Measurement of Air-Water Counter-Current Flow Rates in Vertical Annulus Using Multiple Differential Pressure Signals and Machine Learning

Abstract

Gas-liquid counter-current flow in vertical annulus is involved in multiple industrial fields such as petroleum engineering. For instance, in coalbed methane wells where liquid pumping is utilized, obtaining real-time gas-liquid flow in the annulus is crucial for the development and management of coalbed methane wells. However, due to complex flow conditions, this requirement is difficult to achieve through traditional flow measurement means. Therefore, this paper proposes a flow prediction method based on multiple sets of differential pressure signals and machine learning techniques. Experiments on air-water two-phase flow were conducted on a vertical annulus pipe with an inner/outer diameter of 75mm/125mm and adjustable eccentricity. The probability density function and power spectral density function of three sets of differential pressure signals collected at different heights in the annulus pipe were used as model inputs, and gas-liquid flow rate as output. A gas-liquid two-phase flow prediction model was constructed based on the artificial neural network model, and the hyper-parameters of the model were optimized using Bayesian optimization. The results show that on a test dataset of 440 combinations of conditions with air superficial velocity of 0.06~5.04m/s, water superficial velocity of 0.03~0.25m/s, and pipe eccentricity of 0, 0.25, 0.5, 0.75, 1, the model can achieve average prediction errors of 9.12% and 29.34% for gas and water flow, respectively. This indicates that the method can be applied to non-throttling, non-intrusive measurement of phase flow under annulus gas-liquid counter-current flow conditions.