Water Leaks localization system within water supply networks
Funding programme: POR FESR 2007-2013 Bando ATP (Aree Tematiche Prioritarie)
Start – end year: 2010 – 2011
S.D.I. Automazione Industriale S.r.l.
Pride Global Services S.r.l.
Dip.to Informatica Statistica Comunicazione, Disco – Università di Milano-Bicocca
Comune di Torbole (VR)
Consorzio Milano Ricerche (acting as subcontractor)
Water is an indispensable resource for the life of the people and the exercise of economic activities. The climate changes, natural and anthropogenic modifications lead to climate warming and aridity: it is necessary to take serious measures to ensure the efficient use of water resources. Along with reducing fuel consumption, associated to the rationalisation of the water consumption, the efficiency of the water distribution network is certainly the key factor: the reduction of losses of water mostly influence the efficiency of a water system. In technical terms, these losses are measured in ‘dispersed water’.
The H2O Leak project is addressing innovative solutions to assist the water systems Operators in the pursuit of a more efficient distribution networks. As a consequence, the anticipation of gradual price increase in its use, is increasingly important to guarantee proper management of water resources.
The experience gained by several operators of water systems has shown that the procedure adopted for the detection of leaks through:
- distribution districts networking,
- night water flow analysis of the distribution districts,
- identification of the physical loss with the use of electro-acoustic instruments.
Operatively, H2O Leak project aims at designing and developing a tool for localizing leaks by providing a suitable clustering of the network into independent sub-sectors (District Meter Areas, DMA). Input data will be acquired by a Supervisory Control & Data Acquisition system (SCADA) deployed for continuous online monitoring of key features (pressure and flow) at crucial points of the net (e.g., at the entry and exit of districts).
Within this project we propose a leak localization approach based on Supervised and Unsupervised Learning together with an intensive use of hydrologic simulation. First, several leaks (different for location and intensity) are simulated both on the entire network and on isolated districts; the related key measures obtained through simulation are compared to those obtained on the network without leaks. These data are used for training a classification model for a first-level localization aimed at identifying the district that is most probably affected by the leaks. At a second-level, a clustering for identifying which set of pipes, within the selected district, are most probably associated to the observed variations. The performances of the proposed approach are validated on a set of artificially simulated data and tested on real data acquired from the water distribution network of the City of Torbole, a little town on the shore of Lake Garda.