Identification of Gas-Water Two-Phase Flow Patterns Based on Distributed Acoustic Sensing

Influenced by different flow patterns, the interpretation of production profile logging needs to select different interpretation models to ensure interpretation accuracy. In order to ensure the rationality of the selection of interpretation models, this paper proposes a continuous wavelet transform, convolutional neural network and long short-term memory recurrent neural network (CWT-CNN-LSTM) flow patterns identification method based on distributed acoustic sensing (DAS) data. Firstly, the DAS response signals under five different flow patterns are obtained through the experiment of gas-water two-phase flow. Secondly, the original signal is converted into time-frequency graph based on continuous wavelet transform, so as to capture the time domain and frequency domain characteristics of DAS signal more comprehensively. Finally, Convolutional neural network (CNN) is combined with long short-term memory recurrent neural network (LSTM) to improve the accuracy and reliability of flow patterns identification. The test results show that the flow patterns identification accuracy of the model selected in this paper is high, and the overall identification accuracy is 98.67%. The research results can provide some guidance for the interpretation of distributed fiber production profile data.