Prediction of internal corrosion rate for gas pipeline: A new method based on transformer architecture

Abstract

Accurate assessment of internal corrosion rates in steel natural gas pipelines is a critical process in oil and gas pipeline integrity management. However, the existing models used for predicting internal corrosion rates often suffer from various issues, such as low accuracy, poor generalization, and a lack of interpretability. In order to appropriately address these challenges, we propose CNN-BO-Transformer, and employ DeepSHAP for enhancing the interpretability of the model. The proposed CNN-BO-Transformer is used to predict the corrosion rate in natural gas pipelines, while DeepSHAP is utilized to analyze the causal relationships between input variables and model's predictions. The proposed model is validated by using a real pipeline excavation dataset obtained from a gas field located in Northwest China, achieving an average error of 0.21mm/y. This represents reductions of 69.74 % and 66.67 % as compared to the errors of support vector regression (SVR) and the Transformer model, respectively. The proposed method significantly improves the accuracy and reliability of corrosion rate predictions in natural gas gathering and transportation pipelines, thus providing an effective approach for predictive maintenance and repair of steel gathering in transmission pipelines in gas fields.