Objective Carbon capture, utilization and storage (CCUS) technology has become a critical enabler for achieving carbon neutrality. However, tubing corrosion caused by CO₂ flooding and storage severely threatens the operational safety of oil wells. Existing mechanistic models exhibit insufficient prediction accuracy under extreme operating conditions, while purely data-driven models lack physical interpretability. There is an urgent need to develop a hybrid model that integrates theoretical reliability and predictive precision.

Method Based on the electrochemical mechanism of CO₂ corrosion, a temperature-dependent theoretical model for low/high-temperature corrosion rates was developed using the Arrhenius equation to dynamically correct the coupling effects of ferrous ion concentration and temperature. For extreme corrosion rates, a Random Forest sub-model with grid search-optimized hyperparameters was implemented. Grid search was employed to optimize hyperparameter combinations, enabling precise prediction of extreme corrosion rates. Through threshold partitioning, the mechanistic model and data-driven model were dynamically coupled, enhancing prediction capabilities across all operating conditions.

Result Validation on production wells in a CCUS block showed that the mechanistic model achieved a mean relative error (MRE) of 0.27 and a root mean square error (RMSE) of 0.00067 mm/a under normal conditions. The hybrid model reduced MRE by 16% and RMSE to 0.00066 mm/a through extreme-value correction.

Conclusion This model provides a physically interpretable and high-precision tool for predicting CO₂ corrosion rates in tubing, providing a reliable tool for managing well integrity in CCUS systems, and reducing corrosion-related operational costs and safety risks.