Abstract: Liquid accumulation in gas wells significantly restricts efficient natural gas extraction, presenting a substantial challenge in the industry. Hydraulic jet pumps, recognized for their simple design, lack of mechanical moving parts, and cost-effectiveness, are increasingly adopted to address this issue. However, jet pumps often exhibit limited operational efficiency, hindering broader application. This study aims to optimize the gas-liquid two-phase flow characteristics within jet pumps, enhancing both their operational efficiency and fluid discharge capacity. To achieve this, computational fluid dynamics (CFD) modeling was combined with advanced machine learning techniques. The research initially identified critical operational and structural parameters, such as the area ratio, power fluid pressure, gas volume fraction, nozzle-throat gap distance, throat length, and diffuser angle. A support vector machine (SVM) model, optimized by Bayesian methods, was then employed to predict the complex interactions among these parameters, facilitating targeted optimization. Subsequently, the NSGA-II multi-objective optimization algorithm was applied to simultaneously improve pump efficiency and discharge capacity, resulting in a Pareto optimal solution. Through detailed numerical simulations and validations, the study achieved significant performance enhancements. Specifically, the optimized jet pump configuration improved the liquid discharge rate from 42.05 m3/d to 55.24 m3/d, representing an increase of 31.40%. Additionally, operational efficiency improved from 26.50% to 30.46%, an increment of approximately 3.96 percentage points. The results conclusively demonstrate that combining numerical simulation techniques with optimized machine learning models can effectively address critical performance constraints in jet pumps. The proposed methodology not only significantly enhances fluid transport capacity but also notably improves energy efficiency. This approach provides practical insights and a robust framework for further improving hydraulic jet pump technology in complex gas-well discharge scenarios.

Key words: jet pump, gas well deliquification, two-phase flow simulation, parameter optimization, machine learning

摘要： 井筒积液问题已成为制约天然气田高效开发的关键技术难题，而射流泵凭借无须机械驱动、运行成本低等优势，成为气井排水采气工艺的重要设备。然而，其运行过程中通常面临效率偏低的技术瓶颈。本工作结合气液两相流数值模拟与机器学习方法，对射流泵内部结构、工况参数进行优化设计。结果表明，采用支持向量机模型及优化算法能够实现效率与排液量协同优化。排液量由42.05 m3/d提升至55.24 m3/d，增幅31.40%；效率从26.50%提升至30.46%，增幅3.96个百分点。该优化策略有效提升了射流泵的输送能力和能量利用效率。

关键词: 射流泵, 排水采气, 两相流模拟, 参数优化, 机器学习