The reliability prediction of corroded pipelines is an important part of ensuring the safe operation of pipeline systems. Aiming at the deficiencies of the existing reliability evaluation methods for corroded pipelines in dealing with the correlation of multiple defects and computational efficiency, a method for predicting the reliability of corroded pipelines based on the model of LSTM(long short-term memory) integrated with the attention mechanism was proposed. First, this method constructed the limit state equation of the pipeline section failure and calculated the failure probability of a single corrosion defect. Second, the Gaussian Copula function was used to characterize the correlation of multiple defects, and a joint failure density function was established. Then, failure samples were generated through Monte Carlo sampling, and the dynamic prediction of the failure probability was realized by the model of LSTM integrated with the attention mechanism. Combined with the pipeline internal inspection data of the China-Russia Eastern Gas Pipeline, the influences of the number of defects and the corrosion growth model on the reliability of the pipeline section were analyzed. The results show that, compared with the traditional sampling algorithm, the average absolute error of the proposed method for predicting the reliability of corroded pipelines is only 0.039, and the calculation time is only 1% of that of the Monte Carlo method. The research achievements provide a new technical idea for the reliability evaluation of corroded oil and gas pipelines with multiple defects.