Signal selection for identification of multiphase flow patterns in offshore pipeline-riser system

Multiphase flow pattern identification is an urgently desirable approach for flow assurance of gas-liquid trans- portation in deep-sea pipelines. Experiments are performed on a system with 1657 m horizontal pipe and 16.7 m S-shaped riser. A new method of sample preparation using only data from liquid accumulation stage is proposed for identification of early-stage severe slugging. A set of signal evaluation and selection criteria covering the recognition effect and practicability of signals are proposed for identification of severe slugging, oscillating flow and stable flow. For oilfield applications including above-water and underwater scenarios, pressure signals at different locations of the pipeline-riser system are selected to meet the actual demand with the shortest sampling duration. Using the combination of above-water signals at the top of the riser, the recognition rate of 90.0% is achieved at a sampling duration of 37.2 s. Feature dimension is reduced by principal component analysis, and good recognition results are obtained using the optimal signal combinations in different classifiers.