Identification of flow regimes using platform signals in a long pipeline with an S-shaped riser

•Gas-liquid flow regimes are studied in a 1657 m long pipeline with an S-shaped riser.•Regime boundaries and transition velocities are presented in a flow regime map.•Parameters from wavelet analysis are input into a decision tree to identify regimes.•Minimum sample duration obviously drops with the increase of signal number.•Principal component analysis significantly reduces statistical parameters’ dimension. In this paper, we experimentally investigate gas–liquid flows in a 1657 m long pipeline with an S-shaped riser. Boundaries of three different flow regimes and the transition velocities are determined and presented on the flow regime diagram. Differential pressure signals and pressure signal near the platform are firstly decomposed into six scales via a wavelet analysis, and then statistical parameters on each scale are extracted and further input into a decision tree classifier. For a sample duration of 18.6 s, the highest recognition rates of one signal, two-signal, and three-signal rise from 93.2%, 95.0%, to 96.8%. For one signal, two-signal, and three-signal, the shortest sample durations required for a satisfactory recognition rate of 95% obviously become shorter from 44 s, 28 s, to 2.3 s. On the premise of the highest recognition rate 94.5% to 95.0%, one achieves a dramatic dimension reduction of 81%-93% for statistical parameters via a principal component analysis.