Modelling over-reading correction factors for ultrasonic flow meters in wet gas measurement using advanced regression and machine learning techniques

Ishigita Lucas Shunashu, Osmund Kaunde,
Modelling over-reading correction factors for ultrasonic flow meters in wet gas measurement using advanced regression and machine learning techniques,
Next Research,
Volume 2, Issue 4,
2025,
100915,
ISSN 3050-4759,
https://doi.org/10.1016/j.nexres.2025.100915.
(https://www.sciencedirect.com/science/article/pii/S3050475925007821)
Abstract: Accurate wet gas measurement is essential for optimizing production, transmission, and reservoir management in oil and gas operations. Ultrasonic flow meters, though non-intrusive and versatile, often overestimate flow rates due to the presence of liquid phases, leading to significant operational and economic errors. To address this, a data-driven correction model was developed using computational techniques to predict and compensate for over-reading. This study evaluates the performance of several advanced regression and machine learning approaches, including polynomial regression, random forest regression, nonlinear curve fitting, neural networks, multiple linear regression, ridge regression, and lasso regression, using an extensive experimental dataset. Key input variables include liquid volume fraction, Lockhart–Martinelli parameter, Froude number, Weber number, slip ratio, and density ratio. Among the models tested, random forest regression and multiple linear regression achieved the highest accuracy, with average relative absolute errors of 3.02% and 3.20 % respectively. These findings demonstrate the potential of data-driven modeling to enhance the reliability of ultrasonic flow meters in complex wet gas environments.
Keywords: Machine learning; Ultrasonic flow meter; Wet gas measurement; Over-reading correction; Advanced regression