Objective The integrity management of urban gas pipeline networks demands effective risk assessment methods. Corrosion leakage risk assessment necessitates the comprehensive integration of risk assessment factors with various detection operations. Current detection tasks face challenges due to data complexities and significant data deficiencies. Therefore, it is vital to develop a method for predicting and evaluating corrosion leakage risks.

Methods Key indicators associated with corrosion leakage risks were selected through a correlation analysis. These identified indicators were then employed to develop an intelligent soft detection model that integrates pipeline and environmental data, based on the K-Nearest Neighbor (KNN) and Random Forest algorithms.

Results The model conducted predictions on missing detection data and achieved indirect measurements of key indicators, with a relative error between predicted and measured values staying below 25%, meeting acceptable standards. It effectively forecasts pipeline corrosion leakage risks in instances of missing data, paving the way for additional quantitative assessments. In comparison to prior research, the model displayed enhanced prediction accuracy and reliability, attributed to innovations in extracting multi-factor coupling relationships and algorithm choices. Nonetheless, the emergence of some abnormal data suggested constraints on its predictive capacity under specific circumstances and its dependence on complete and precise data. Consequently, enhancing both the quantity and quality of detection data, along with refining the feature extraction approach for key risk indicators, is anticipated to further boost the accuracy of the model.

Conclusion This research enriches the risk prediction theory concerning corrosion leakage in gas pipelines and offers practical benefits in enhancingpipeline operation safety and reliability. Future research efforts should focus on enhancing data acquisition and analysis techniques, optimizing the model structure, and improving the model adaptability and accuracy across various application scenarios.