Oil-Gas-Water Three-Phase Flow Pattern Identification Through Parallel Decision Trees With Differential Pressure and Ultrasonic Sensors

• Published in: IEEE transactions on instrumentation and measurement Vol. 73; pp. 1 - 15
• Main Authors: Shi, Xuewei, Tan, Chao, Dong, Feng
• Format: Journal Article
• Language: English
• Published: New York IEEE 2024; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Acoustics; Cost analysis; Decision tree; Decision trees; Differential pressure; differential pressure (DP) sensor; Feature extraction; Flow characteristics; Flow distribution; flow pattern identification; Flow velocity; Fluids; Frequency measurement; Intelligibility; oil–gas–water three-phase flow; Performance evaluation; Pressure measurement; pulse wave ultrasonic Doppler (PWUD) sensor; Sensor phenomena and characterization; Sensors; Ultrasonic imaging; Ultrasonic variables measurement; Velocity distribution
• ISSN: 0018-9456; 1557-9662
• DOI: 10.1109/TIM.2024.3470252

A deep understanding of oil-gas-water three-phase flow behaviors and mechanics has great significance for flow controlling and modeling, production efficiency improvement, and operation safety assurance. To accurately identify horizontal oil-gas-water three-phase flow patterns, a decision-level fusion method with a parallel-tree structure is proposed based on a combined differential pressure (DP) and pulse wave ultrasonic Doppler (PWUD) sensor. With advantages of nonintrusive, easy-to-mount operation, and low cost, the combined sensor can simultaneously acquire various flowing information, including instantaneous DP fluctuations, time-varying echograms, and velocity profiles. By analyzing the dynamic sensor responses to different flow behaviors, typical flow characteristics of oil-gas-water three-phase flow are first revealed. On this basis, several features are extracted from the time-accumulated echograms, average flow velocity time series, and decomposed DP fluctuations to objectively characterize the flow patterns from different perspectives in the time, space, and time-frequency domains. Then, two parallel decision trees are constructed to, respectively, identify the gas-liquid interphase structure and oil-water interphase structure. Using the visibility of tree structures, the ability of these extracted features for flow structure distinguishment is analyzed. Finally, the final decision on the oil-gas-water three-phase flow pattern was generated by fusion the outputs of two decision trees. Tenfold cross-validation method is adopted for model performance evaluation. The results show that 12 types of oil-gas-water three-phase flow pattern are identified with the overall identification accuracy of 92.1%. This study provides a simple, affordable, and nonintrusive solution with good intelligibility and interpretability for flow pattern identification of complex industrial three-phase flows.