Predicting CO₂ Corrosion of Natural Gas Pipeline Transport using Supervised Machine Learning Models

Abstract This study develops a robust machine learning model to predict the CO₂ corrosion rates of wet natural gas pipelines, a crucial aspect of pipeline management in energy systems. It uses a combination of field inspection and HYSYS simulation data to train and evaluate eight regression models, including Linear Regression (LR), Ridge (R), Decision Trees (DT), Bagging (B), Extra Trees (ET), Gradient Boosting Decision Tree (GBDT), Extreme Gradient Boosting (XGBoost) and Random Forest (RF). The expectation-maximization method is applied to fill in the missing values in the combined dataset. Additionally, each model undergoes k-fold cross-validation and hyperparameter tuning to ensure high performance and accuracy. Feature selection identified key corrosion predictors such as temperature, CO₂ partial pressure, pH, and inhibitor efficiency for predicting the CO₂ corrosion rate. The bagging regression model outperformed the other models, achieving R² scores of 0.978 for the combined datasets. The proposed machine learning framework offers a cost-effective, data-driven approach for improving pipeline design and integrity management.