Ensemble machine learning-based virtual multiphase flow metering in high gas/oil ratio and water-cut reservoirs

By combining data-driven ensemble machine-learning algorithms and historical oil field portable test reports, this paper proposes a Data-Drive Multiphase Virtual Flow Meter (DD-MVFM) that estimates oil, gas, and water flow rates, provides real-time monitoring, and predicts future production for a 6-month period with appropriate accuracy. The proposed DD-MVFM utilizes the existing hardware used for measurements of basic variables such as temperature, and pressure at different locations at the well-head structure. The DD-MVFM can be employed in three ways. The first way is to be used as a verification tool for multiphase physical flow meters (MPFMs), making sure they are working properly and increasing confidence in the collected readings. The second way is to use the DD-MVFM as a redundant tool when the MPFMs are not available or going through maintenance. The third way, which is the main objective of our research, is to employ the proposed DD-MVFM as a stand-alone for the complete replacement of current and future MPFM installments. This, significantly lowers the operating cost, reducing the required portable field tests, and saving the need to build a major infrastructure for the set-up of MPFMs for new oil wells. Consequently, this contributes to the ambitious goal of reducing CO₂ emissions. The DD-MVFM's development involves the fusion of data wrangling and machine learning algorithms for optimal performance. Initial testing indicates an 85 % correlation with the actual production rates, with potential for further improvement as more field test data is incorporated, making it a pioneering solution in the field of oil and gas management.