Morphing of Water Networks

Along with increases in the level of research activity surrounding sensor placement and event detection algorithms come greater needs for realistic datasets. In particular, realistic water network models are hard to obtain. Utilities are justifiably reluctant to release their models because of security concerns. The result is that all too many papers have presented algorithmic work evaluated only on networks with fewer than 100 nodes. In this paper, we explore the idea of "morphing" a water distribution network in order to defeat state-of-the-art layout reconstruction methods. Specifically, we solve the problem of finding new lengths and diameters for the pipes that are maximally different from the original parameters, subject to the constraint that travel times through the network are approximately preserved. The resulting networks are appropriate for evaluating sensor placement algorithms, but otherwise have strange characteristics. For example, some pipes may be extremely long or short, and the pressures and heads may be quite confusing to an adversary attempting to identify the network. We give results for small networks and discuss the challenges of scaling our algorithm up to larger networks.