Abstract

Water distribution networks (WDNs) play a vital role in reliably delivering clean potable water to the society. The deterioration of water infrastructures that have drastically increased throughout the major urban centers has caused increasing water-main (WM) failures with severe consequences such as disruption of services and revenue losses. Effective management of WMs is essential. This involves repairing the WMs and implementing strategies to minimize water loss. Models should be used to predict breaks ahead of their occurrences and to plan rehabilitation. The use of these models would promote sustainable infrastructures and save costs. This research integrates the probability of failure (POF) derived from a random forest failure prediction model (predictive analytics) with a risk management strategy (prescriptive analytics) in a cold region in Canada. To investigate the effect of environmental factors, freezing index was considered and found to be among the top three most important attributes. In the proposed predictive analytics process Principal Component Analysis (PCA) was implemented for data reduction. Clustering is applied to avoid under/overestimating WM failure prediction and find the most similar cohorts. The results outlined that clustering considerably improved the prediction and risk-analysis outcomes. Finally, with the proposed risk management strategy, results showed that 3.68% of the network's total length is at high risk, and needs immediate action for fixing; however, it is only 0.07 to 1.02% of the network's total length when clustering was performed. Therefore, there was a 67–80% improvement in having WMs with high-rating risk compared to when no clustering was performed.