Risk Assessment of Water Supply Pipe Leakage Based on Random Survival Forest
The issue of water supply network leakage is receiving increasing attention,as pipe leakage not only leads to the waste of water resources but may also trigger urban safety incidents.Risk assessment for network leakage is an essential step for water utilities to control leakage.In the survival analysis of water supply pipes,pipes that do not leak are considered to be censored data,and pipes that are not leaking are also important when building a leakage risk assessment model.Capitalizing on the strengths of survival analysis in addressing the censored pipe data,this study built a leakage risk assessment model for the H City water supply network using the random survival forest algorithm.The results indicated that the model had good predictive accuracy,with a C-index of over 0.75.Moreover,the model enabled the precise identification of key factors of leakage,with inherent pipe attributed having a greater impact than environmental variables through the variable importance result.This study further analyzed the dynamic changes in leakage risk under different temperatures and rainfall conditions categorizing the predicted leakage risk into five levels.The findings revealed that under cold and dry climates,there was a highest number of pipes classified as risk levels Ⅱ and Ⅲ.This indicated that the probability of leakage occurrence in the water supply pipeline network was highest during winter dry weather.Additionally,the model predicted the survival probability of pipes at different ages in the future,showing that the number of leaking pipes in H City's water supply network will significantly increase after ten years.This research provided short-term and long-term leakage risk forecasts for water utilities,considering the impact of environmental factors on leakage,which could help in developing more precise pipe maintenance and renewal strategies.
random survival forestsurvival analysiswater supply networkrisk assessmentcensored data
国家科技期刊平台
NSTL
万方数据
