Prediction of internal corrosion in gas gathering pipelines using an integrated machine learning approach

This is the authors' abstract. We don't add key points for this paper.

Gas field tubing suffers from severe internal corrosion. To guarantee safe and sustainable field operation, this study proposes a novel integrated machine learning prediction method for internal corrosion rates based on simulated corrosion experimental data. The proposed model embeds a self-attention mechanism into a hybrid framework combining temporal convolutional network (TCN) and bidirectional long short-term memory (BiLSTM), enabling the simultaneous prediction of uniform and localized corrosion rates under the coupling effect of multiple corrosion influencing factors. For performance validation, standalone TCN and BiLSTM models were constructed for comparison. The standalone models yielded minimum mean relative errors (MRE) of 25.88% for uniform corrosion and 12.74% for localized corrosion, with corresponding coefficients of determination (R²) of 0.626 and 0.216. In contrast, the self-attention-enhanced integrated model reduces the MRE to 5.14% and 2.04% for uniform and localized corrosion, respectively, while the R² values are improved to 0.974 and 0.98. The results demonstrate that the proposed model achieves a remarkable improvement in the internal corrosion prediction accuracy of gas gathering pipelines.