Application of Machine Learning Algorithms and Integrated Production Modelling to Improve Accuracy of Liquid Production Rate Measurements Using Multiphase Flow Meters

Abstract

Objectives/Scope: During the period of two years the difference between sum of daily oil flow rate measurements of each oil production well using multiphase flow meter (MPFM) and cumulative daily oil production rate measured by custody transfer meter increased overall by 5%. For some wells inaccuracy of MPFM liquid rate measurement could reach 30-50%. The main goal of this research was to improve the accuracy of multiphase flow meter production rate measurements. Methods, Procedures, Process: More than 80 oil production wells were involved in the research, more than 100 flow rate tests were carried out. Machine learning methods such as supervised learning algorithms (linear and nonlinear regressions, method of gradient descent, finite differences algorithm, etc.) have been applied coupled with Integrated production modelling tools such as PROSPER and OpenServer in order to develop a function representing correlation between MPFM parameters and flow rate error. Results, Observations, Conclusions: The difference between cumulative daily oil production rate measured by custody transfer meter and multiphase flow meters decreased to 0.5%. The solution has been officially applied at the oil field and saved USD 500K to the Company. The reliability of the function was then proved by the vendor of MPFMs. Novel/Additive Information: For the first time machine learning algorithms coupled with Integrated Production modelling tools have been used to improve the accuracy of multiphase flow meter production rate measurements.