The Study of Predicting Corrosion Failure Risks in Urban Pipeline Networks Based on Machine Learning

Corrosion is one of the primary causes of pipeline network failures, significantly impacting urban health and safety. To prevent pipeline failures, predicting pipeline corrosion has become a critical requirement for operators in daily operations and asset management. This study uses 20 years of maintenance data from water supply networks with DN300 and above in a certain area of Shanghai as a sample, considering 10 variables as input features, including pipeline attributes (e.g., material, diameter, age) and environmental factors (e.g., burial depth, road type, distance to subway). To address the issue of data imbalance, three data processing methods are employed: random oversampling, synthetic minority over-sampling technique for nominal and continuous (SMOTENC), and Adaptive Synthetic Sampling (ADASYN). Three machine learning models, including Support Vector Classifier (SVC), Logistic Regression (LR), and LightGBM, are developed for comparative analysis to predict pipeline corrosion failure risks. The results show that the LightGBM model with SMOTENC performs excellently in identifying urban pipeline corrosion, with an accuracy of 0.936 and an AUC of 0.925. Finally, the corrosion failure risk of the pipeline network in the region was assessed, and the risk was classified into five levels from high to low using the Jenks Natural Breaks method. The distribution of the pipeline network across these risk levels is as follows: 0.42%, 2.39%, 28.68%, 35.68%, and 32.83%.This study provides valuable insights into the safe operation of urban pipeline networks and helps operators develop more effective prevention and maintenance strategies, enhancing the safety and reliability of urban water supply networks.

Keywords: Pipeline networks, Machine learning, Pipeline corrosion, Data processing, Predictive modeling, Failure prediction.