conducted. Statistical features of the differential pressure signals were extracted and input into a neural network classifier to identify flow regimes. The influences of positions along the loop and sample lengths of the differential pressure signals on the recognition rates were discussed. For the 13 differential pressure signals at various positions along the loop, the recognition rates of the signals on the S-shaped riser section, the inclined section and the horizontal section decreases in turn. Recognition rates higher than 92.2% are achieved with signals near the outlet of the riser. The overall recognition rate increases significantly as the sample length increases, and good recognition rates can be obtained if the sample lengths exceed about 74.4 s.