A grid-based adaptive multi-objective differential evolution algorithm

详细信息    查看全文

• 作者：Jixiang Cheng^a ; ^{chengjixiang0106@126.com" class="auth_mail" title="E-mail the corresponding author} ; Gary G. Yen ; ^b ; ^{gyen@okstate.edu" class="auth_mail" title="E-mail the corresponding author} ; Gexiang Zhang^c ; ^{zhgxdylan@126.com" class="auth_mail" title="E-mail the corresponding author}
• 关键词：Multi-objective optimization ; Multi-objective differential evolution ; Grid ; Parameter adaptation
• 刊名：Information Sciences
• 出版年：2016
• 出版时间：1 November 2016
• 年：2016
• 卷：367-368
• 期：Complete
• 页码：890-908
• 全文大小：941 K

文摘

Differential evolution is an excellent optimizer for single objective optimization problems. To extend its use for multi-objective optimization problems with promising performance, this paper proposes a grid-based adaptive multi-objective differential evolution algorithm. The main feature of the proposed algorithm is its dynamical adjustment of convergence and diversity by exploiting the feedback information during the evolutionary process. In the algorithm, the objective space is divided into grids according to the nondominated solutions in the population. Based on the grid, three indexes including grid fitness, grid density, and grid-objective-wise standard deviation are defined to measure individual rank, individual density, and population search status quo, respectively. Afterwards, three main components of the algorithm, i.e., parents selection, parameter control, and population update, are implemented based on grid index values. To validate algorithm performance, comprehensive experiments are carried out on thirty-one benchmark problems. The results show that the proposed algorithm outperforms nine state-of-the-art competitors in terms of three performance metrics. Also, the effectiveness of three components and the sensitivity of two design parameters in the algorithm are empirically quantified.