Data Driven Water Pipe Failure Prediction: A Bayesian Nonparametric Approach

Water pipe failures can cause significant economic and social costs, hence have become the primary challenge to water utilities. In this paper, we propose a Bayesian nonparametric approach, namely the Dirichlet process mixture of hierarchical beta process model, for water pipe failure prediction. It can select high-risk pipes for physical condition assessment, thereby preventing disastrous failures proactively. The proposed method is adaptable to the diversity of failure patterns. Its model structure and complexity can automatically adjust according to observed data. Additionally, the sparse failure data problem that often occurs in real-world data is tackled by the proposed method via flexible pipe grouping and failure data sharing. An approximated yet computational efficient Metropolis-within-Gibbs sampling method is developed with the exploitation of the failure data sparsity for model parameter inference.The proposed method has been applied to a metropolitan water supply network. The details of the application context are also presented for demonstrating its real-life impact. The comparison experiments conducted on the metropolitan water pipe data show that the proposed approach significantly outperforms the state-of-the-art prediction methods, and it is capable of bringing enormous economic and social savings to water utilities.