Economical Replacement Time Prediction Of Water Supply Network Based On Survival Analysis

Water supply network is one of the city’s important infrastructures, which undertake thetask of water transport. However, pipeline aging, rapid expansion of city and poormanagement lead to pipe break incidents frequently. Pipe break is not only a waste ofresources, but also seriously affect people’s production and life, and bring pipe repair/updateproblems to management. Thus, efficient and economic methodology, such as analyzing themain factors affecting the pipe break, evaluating the economic replacement time under thethreshold break rate and establishing the break prediction model will reduce the pipe breakrate effectively and is expected to be extremely useful in the maintenance of waterdistribution system.This paper described the survival analysis theory and its common models. Thesemi-parametric Cox proportional hazards model is chosen to analyze the pipe life. Firstly, thevariables of Cox model were determined according to basic information of water pipes andhistoric records of pipe breaks in the study area. Secondly, after survival time was defined,ModelⅠand ModelⅡwere established, which the pipes were divided into two groups by thenumber of pipe failure. The standard score residuals was used to examine the proportion riskassumption. As a result, road type in ModelⅠ was the time-dependent variable. Then thesurvival function was obtained, and the distribution of survival times of the pipes wasconsidered as a Weibull distribution. Finally, the model was tested by the deviance residuals.Through analyzing the hazard ratio, the conclusion was that pipe diameter was protectivefactor, pipe length, pipe material and road type were risk factors. In a word, sensitivity onlength and road type of ModelⅠ was greater than ModelⅡ.The pipe replacement time was further studied based on Cox proportional hazards model.Firstly, individual break prediction error and cumulative break prediction error werecalculated. It is concluded that the precision of Cox model was the best when survivalprobability was0.7. Secondly, to different diameter and material, general pipe breakprediction model was established respectively based on the pipe break record and median life.The individual pipe break rate was also obtained. Then, the equation of economic replacementtime was estimated combined with the threshold break rate. It showed that the model couldpredict the trend of pipe breaks effectively. Finally, the distribution of economic life andaverage value for economic life were analyzed.