An Experimental Study on Gas-Liquid Two-Phase Flow-Pattern Classification in Gas Wells Using Machine Learning Techniques

• Item type: SIG Technical Reports(1)
• 公開日: 2021-02-22
• タイトル: An Experimental Study on Gas-Liquid Two-Phase Flow-Pattern Classification in Gas Wells Using Machine Learning Techniques
• タイトル / 言語: en
• タイトル: An Experimental Study on Gas-Liquid Two-Phase Flow-Pattern Classification in Gas Wells Using Machine Learning Techniques
• 言語: eng
• キーワード / 主題Scheme: Other
• キーワード / 主題: SeMI
• 資源タイプ / 資源タイプ識別子: http://purl.org/coar/resource_type/c_18gh
• 資源タイプ: technical report
• 著者所属: Graduate School of Engineering, The University of Tokyo
• 著者所属: Graduate School of Engineering, The University of Tokyo
• 著者所属: Graduate School of Engineering, The University of Tokyo
• 著者所属(英): en
• 著者所属(英): Graduate School of Engineering, The University of Tokyo
• 著者所属(英): en
• 著者所属(英): Graduate School of Engineering, The University of Tokyo
• 著者所属(英): en
• 著者所属(英): Graduate School of Engineering, The University of Tokyo
• 著者名: Meshal, Almashan; Yoshiaki, Naruse; Hiroyuki, Morikawa
• 著者名(英): Meshal, Almashan; Yoshiaki, Naruse; Hiroyuki, Morikawa
• 論文抄録 / 内容記述タイプ: Other
• 論文抄録 / 内容記述: When gas and liquid are flowing simultaneously inside a transport conduit, their spatial distribution is referred to as the flow pattern of a multi-phase flow. In the industry of petroleum engineering, multiphase flow characterization is required in many applications. Therefore, an accurate gas-liquid flow pattern predictive model is what the industry is aiming for in this field. Flow patterns change from one to another based on the gas and the liquid flow rates, the properties of fluid and gas, the diameter and the incline of the pipeline, and based on other fluid mechanical properties. The datasets used in this study were collected from a natural gas facility in Niigata in Japan, by Japan Oil, Gas and Metals National Corporation (JOGMEC). The experimentally derived datasets include the superficial gas velocity, the superficial liquid velocity, the flow pattern, the pressure, the temperature and the liquid holdup. The data acquisition was performed with a pipeline of a large diameter (106.3 mm) and it was inclined at three different angles (0°, 1°, and 3°). The pressure was set at two different rates (592 and 2060 kPa) and the liquid and gas flow rates were covering a wide range of flow rates. In this study, a machine learning based model is trained and tested in predicting gas-liquid two-phase flow patterns. The multiclass decision jungle algorithm was used in building the predictive model. The dataset is imbalanced. Therefore,; over-sampling and under-sampling techniques were used as a preprocessing step before training the predictive model, in order to increase the prediction accuracy. A comparative study has been carried out in this research to compare the performance of the multiclass decision jungle algorithm with other classification machine learning algorithms. Results show that the classified two-phase gas-liquid flow-patterns are in a good agreement with the experimental ones. Furthermore, the feature importance of the trained models is discussed in this study.
• 論文抄録(英) / 内容記述タイプ: Other
• 論文抄録(英) / 内容記述: When gas and liquid are flowing simultaneously inside a transport conduit, their spatial distribution is referred to as the flow pattern of a multi-phase flow. In the industry of petroleum engineering, multiphase flow characterization is required in many applications. Therefore, an accurate gas-liquid flow pattern predictive model is what the industry is aiming for in this field. Flow patterns change from one to another based on the gas and the liquid flow rates, the properties of fluid and gas, the diameter and the incline of the pipeline, and based on other fluid mechanical properties. The datasets used in this study were collected from a natural gas facility in Niigata in Japan, by Japan Oil, Gas and Metals National Corporation (JOGMEC). The experimentally derived datasets include the superficial gas velocity, the superficial liquid velocity, the flow pattern, the pressure, the temperature and the liquid holdup. The data acquisition was performed with a pipeline of a large diameter (106.3 mm) and it was inclined at three different angles (0°, 1°, and 3°). The pressure was set at two different rates (592 and 2060 kPa) and the liquid and gas flow rates were covering a wide range of flow rates. In this study, a machine learning based model is trained and tested in predicting gas-liquid two-phase flow patterns. The multiclass decision jungle algorithm was used in building the predictive model. The dataset is imbalanced. Therefore,; over-sampling and under-sampling techniques were used as a preprocessing step before training the predictive model, in order to increase the prediction accuracy. A comparative study has been carried out in this research to compare the performance of the multiclass decision jungle algorithm with other classification machine learning algorithms. Results show that the classified two-phase gas-liquid flow-patterns are in a good agreement with the experimental ones. Furthermore, the feature importance of the trained models is discussed in this study.
• 書誌レコードID / 収録物識別子タイプ: NCID
• 書誌レコードID / 収録物識別子: AA11838947
• 書誌情報: 研究報告ユビキタスコンピューティングシステム（UBI）; 巻 2021-UBI-69, 号 7, p. 1-6, 発行日 2021-02-22
• ISSN / 収録物識別子タイプ: ISSN
• ISSN / 収録物識別子: 2188-8698
• Notice: SIG Technical Reports are nonrefereed and hence may later appear in any journals, conferences, symposia, etc.
• 出版者 / 言語: ja
• 出版者: 情報処理学会