Abstract:The pipe failure data from 2004 to 2008 in a southern city were collected and average pipe age, tensile strength and pipe failure rate were obtained by data analysis, average pipe pressure was obtained by water distribution simulation. With the diameter classification of training samples, predictive pipe failure models which described the pipe failure rate correlations among diameters, average pipe pressure, average pipe age and tensile strength were established on the basis of the genetic programming. The feasibility and effectiveness of the models were verified by the pipe failure data from 2009. The results show that predictive pipe failure models can well describe actual conditions.
