Water Distribution Network Clustering: Graph Partitioning or Spectral Algorithms?

Abstract

Water Network Partitioning (WNP) is among the most attractive and studied strategies for the improvement of the Water Distribution Network (WDN) management. The proper definition of sub-regions (called clusters or districts) with high link density between nodes in the same group, and a relatively low link density between nodes in different groups, is a crucial aspect for the partitioning of a water system. If on one hand the definition of these monitored sub-areas, called District Metered Areas (DMAs), allows simplifying the water balance, the pressure control, the water leaks identification and the water quality protection, on the other hand it may worsen the hydraulic performance and the reliability of the system. In this paper, two clustering algorithms, graph partitioning based on a Multi-Level Recursive Bisection and Spectral Clustering, were used to define the districts. Some of the major geometrical and hydraulic characteristics of the network has been adopted as weights in the partitioning procedure. A comparison between the two clustering methods was made for a real water network of the South Italy, Parete, evaluating some clustering quality and hydraulic indices, in order to define the algorithm and the weight which work better for the definition of the optimal clustering layout.