In water distribution systems, pressure transients provide information about the state of the system and its response to changes in operations, demand fluctuations, and pipe failures. Advanced sensing and data logging techniques enable exploring and characterizing pressure transient patterns, which are typically estimated through modeling and simulation, if not ignored. In this study, an approach based on time-series clustering is proposed to discover patterns in pressure transients from high-resolution pressure signals, which are collected by a network of pressure sensors. Six widely used clustering algorithms are tested to identify repeating patterns: k-means, k-medoids, hierarchical clustering, density-based spatial clustering of applications with noise, affinity propagation, and clustering by fast search and find of density peaks (SFDP). Three performance scores are suggested to evaluate and compare the quality of the results, including Silhouette coefficient, sum of squared errors, and Calinski-Harabaz index. Initial simulations indicate that k-means and SFDP consistently provide better performance results compared to other algorithms as well as being computationally efficient.
展开▼
