Clustering Analysis in Water Distribution Systems for Enhanced Metering Infrastructure Retrofitting

As the population increases and covers more land, water distribution systems (WDSs) also expand to deliver potable water at adequate pressure, an essential service to communities around the world. The hydraulic networks that represent WDSs are large and complex dynamic systems. Hence, appropriate modeling methods are needed to identify influential zones or areas where water utilities can implement comprehensive measurement programs and infer the results to the rest of the network. This work presents an effective clustering method to partition water distribution systems into sub-networks and identify influential areas that can be used as hot spots by water utilities to retrofit their systems with advanced metering infrastructure components. An adapted K-means clustering algorithm analysis is applied to two benchmark hydraulic networks to highlight the effects of the different parameters of interest for clustering, that is, pressure and demand, different weights used as edge distances, and network topologies. An exhaustive search method algorithm is implemented to minimize the variation of the parameter of interest among clusters. The results show influential areas at the sub-network level represented by the clusters' centroids that can be used to infer the hydraulic conditions of other areas with different levels of accuracy. Minimum demand variation is achieved when using a combination of hydraulic and topological characteristics as edge weights. The clustering models will help researchers and practitioners select an effective partitioning tool to improve the management of water distribution systems.