Data-driven intelligent corrosion risk assessment for long-distance natural gas pipelines

Corrosion in long-distance natural gas pipelines presents significant challenges to their safe operation and the sustainability of complex multi-segment pipeline networks. Currently, the absence of a unified predictive model for the burst pressure of corroded buried pipelines forces the industry to rely on complex physical analyses for corrosion risk assessment, resulting in inefficiency and limited accuracy. An intelligent risk assessment model for corrosion and burst failure of long-distance natural gas pipelines was developed in this study based on a data-driven approach, providing a feasible alternative to existing assessment models. In the study, 624 sets of defective pipeline burst pressure test data were collected, six prediction accuracy judgment indexes were selected, and an intelligent prediction model for the burst pressure of corroded pipelines was established using a Support Vector Machine (SVM) optimized by the Artificial Bee Colony (ABC) algorithm, referred to as the"ABC-SVM"model. The superiority of the ABC-SVM model was verified by comparing its prediction accuracy with that of the existing standard model and 15 other machine learning models, leading to the establishment of a unified burst pressure prediction model for corroded pipelines. The subset simulation method was employed to reduce the time required for calculating the probability of burst failure in corroded pipelines. Based on the calculated failure probability range, four risk levels were defined, allowing for the rapid assessment of risk levels according to the burst failure probability determined under actual working conditions of corroded pipelines. The application results demonstrated that the ABC-SVM model achieves the highest prediction accuracy and broadest applicability, with R² values of 0.97 for the training set and 0.95 for the prediction set, along with global performance indexes of 51.61 and 35.13, respectively, which are the lowest among all models evaluated. The developed risk assessment model offers a feasible alternative to existing physics-based risk assessment models.