Characteristics and flow measurement of gas-liquid two-phase counter-current flow in vertical annulus
- 1Gansu Key Laboratory of Efficient Utilization of Oil and Gas Resources in Longdong, College of Petroleum and Chemical Engineering, Longdong University
- 2College of New Energy, Longdong University
- 3PetroChina Changqing Oilfield Co
Abstract
Gas–liquid two-phase flow measurement in vertical annular pipes remains a challenge in petroleum engineering and related fields, due to the complex flow regimes and asymmetric cross-sectional geometry. In this study, a non-invasive soft-sensing method based on multi-scale differential pressure signals and machine learning is developed for flow rate prediction in vertical annular (inner/outer diameters 75/125 mm) counter-current gas–liquid two-phase flow. Experiments were conducted over a gas superficial velocity range of 0.06–2.91 m/s, a liquid superficial velocity range of 0.02–0.24 m/s, and five eccentricity levels (0–1.0). Differential pressure signals were simultaneously acquired at two spatial scales: a short-span (50 mm, approximately 1 D) and a long-span (1000 mm, approximately 20 D). A total of 53 features were extracted from the time domain, frequency domain (power spectral density), and time-frequency domain (discrete wavelet transform). Three predictive models – Random Forest (RF), XGBoost, and Multilayer Perceptron (MLP) – were constructed and hyperparameter-optimized via grid search. Results show that the XGBoost model exhibits the best generalization performance, achieving mean absolute percentage errors (MAPE) of 6.04% for gas superficial velocity and 27.17% for liquid superficial velocity. SHAP-based feature importance analysis indicates that the long-span (1000 mm) differential pressure features play a dominant role in flow rate prediction. Low-frequency energy and the signal mean are crucial for liquid velocity prediction, while high-frequency detail energy governs gas velocity prediction. Although the short-span (50 mm) features rank lower in global importance, they provide critical non-linear corrections under extreme conditions such as high gas velocity. The eccentricity feature (ecce) exhibits a data-driven nonlinear tendency: once the eccentricity exceeds 0.5, the SHAP values for both phases shift markedly negative. This trend is qualitatively consistent with the understanding that wall friction in the narrow gap increases under highly eccentric annular geometries. This study demonstrates that multi-scale differential pressure sensing combined with interpretable machine learning offers a feasible pathway for flow rate measurement in complex wellbore annuli without relying on specific flow regimes (e.g., Taylor bubble flow).
Related
- Integrated counter current separation and omics approaches for discovery of bioactive compounds from medicinal plants
- Factors associated with current non-nutritive sweetener use among adults in Chungcheong, Korea: focusing on the sociodemographic and dietary characteristics
- Transcatheter Aortic Valve Implantation in Complex Anatomy: An Anatomy-Driven Framework for Bicuspid Stenosis, Pure Aortic Regurgitation, and the Small Annulus in the Era of Lifetime Management
- Association of current HbA1c with concurrent multiorgan screening abnormalities and the incremental value of routine clinical characteristics in type 2 diabetes: a single-center cross-sectional study in Wenzhou, China.
- Pulmonary Valve Annulus Z-Score as a Predictor of Transannular Patch Insertion in Tetralogy of Fallot Repair: A Cross-sectional Study at National Cardiovascular Center Harapan Kita
BibTeX
@article{cao2026characteristics,
title = {{Characteristics and flow measurement of gas-liquid two-phase counter-current flow in vertical annulus}},
author = {Feng Cao and Anzhao Ji and Zhanjun Chen and Xuefen Liu and Peng Zhang and Fengfeng Yang and Huifeng Zheng},
journal = {Scientific Reports},
year = {2026},
month = sep,
doi = {10.1038/s41598-026-68538-y},
url = {https://doi.org/10.1038/s41598-026-68538-y},
note = {CC BY},
}
RIS
TY - JOUR
TI - Characteristics and flow measurement of gas-liquid two-phase counter-current flow in vertical annulus
AU - Feng Cao
AU - Anzhao Ji
AU - Zhanjun Chen
AU - Xuefen Liu
AU - Peng Zhang
AU - Fengfeng Yang
AU - Huifeng Zheng
PY - 2026
DA - 2026/09/01
JO - Scientific Reports
DO - 10.1038/s41598-026-68538-y
UR - https://doi.org/10.1038/s41598-026-68538-y
AB - Gas–liquid two-phase flow measurement in vertical annular pipes remains a challenge in petroleum engineering and related fields, due to the complex flow regimes and asymmetric cross-sectional geometry. In this study, a non-invasive soft-sensing method based on multi-scale differential pressure signals and machine learning is developed for flow rate prediction in vertical annular (inner/outer diameters 75/125 mm) counter-current gas–liquid two-phase flow. Experiments were conducted over a gas superficial velocity range of 0.06–2.91 m/s, a liquid superficial velocity range of 0.02–0.24 m/s, and five eccentricity levels (0–1.0). Differential pressure signals were simultaneously acquired at two spatial scales: a short-span (50 mm, approximately 1 D) and a long-span (1000 mm, approximately 20 D). A total of 53 features were extracted from the time domain, frequency domain (power spectral density), and time-frequency domain (discrete wavelet transform). Three predictive models – Random Forest (RF), XGBoost, and Multilayer Perceptron (MLP) – were constructed and hyperparameter-optimized via grid search. Results show that the XGBoost model exhibits the best generalization performance, achieving mean absolute percentage errors (MAPE) of 6.04% for gas superficial velocity and 27.17% for liquid superficial velocity. SHAP-based feature importance analysis indicates that the long-span (1000 mm) differential pressure features play a dominant role in flow rate prediction. Low-frequency energy and the signal mean are crucial for liquid velocity prediction, while high-frequency detail energy governs gas velocity prediction. Although the short-span (50 mm) features rank lower in global importance, they provide critical non-linear corrections under extreme conditions such as high gas velocity. The eccentricity feature (ecce) exhibits a data-driven nonlinear tendency: once the eccentricity exceeds 0.5, the SHAP values for both phases shift markedly negative. This trend is qualitatively consistent with the understanding that wall friction in the narrow gap increases under highly eccentric annular geometries. This study demonstrates that multi-scale differential pressure sensing combined with interpretable machine learning offers a feasible pathway for flow rate measurement in complex wellbore annuli without relying on specific flow regimes (e.g., Taylor bubble flow).
SN - 2045-2322
ER -