Experimental justification of the choice of models for predicting the efficiency of a gas separator as part of an electric submersible pump system

UDK: 622.276.53.001.57

DOI: 10.24887/0028-2448-2025-12-82-86

Key words: gas separator, electric submersible pump (ESP) system, machine learning, gas separation, TabPFN, experimental research

Authors: E.V. Yudin (Gazprom Neft Companу Group, RF, Saint Petersburg); V.A. Kupavykh (Ufa State Petroleum Technological University, RF, Ufa); B.M. Latypov (Ufa State Petroleum Technological University, RF, Ufa); V.E. Chernyshov (Association «Digital technologies in industry», RF, Saint Petersburg); M.D. Shabunin (Research and Education Center «Gazprom Neft-UGNTU», RF, Ufa); I.V. Grigoriev4,5 (Research and Education Center «Gazprom Neft-UGNTU», RF, Ufa; Gubkin University, RF, Moscow); M.V. Verbitsky (Gubkin University, RF, Moscow)

This article presents an experimental and model-based justification for selecting optimal predictive approaches to estimate the gas separation efficiency of downhole separators used in electrical submersible pump (ESP) systems operating under high free-gas conditions. Gas separation remains a critical challenge in ESP applications, as excessive free gas leads to degradation of pump head–flow performance, cavitation, unstable flow regimes, and reduced run life. To address this issue, a comprehensive experimental investigation was carried out using a patented laboratory flow loop at Gubkin University, enabling controlled generation of gas–liquid mixtures and evaluation of separators of different sizes under varying liquid flow rates, gas rates, and shaft speeds. Based on the collected dataset, the study compares two major classes of predictive models: classical statistical approaches (polynomial regression, Ridge, Lasso, ElasticNet) and a modern machine-learning method, TabPFN - modern machine learning algorithm tailored for small tabular datasets. Model performance was assessed using coefficient of determination R2 and mean absolute percentage error (MAPE) with cross-validation. The results reveal that TabPFN significantly outperforms all classical models in predicting residual gas content. When using all eight input features, TabPFN achieves MAPE equal to 1,45 % and R2 – 0,96. Although classical regression methods show lower accuracy, they provide interpretability and enable identifying key influencing factors, such as inlet gas fraction and separator size. The findings confirm the high potential of advanced machine-learning methods for predictive analytics, enabling development of digital twins and real-time operational optimization tools for ESP systems working in high-gas environments.

References

1. Wilson B.L., ESP Gas separator’s affect on run life, SPE-28526-MS, 1994, DOI: https://doi.org/10.2118/28526-MS

2. Kuz’min N.I., Verbitskiy V.S., Khabibullin R.A. et al., Analysis of oil wells operation parameters and modes effects on electric submersible pumps reliability (In Russ.), Neftyanoe khozyaystvo = Oil Industry, 2024, no. 12, pp. 106–111, DOI: https://doi.org/10.24887/0028-2448-2024-12-106-111

3. Yudin E.V., Gorbacheva V.N., Smirnov N.A., Modeling and optimization of wells operating modes under annular flow conditions (In Russ.), Neftyanoe khozyaystvo = Oil Industry, 2022, no. 11, pp. 122–126, DOI: https://doi.org/10.24887/0028-2448-2022-11-122-126

4. Yudin E. et al., Modeling and optimization of ESP wells operating in intermittent mode, SPE-212116-MS, 2022, DOI: https://doi.org/10.2118/212116-MS

5. Yudin E. et al., New applications of transient multiphase flow models in wells and pipelines for production management, SPE-201884-MS, 2020,

DOI: https://doi.org/10.2118/201884-MS

6. Harun A.F., Prado M.G., Serrano J.C., A mechanistic model to predict natural gas separation efficiency in inclined pumping wells, SPE-67184-MS, 2001,

DOI: https://doi.org/10.2118/67184-MS

7. Lea J.F., Bearden J.L., Gas separator performance for submersible pump operation, Journal of Petroleum Technology, 1982, V. 34, no. 6, pp. 1327–1333,

DOI: https://doi.org/10.2118/9219-PA

8. Gadbrashitov I.F., Sudeyev I.V., Generation of curves of effective gas separation at the ESP intake on the basis of processed real measurements collected in the Priobskoye oil field, SPE-102272-RU, 2006, DOI: https://doi.org/10.2118/102272-RU

9. Gorid’ko, K.A., Kobzar’ O.S., The approach to determine the gas separator efficiency as a part of an electric submersible pump unit (In Russ.), Nauchnye trudy NIPI Neftegaz GNKAR, 2023, no. S1, pp. 9–20, DOI: https://doi.org/10.5510/OGP2023SI100831

10. Mikhaylov V.G., Petrov P.V., Mathematical model of separation of gas in the working chamber of the rotary gas separator (In Russ.), Vestnik UGATU, 2008, V. 10,

no. 1, pp. 21–29.

11. Harun A.F., Prado M.G., Shirazi S.A., An improved model for predicting separation efficiency of a rotary gas separator in ESP systems, SPE-63044-MS, 2000,

DOI: https://doi.org/10.2523/63044-MS

12. Ojeda, L.C.O., Olubode M., Karami H., Application of machine learning to evaluate the performances of various downhole centrifugal separator types in oil and gas production systems, SPE-213059-MS, 2023, DOI: https://doi.org/10.2118/213059-MS

13. Sharma A., Ojeda C.S., Yuan N., Predicting gas separation efficiency of a downhole separator using machine learning, Energies, 2024, V. 17, no. 11,

DOI: https://doi.org/10.3390/en17112655

14. Okafor C.C., Verdin P.G., 3D computational fluid dynamics analysis of natural gas separation efficiency in multiphase pumping wells with heterogeneous flow regime, Engineering Applications of Computational Fluid Mechanics, 2024, V. 18, no. 1, pp. 2395452–2395473, DOI: https://doi.org/10.1080/19942060.2024.2395452

15. Okafor C.C., Verdin P.G., Hart P., CFD investigation of downhole natural gas separation efficiency in the churn flow regime, SPE-204509-MS, 2021,

DOI: https://doi.org/10.2118/204509-MS

16. Liu B., Prado M., Application of a bubble tracking technique for estimating downhole natural separation efficiency, Journal of Canadian Petroleum Technology, 2004, V. 43, no. 5, DOI: https://doi.org/10.2118/04-05-05

17. Drozdov A.N., Stand Investigations of ESP’s and gas separator’s characteristics on gas-liquid mixtures with different values of free-gas volume, intake pressure, foaminess and viscosity of liquid, SPE-134198-MS, 2010, DOI: https://doi.org/10.2118/134198-MS

18. Drozdov A.N., Verbitckiy V.S., Arseniev A.A., Rotary gas separators in high GOR wells, field and lab tests comparison, SPE-117415-MS, 2008,

DOI: https://doi.org/10.2118/117415-MS

19. McCoy J.N., Podio A.L., Lisigurski O., A laboratory study with field data of downhole gas separators, SPE-96619-PA, 2007, DOI: https://doi.org/10.2118/96619-PA

20. Patent RU2075654C1. Method of tests of hydraulic machines and electric motors to them and test bed for realizing the method, Inventors: Drozdov A.N., Dem’yanova L.A.