The use of statistical methods to discern patterns of historical breakage rates and use them to predict water main breaks has been widely documented. Particularly challenging is the prediction of breaks in individual pipes, due to the natural variations that exist in all the factors that affect their deterioration and subsequent failure. This paper describes alternative models developed into operational tools that can assist network owners and planners to identify individual mains for renewal in their water distribution networks. Four models were developed and compared: a heuristic model, a naïve Bayesian classification model, a model based on logistic regression and finally a probabilistic model based on the non-homogeneous Poisson process (NHPP). These models rank individual water mains in terms of their anticipated breakage frequency, while considering both static (e.g. pipe material, diameter, vintage, surrounding soil, etc.) and dynamic (e.g. climate, operations, cathodic protection, etc.) effects influencing pipe deterioration rates.
- logistic regression
- naïve Bayesian classification
- non-homogeneous Poisson process
- pipe break forecast
- Received March 8, 2011.
- Accepted August 18, 2011.
- © National Research Council of Canada 2012