Machine learning approaches and multiphase flow characteristics for advancing virtual flow metering

Accurate real-time measurement of pipeline multi-phase flow is crucial for effective reservoir management and production optimization. However, the multi-phase flow measurement is challenging due to the complex flow and fluid distribution. Conventional multi-phase flow meters are complex and expensive and require specific installation, regular calibration, and maintenance. Physical multi-phase flow meters and periodic well testing are widely used as benchmarks; however, phase-rate uncertainty can increase in multi-phase service and is sensitive to operating envelope, flow regime, and calibration/verification practices. The need for an easy and accurate measurement of multi-phase flow is essential, and virtual flow meters (VFMs) can provide a cost-effective, low-maintenance complementary approach to estimating multi-phase flow rates by leveraging machine learning (ML) techniques with available flow data. VFMs are software-based models that utilize mathematical models and readily available sensor data (such as pressure, temperature, and differential pressure) to determine flow rates without the need for physical flow meters. This comprehensive review explores the recent technologies, advancements, and approaches of VFMs. Different types of VFM models and methodologies to develop the VFM model were highlighted with a focus on ML techniques integrated with multi-phase flow characteristics. The challenges, limitations, and future directions of VFM models were also discussed, especially regarding data-driven approaches, such as data quality, model generalization, and computational constraints, which are the main challenges faced by VFM. Further future direction for the development of a virtual flowmeter, emphasizing the flow characteristics, was highlighted in the manuscript.