Integration of multi-domain das data analysis and machine learning for wellbore flow regimes identification in shale gas reservoirs

Distributed Acoustic Sensing (DAS) technology offers a novel method for wellbore flow monitoring in shale gas reservoirs with its high dynamic range and real-time continuous monitoring capabilities. This study presents a multiphase flow monitoring platform based on DAS, analyzing signals in time, frequency, and time-frequency domains to extract fluid flow regimes features. Machine learning models, including Random Forest (RF), Back Propagation Neural Network (BPNN), and Decision Tree (DT), were developed for flow regime identification. Results reveal that DAS data primarily responds within 0–100 Hz, peaking at 20–50 Hz. Multi-domain feature fusion achieves high accuracy in identifying inclined well flow regimes, with RF reaching 100 %, BPNN 99.22 %, and DT 96.88 %. Both time-domain and frequency domain features provide critical support for online auxiliary identification of wellbore flow regimes. This study offers technical support for monitoring and identifying wellbore flow regimes in shale gas reservoirs, has certain reference value for deeply exploring fluid flow regimes within wellbores, and opens up new pathways for improving the efficiency of oil and gas resource exploration and development.