Experimental investigation and modeling of CO 2 corrosion in enhanced oil recovery surface gathering pipelines

To systematically investigate the CO2 corrosion behavior of 20# steel and to predict the corrosion rate, an autoclave system was employed to conduct weight loss measurements. Microstructural evolution of the corrosion product film was characterized by SEM and EDS, providing direct microscopic support for analysis of corrosion mechanisms and influencing factors. It was found that corrosion rate increases with temperature and CO2 partial pressure. Increasing water cut enhances electrochemical corrosion. Cl− concentration severely impairs the compactness of the product film and thus causes severe localized corrosion. To address the challenge of limited experimental data, Synthetic Minority Over-sampling Technique for Regression (SMOTE-R) was employed for data augmentation. For corrosion rate prediction, the Whale Optimization Algorithm (WOA) was introduced to optimize the hyperparameters of Gradient Boosting Regression Tree (GBRT). In leave‑one‑out cross‑validation (LOOCV), the coefficient of determination (R2) improved to 0.934, while RMSE and MAE decreased by 31.6% and 26% when compared with traditional GBRT. For comparison, Particle Swarm Optimization (PSO), Genetic Algorithm (GA), and Bayesian optimization were also applied to tune the GBRT hyperparameters. The WOA-GBRT model consistently outperformed in prediction accuracy, demonstrating its high accuracy in predicting the CO2 corrosion rate of 20# steel.