A deep learning framework integrating Transformer and LSTM architectures for pipeline corrosion rate forecasting

Abstract

Accurately predicting the corrosion rate is crucial for ensuring the safe operation of buried pipelines. Currently, research on pipeline corrosion prediction is largely confined to static methods, which do not fully capture dynamic safety considerations. In contrast, machine learning techniques can more effectively process experimental data and comprehend its complex characteristics. Based on this, this paper proposes an interpretable Transformer-LSTM (Long Short-Term Memory) model for predicting the corrosion rate of buried pipelines. Its core innovation lies in modifying the Transformer architecture by replacing the decoder layer of the traditional Transformer model with a fully connected layer and substituting the original attention layer with an LSTM layer. This modification allows the model to utilize the storage units of LSTM to effectively store and update information within the sequence. Finally, two cases were combined for case verification. Taking Case 1 as an example, the research results indicate that, compared to the LSTM model, the Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), and Root Mean Square Error (RMSE) of the Transformer LSTM model are reduced by 85.5 %, 89.8 %, and 83.2 %, respectively. In comparison to the Transformer model, the MAE, MAPE, and RMSE of the Transformer LSTM model decreased by 73.8 %, 80.5 %, and 68.6 %, respectively. Additionally, the SHapley Additive exPlanations (SHAP) method is employed to provide a global and intuitive explanation of the model, aiding in the understanding of the contribution of input features. These research findings will assist pipeline operators in better planning the operation and maintenance of pipelines.