OntheDifferentialEvolutionSchemesinMOEAD.pdfVIP

2010 Sixth International Conference on Natural Computation (ICNC 2010) On the Differential Evolution Schemes in MOEA/D Waibin Huang Hui Li School of Computer Computing Science School of Computer Science Zhejiang University City College, Hangzhou, China University of Nottingham, Nottingham, UK Email: huangwb@zucc.edu.cn Email: hzl@cs.nott.ac.uk Abstract—Recently, MOEA/D has gained some success in tack- whole population. It is also pointed out that the performance ling a class of continuous multi-objective optimization problems of MOEA/D-DE is sensitive to the parameter setting in DE with complicated pareto sets. It uses a simple differential evo- scheme. In fact, a number of DE schemes have been proposed lution scheme (i.e., DE/BEST/1) to produce offspring solutions. However, the performance of MOEA/D is very sensitive to the for single objective optimization [9]. It is quite interesting parameters in differential evolution schemes, such as scaling to investigate the combination of MOEA/D with these DE factor. In this paper, we further investigate the influence of schemes. various DE schemes in MOEA/D. Our experimental results show The remainder of this paper is organized as follows. Sec- that the performance of MOEA/D relies on the choice of mating tion II introduces some basic definitions in multi-objective parents in DE schemes. optimization. In the following section, a brief