多源数据融合驱动的城市管网腐蚀风险预测与评估OA

Urban buried pipeline network corrosion failure prediction is a critical technical challenge for ensuring water supply safety.To address the limitations of traditional physical models and the lack of physical constraints in machine learning methods,this paper proposes a physics-informed machine learning approach that integrates finite element simulation data with actual operational data.First,a three-dimensional finite element model for buried water supply pipelines was established and validated through full-scale experiments,with Morris sensitivity analysis identifying six key influencing factors including corrosion depth and internal pressure.Next,operational data from 114702 water supply pipelines in Shanghai were analyzed,and a comprehensive risk index(CRI)incorporating physical mechanisms was constructed through parameter mapping.Finally,the synthetic minority over-sampling technique for nominal and continuous(SMOTENC)technique was employed to address data imbalance,and a light gradient boosting machine(LightGBM)prediction model was established.The results demonstrate that the CRI,as the most important feature effectively distinguishes corrosion failure pipelines,with failed samples showing significantly higher CRI values than normal samples.The LightGBM-SMOTENC model achieved accuracy,recall,and AUC values of 0.928,0.892,and 0.943,respectively,significantly outperforming traditional methods.The proposed five-level pipeline network corrosion risk classification system identifies 0.42％of pipelines as extremely high-risk,providing scientific basis for precision maintenance.