Flow Characteristics of Gas and Liquid in Pipeline Revealed by Machine Learning on Distributed Acoustic Sensing Data

The two-phase flow rates of gas and liquid in a pipeline are crucial parameters for optimizing gas-oil production strategies and ensuring the reliability of gas-oil transportation systems. Although available measurement techniques, such as various flow meters, offer accurate flow rate data, they might face limitations in providing distributed and real-time information at multiple points. Distributed Acoustic Sensing (DAS) offers a viable alternative for long-term, multipoint dynamic monitoring of the flow. However, the data acquired through fiber optic monitoring techniques are often difficult to analyze and process in real-time, while machine learning offers automatic identification of complex patterns and relationships within the data, enabling more precise predictions and classifications. To evaluate the feasibility of DAS technology combined with machine learning methods to estimate the gas-liquid flow rate in pipelines, an experimental loop that utilized DAS was developed to measure gas-liquid two-phase flow signals in pipelines. The machine learning method was then applied to analyze the DAS signals, based on which models were established to predict flow rates and regimes. Furthermore, validation experiments were conducted to assess the predictive performance of these models. Compared to the actual flow rates measured by electronic flowmeters, the results by integration of DAS and machine learning show the predictive accuracy of two models reach 97%. In the subsequent validation experiments, both the goodness of fit for the flow rate prediction model and accuracy for the flow regime prediction model exceeded 85%. Thus, compared to current flow measurement methods, the integration of DAS and machine learning not only provides accurate flow rate estimations but also offers available prediction of flow regimes, enhancing the measurement capabilities and technology insights.