Abstract: Corrosion-induced wall thinning in buried metal pipelines poses a significant threat to the long-term safety and integrity of pipeline networks operating in complex soil environments. To predict the remaining service life of such pipelines, this study constructs a deep learning framework that integrates spatial feature extraction and time-series modeling by analyzing pipeline corrosion monitoring data. An initial quantitative analysis of the influencing factors using random forest (RF) identifies service time and soil pH as the most critical features, with important scores of 31.9% and 21.7%, respectively. Compared with the convolutional neural network (CNN), long short-term memory (LSTM), and convolutional neural network - long short-term memory (CNN-LSTM) models, the BKA-CNN-LSTM model of the black kite algorithm (BKA) shows significant advantages in convergence speed and prediction accuracy. The model achieves a coefficient of determination (R²) of 0.967 5, with mean absolute error (MAE), mean square error (MSE), and root mean square error (RMSE) values of 2.314, 0.238, and 0.350 9, respectively. All predicted values fall within the 95% error band. The results indicate that the proposed model can accurately characterize pipeline corrosion development, offering considerable value for pipeline safety assessment and maintenance decision-making.