Least-cost robust design optimization of water distribution systems under multiple loading

Least-cost design of water distribution system is a well-known problem in the literature. The formulation of the least-cost design problem started by deterministic modeling and later by more sophisticated stochastic models that incorporate uncertainties related to the problem's parameters. Recently, a new nonprobabilistic modeling, titled the robust counterpart (RC) approach, has been developed for the least-cost design problem to incorporate the uncertainty without the need for full stochastic information. These nonprobabilistic methods, developed in the field of robust optimization, were shown to be advantageous over classical stochastic methods in the following aspects: tractability and computation time, nonnecessity of full probabilistic information, and the ability to integrate correlation of uncertain parameters aspects without adding complexity. Former studies have considered the RC approach for a special case of the least-cost problem with a single load demand uncertainty, and single gravitational source to simplify the problem formulation and facilitate the use of the method. This special case does not handle the joint temporal and spatial correlations between the problem uncertainties and does not include components such as pumping stations and storage facilities. These new components require trading off capital and operation (i.e., energy) costs in the objective function, as the design cost is explicitly influenced by the demand uncertainty, unlike the situation where only capital cost is considered. In this study, the RC approach is expanded to cover the general least-cost design problem, including (1) multiload patterns, (2) pumping stations, and (3) storage facilities. The unknowns are the pipe diameters, pump and tank capacities, and the heads added by the pumping stations. The problem is solved using the cross-entropy method for several possible protection levels, which are defined by the size of the uncertainty set. The results are demonstrated on two examples to show the trade-off between cost and reliability and test the network's ability to cope with unexpected scenarios.