Exploring The Effects Of Constraints On Multiobjective Evolutionary Algorithm Optimization Efficiency And Effectiveness In Water Resources
Clarkin, Timothy 1 ; Raseman, William 2 ; Kasprzyk, Joseph 3 ; Herman, Jon 4
1 ¾«Æ·SMÔÚÏßӰƬ
2 ¾«Æ·SMÔÚÏßӰƬ
3 ¾«Æ·SMÔÚÏßӰƬ
4 UC Davis
This research explores how user-defined constraints affect the efficiency and effectiveness of multi-objective evolutionary algorithm (MOEA) optimization in water resources. Constraints in MOEA optimization commonly represent limits on acceptable performance, but their effect has not been extensively researched. The study considers two water resources problems: a water supply portfolio planning model, and an economic development and environmental water quality model. These models are optimized in two different cases: one with constraints – as the problems were originally formulated – and one without constraints. For each model, the effectiveness (fig. 1) and efficiency of search (fig. 2) on the constrained and unconstrained problems are compared. The reference sets of solutions are compared (fig. 1a) and the original set of constraints are applied a posteriori (fig. 1b). Plots of hypervolume metric versus functional evaluations (fig. 2) are used to determine search progress. Initial results suggest that constraints aid in the search process by favoring selection of solutions that meet the decision maker’s preferences. While in some cases constraints can make a problem harder to solve, they allow the search process to more efficiently and effectively produce acceptable solutions.