An active learning framework assisted development of corrosion risk assessment strategies for offshore pipelines

Aggressive fluids and prolonged service life result in an increasing internal corrosion risk of offshore pipelines, especially for perforation. A framework was constructed by active learning (AL) aimed at assessing the future corrosion risk of offshore pipelines and rapidly developing in-line inspection (ILI) strategies. Compared with Support Vector Regression (SVR), K-Nearest Neighbors (KNN) and Random Forest (RF), the Gradient Boosting Regression (GBR) model exhibits greater robustness and stability under both 10-fold cross-validation (CV) and bootstrap method. The RMSE and R2 are 1.48 and 0.83, respectively. Through the double iterative operation of input parameters and models, the AL framework can significantly reduce the demand for training samples, while maintaining accurate predictions. This study contributes to promoting the application of active learning methods in pipeline integrity management and planning, and providing assistance for the construction of intelligent and digital oilfields.