Optimal placement of quality sensors in water distribution systems

Water supply infrastructures are crucial for the sustainable existence and development of modern cities [1,2]. Water distribution systems (WDSs) are complex structures formed by many elements designed and erected to transport water of sufficient quality from water sources to consumers. The amount of the above elements, which can reach up to tens of thousands of links and junctions, their frequently wide spatial dispersion and the WDS characteristic of being very dynamic structures make the management of real WDSs a complex problem [3-5]. Moreover, although the main objective is to supply water in the quantity and quality required, other requirements are essential, namely maintaining conditions far from failure scenarios [6,7], ability to quickly detect sources of contamination intrusion [8,9], minimization of leaks [10-12], etc. Most of these objectives may be achieved through suitable location of sensors along the network and, currently, an increasing number of efforts are carried out in this direction [12-14]. The identification of potential contaminant intrusion in water networks is a crucial point to fully guarantee water quality in WDSs. As a consequence, water utilities are bound to measure water quality parameters continuously, so that quality can be adequately monitored. To this end, an optimal lattice of sensors should be designed that covers strategical points of the water network [15]. It is a matter of safety and security arrangement in WDS management, and sensors cannot be randomly placed along the network. Placing sensors may seem simple at the beginning, but considering sensor station costs and the extension of the network that should be covered, it turns out to be a challenging problem. The plurality of potential contaminants, the identification of the contaminant sources in the network, and the reaction time of the utilities to deal with a contamination event are also important elements to consider. This work is not intended to cover all the aspects related to network protection against potential contaminant intrusion. It will rather concentrate on proposing a solution just for the sensor placement problem, namely, optimally determining the number of sensors and their locations. And we address this optimization problem from a multi-objective perspective. Several goals should be taken into account when placing water quality sensors. Optimal sensor placement aims to achieve early contaminant detection and seclusion of affected areas so that the public exposure to contamination be minimum. First, it is desired to identify quality problems as soon as possible, it means, to minimize the detection time. Second, irrespective of the location of the contaminant source, at least one sensor should always be able to identify a quality problem; this amounts to maximizing the coverage of protection. Additionally, the bulk of poor or bad quality water consumed should be minimized; this, specifically, involves that high population density areas have to receive special attention compared to other areas with much lower consumption rate. And, importantly, the cost, which is directly proportional to the number of installed sensors, should be kept to a minimum. These objectives are mutually conflicting and improving one of them will probably result in a detriment for another. The rationale is clear. For example, maximizing the protection coverage in the network will require either to increase the number of sensors (it means the cost) or to probably be bound to accept larger detection times. Consequently, the final solution will result from a compromise among objectives rather than from a unique “best alternative”. Suitably solving problems of this nature requires the use of a multi-objective approach. Such an approach is able, for example, to answer marginal cost questions, such as if it is worth buying an additional sensor to get a reasonable improvement in another objective, because there is no way to know how much improvement in protection coverage and detection time will bring that additional sensor. Those are the kinds of questions that a multi-objective approach helps to answer. We claim that those are the kind of questions and answers needed to eventually find a sensor placement solution that represents a good trade-off among all the objectives involved. In this contribution we present the necessary materials and methods. Then, we develop contaminations scenarios and evaluate the considered objectives based on the so-called contamination matrix concept. Next, we develop a multi-objective solution using a well-known multi-objective optimization algorithm [16]. A use case corresponding to a medium-size water distribution network is presented together with the obtained results and a thorough discussion.