Application of Machine Learning in Wet Gas Measurement Predictions

Abstract

The Venturi tube, a classic single-phase flow meter, proved to be a reliable and accurate wet gas flow meter, but it requires correction. The presence of liquid content increases the measured differential pressure across the Venturi tube and causes over-reading (OR). Developing novel methods to correct the Venturi tube readings has become a target in the oil & gas industry to quantify production from individual wells. This work is proposing a new approach to overcome the OR challenge using machine learning (ML) models. The ML model to predict OR was developed using the random forest (RF) technique. Initially, synthetic dataset of producing wells were generated. A variation on the type of gasses and liquids were studied including nitrogen and benzene, argon and water, natural gas and water and natural gas and decan. The input parameters consist of fluid properties and fluid conditions. The fluid properties are including gas and liquid phases. The target for these inputs is to predict the OR. The results were then evaluated based on root-mean-square error (RMSE), and the fitness of the model was assessed by the coefficient of determination R2. The ML model showed high accuracy results for the OR prediction of the testing data of R2=0.997 and RMSE of 0.7%. The developed model was then applied to a new set of data of air and water for further validation. This validation resulted in R2= 0.998 and RMSE of 0.5%. This shows that the RF technique is capable of predicting the OR in wet gas when the aforementioned input parameters are used. The uniqueness of this ML model is that the inputs are all measurable in the field. Once the OR is predicted, the "real" gas mass flow rate can be calculated directly. The novelty of this work lies in providing a robust method to calculate the "real" gas mass flow rate real-time which ultimately diminishes the need to use published correlations derived from experimental conditions that are not necessarily representative of the oil field conditions.