Abstract: As the development period of shale gas reservoirs extends,factors such as string corrosion and erosion have led to increasingly serious wellbore integrity problems in shale gas wells,which seriously affects the liquid unloading process and suppresses gas well productivity.To address this issue,a real-time wellbore integrity diagnostic model was established using a logistic regression algorithm and field monitoring data.On this basis,a long short-term memory(LSTM)neural network algorithm was applied to improve diagnostic accuracy,ultimately achieving real-time,precise diagnosis of wellbore integrity.Applying this method enabled automatic and effective online diagnosis of tubing perforation and break-off in 23 shale gas wells in the test area,with an accuracy of 100%.The diagnostic efficiency for wellbore integrity issues was improved by 96.7%,the frequency of anomalies was significantly reduced,and production declines caused by wellbore integrity issues were effectively controlled,with the production loss attributable to wellbore integrity problems reduced by 78%.This study provides reference for the efficient and rapid identification of wellbore integrity issues in shale gas wells and for the intelligent application of gas well production monitoring,diagnosis,and analysis.

Key words: wellbore integrity, real-time diagnosis, deep learning, logistic regression algorithm, shale gas