Machine Learning Models For Risk Modelling Of Pipeline Systems

Paper No. IPC2022- 87258

This paper explores using statistical and machine learning models, like logistic regression and random forest, to quantify pipe failure probability in distribution pipelines. Trained on historical incident records, model performance is compared using lift charts. The study discusses each model's strengths, limitations, and data integration. Case studies show how training data quantity and external datasets impact effectiveness. The results aim to guide operators in developing machine learning models for pipeline risk assessment and integrity management.