Multi-objective optimization for water distribution system design using a meta-metaheuristic

The design of a water distribution system (WDS) involves finding an acceptable trade-off between multiple conflicting objectives, particularly in terms of minimizing cost and maximizing system benefits (such as hydraulic reliability). The goal in multi-objective optimization is to find a set of design solutions which embodies an acceptable trade-off between these costs and benefits, enabling the decisionmaker to make an informed decision. In addition to satisfying consumer water needs within specified pressure ranges, a system must be built to accommodate multiple demand loading conditions, withstand component failures and allow surplus capacity for growth. Towards this end, a surrogate measure for hydraulic reliability named network resilience (Prasad and Park 2004) is employed in this paper. Recent multi-objective WDS design optimization studies have achieved success using multi-objective evolutionary algorithms, such as the Non-dominated Sorting Genetic Algorithm II (NSGA-II). However, the multi-objective design of a large WDS within a reasonable timeframe remains a formidable problem, owing to the extremely high computational complexity of the problem. In this paper, a recent evolutionary meta-metaheuristic called AMALGAM (Vrugt and Robinson 2007) is applied for the first time to WDS design optimization, and compared against several popular multi-objective optimization algorithms, including NSGA-II, Differential Evolution, a Univariate Marginal Distribution algorithm, and a novel greedy algorithm developed to mimic decisions typically made by a design engineer. AMALGAM employs multiple metaheuristics simultaneously in an attempt to improve optimization performance. These algorithms were tested with respect to a number of benchmark systems documented in the literature, and AMALGAM consistently demonstrated superior performance. This was particularly apparent for the large Exeter WDS benchmark, for which AMALGAM was able to find feasible solutions within the allotted timeframe, whereas the closest competitor was unable to do so.