Multi-objective Optimization Strategies based on Decomposition for the Design of Water Distribution Systems

The use of Pareto dominance for evaluating the solutions has been the mainstream in evolutionary multi objective optimization for the last two decades. An alternative is multi objective evolutionary algorithm based on decomposition (MOEA/D) which uses scalarizing the objective functions. In this paper, decomposition strategies are developed for water distribution network (WDN) design problems by integrating the concepts of genetic algorithm (GA) within the MOEA/D framework. The proposed algorithms are then compared with the two well-known non-dominance based MOEAs: NSGA2, and SPEA2. This comparison is made by plotting the Pareto fronts and evaluating the hyper-volume and two-set coverage metrics across two large scale WDN design problems. Experimental results show that MOEA/D outperform the Pareto dominance methods in terms of both non-domination and diversity criteria. It suggests that decomposition based multi objective evolutionary algorithms are very promising in dealing with real-world water engineering optimization problems.