Submarine pipeline corrosion rate prediction model based on high-dimensional mapping augmentation and residual update gradient forest

Abstract

Pipelines play a crucial role in the transportation of oil and gas, corrosion is a prevalent issue in submarine pipelines, and accurately predicting the corrosion rate is crucial for ensuring their safe operation. In light of the challenges posed by the scarcity and imbalance of corrosion data samples, this study develops a data-driven hybrid model for pipeline corrosion prediction. Firstly, grey relational analysis is employed to validate the nonlinear relationship between corrosion factors and corrosion rate. Subsequently, this study innovatively combines Kernel Principal Component Analysis (KPCA) with Variational Autoencoder (VAE) to capture the nonlinear relationships within the data and augment the sample size. The proposed Residual Update Gradient Forest (RUGF) model is then utilized to predict the augmented data. Finally, through SHapley Additive exPlanations (SHAP) and Fourier Amplitude Sensitivity Test (FAST), this study demonstrates that the model effectively identifies key principal components and provides explanations for the prediction results. Case studies reveal that the proposed model exhibits robust generalization capabilities and significantly outperforms classical regression algorithms, achieving highly accurate predictions of corrosion rate in Submarine pipeline.