A Deep Learning-Based Intelligent Quantification Method for Wellbore Flow Rate

Abstract This paper proposes a classification and recognition method for Distributed Acoustic Sensing (DAS) signals based on a two-dimensional attention-enhanced YOLOv9 network, aimed at intelligent monitoring and identification of downhole flow in wells. Addressing the issues of low accuracy and poor real-time performance of traditional flow monitoring technologies in complex downhole environments, this invention transforms the one-dimensional time-series vibration signals collected by DAS into two-dimensional time-frequency spectra through Short-Time Fourier Transform (STFT). It then combines the YOLOv9 network with its advanced detection capabilities to adaptively extract temporal, frequency, and spatial features of the signals, achieving precise classification of low, medium, and high flow rate vibration patterns. Experimental results demonstrate that this method significantly enhances the accuracy and robustness of signal recognition, resolving the issues of low precision and poor real-time performance encountered by traditional methods in complex downhole environments. The application of this study can provide reliable technical support for monitoring liquid profiles during fracturing processes and interpreting production profiles in the production phase, thereby facilitating real-time control and efficiency improvement in unconventional oil and gas development.