An approach to water supply clusters by semi-supervised learning

The rational distribution of water in a water supply network (WSN) is a complex problem, especially for systems of large scale. Its complexity is continually increasing from the point of view of technical management. The division of WSN into hydraulic zones is a partition of the supply network into subsystems with controlled inputs and outputs, building smaller independent networks. This solution is a strategic option used in many cities worldwide to control and operate their systems seeking to improve the WSN management, working with each part as a whole. Looking for leaks, detecting water distribution anomalies or carrying out rehabilitation plans, are instances of the aspects that can be technically improved by this reduction of the inspection area. For these reasons, it is important to design the hydraulic zones structure in some optimal way. In this paper, we propose a semi-supervised learning to approach it. To do it we add the different supply constraints to the adjacency matrix of the graph and then gathering the reality of the hydraulic zones in a single matrix. The next step splits the network, applying to it a spectral clustering algorithm. This methodology offers an adequate solution to the hydraulic zones paradigm through clusters that allow the conditions for the zones to become small quasi-independent water supply networks.