Solving Multiobjective Optimization Problem By Constraint Optimization

Multiobjective optimization problems (MOPs) have attracted intensive efforts from AI community due to their wide science and engineering applications.Solving MOPs has great theoretical and practical significance in recent years.Numerous multiobjective evolutionary algorithms (MOEAs) were proposed to tackle MOPs in the literature.In addition, many researchers exploited MOEAs to solve constraint optimization problems (COPs) in such a way that those constraints in COPs were transformed into objective functions.In this paper, we investigate a new framework, consider how to tackle a MOP by iteratively solving a series of COPs and propose the algorithm named multiobjective evolutionary algorithm based on constraint optimization (MEACO).In contrast to existing MOEAs, MEACO requires no complex selection mechanism or elitism strategy in solving MOPs.Given a MOP,MEACO firstly constructs a new COP by transforming all but one of objective functions into constraints. Then, the optimal solution of this COP is computed by a subroutine evolutionary algorithm so as to determine some Pareto-optimal solutions.After that, a new COP with dramatically reduced search space can be constructed using existing Pareto-optimal solutions.This new generated COP will be further solved to find more Pareto-optimal solutions.This process is repeated until the stopping criterion is met. In MEACO, we adopt Jump local search based evolutionary algorithm as constraint optimization method. Experimental results on 9 well-known MOP test problems demonstrate that our new algorithm outperforms three existing MOEAs in terms of convergence and spacing metrics.Moreover, we investigate the effect by modifying parameters and much better performance of MEACO is observed.