Corrosion prediction of deep oil well tubing based on improved Stacking ensemble learning approach

To enhance the prediction accuracy for both average corrosion rate and pitting corrosion rate of oil well tubing in deep complex environments, and to address the issue of insufficient consideration of base learner heterogeneity in traditional Stacking ensemble learning, an improved Stacking ensemble learning algorithm based on the coefficient of determination (R2) is proposed. This algorithm integrates four machine learning models as base learners: extreme gradient boosting (XGBoost), random forest (RF), support vector regression (SVR), and gradient boosting decision tree (GBDT). The outputs of these base learners are weighted according to their respective R2, and the weighted combination forms the input dataset for the meta-learner. Experimental results demonstrate that, compared with the traditional Stacking ensemble method, the improved model achieves a 25.9% reduction in mean absolute error (MAE) and a 9.7% reduction in mean squared error (MSE) for average corrosion rate prediction, alongside a 2.3% increase in the R2. For pitting corrosion rate prediction, it yields reductions of 11.6% for MAE and 2.0% for MSE, respectively, with a 2.7% increase for R2. These results validate the effectiveness of the proposed algorithm, and the research findings provide valuable support for corrosion prevention, control and safe operational maintenance of deep oil well tubing.