Single well virtual metering research and application based on hybrid modeling of machine learning and mechanism model

Abstract

As a supplement or alternative metering method for physical metering devices, Virtual Flow Metering (VFM) is more and more widely used in the upstream metering scenario of oil and gas fields. To enhance the accuracy of single well VFM system and improve the system maintenance convenience, a new VFM method based on “hybrid model” which combines mechanism model and data model is proposed on the basis of the traditional mechanism model of dynamic multi-phase flow, the interactive logic between machine learning model of big data on single well production condition and mechanism model is established. The new VFM system architecture can realize the stable and accurate operation of the new “hybrid model” VFM system driven by real-time measurement data. The research show that: (1) machine learning model of big data on single well production condition can realize the second index output of three-phase flow under real-time measurement data input; (2) The field data are used to evaluate the accuracy of new VFM results, shows the average errors of liquid flow and gas flow are less than 5% and 3% respectively, and the metering accuracy is improved to a certain extent compared with traditional VFM method; (3) Compared with virtual metering based on the traditional mechanism model, the new VFM system is more convenient in later model maintenance and only adjust few model setting according WC and GOR changing in field data.