Community detection as a tool for DMA identification

Water losses, the portion of water introduced in a pipe network but not consumed by users, represent a significant problem in water distribution system (WDS) management. Modern guidelines suggest to divide the pipe network in clusters, in order to compute a water balance and measure water consumption by each group. These clusters are called district metered areas (DMAs). The division of a pipe network in DMAs is usually realized with a visual exam supported by technical experience. This approach, which is convenient for small WDSs, becomes dicult to apply to large WDSs characterized by thousands of user nodes and pipes. Therefore, it is necessary to have an automatic tool to recognize the affinity degree of neighbouring nodes and to decide how to assign a node to a particular DMA. We propose an automated approach to subdivide pipes, that only requires flow rates through the network. The method has been tested to a large WDS often used as benchmark. The approach successfully divides the pipe network in an acceptable number of DMAs. Each resulting DMA is characterized by a low number of external links and by a proper number of users.