First Principles and Machine Learning Virtual Flow Metering: A Literature Review

Virtual Flow Metering (VFM) is an increasingly attractive method for estimation of multiphase flowrates in oil and gas production systems. Instead of using expensive hardware metering devices, numerical models are used to compute the flowrates by using readily available field measurements such as pressure and temperature. Currently, several VFM methods and software are developed which differ by their methodological nature and the industry use. In this paper, we review the state-of-the-art of VFM methods, the applied numerical models, field experience and current research activity. In addition, we identify gaps for future VFM research and development. The review shows that VFM is an active field of research, which has the potential to be used as a standalone metering solution or as a back-up for physical multiphase flow meters. However, to increase the value of VFM technology for oil and gas operators, future research should focus on developing auto-tuning and calibration methods which account for changes of fluid properties and operation conditions. In addition, the review shows that the potential of machine learning methods in VFM is not fully revealed, and future research should focus on developing robust methods which are able to quantify flow estimation uncertainties and incorporate first principle models that will result in more accurate and robust hybrid VFM systems. Finally, our review reveals that dynamic state estimation methods combined with first principles and machine learning models could further improve the VFM accuracy, especially under transient conditions, but implementation of these methods can be challenging, and further research is required to make them robust.