基于动态多策略差分进化模型的MOEA/D算法  

MOEA/D based on dynamic multi-strategy differential evolution model

作　　者：林震[1] 侯杏娜[1] 韦晓虎[1] Lin Zhen Hou Xingna Wei Xiaohu(Dept. of Experiential Practice, Guilin University of Electronic Technology, Guilin Guangxi 541004, China)

机构地区：[1]桂林电子科技大学教学实践部,广西桂林541004

出　　处：《计算机应用研究》2017年第9期2624-2628,共5页Application Research of Computers

基　　金：国家自然科学基金资助项目(61261017);桂林电子科技大学教育教学改革项目(JGB201431;JGB201530;ZJW43030)

摘　　要：在基于分解技术的多目标进化算法的框架中,引入一种动态多策略差分进化模型。该模型在分析不同差分进化策略的特点基础上,选择了三种差分进化策略,并对每种策略分配一子种群。在进化过程中,依据每种策略对邻域更新的贡献度,动态地调整其子种群的大小。对比分析采用不同差分进化算法的性能,结果表明运用多个策略之间相互协同进化,有利于提高算法性能。将新算法同NSGA-Ⅱ与MOEA/D算法在LZ09系列基准函数上进行性能对比,实验结果显示该算法的收敛性和多样性均优于对比算法。将新应用于Ⅰ型梁多目标优化设计问题中,获得的Pareto前沿均匀,且解集域较宽广,对比分析表明了算法的工程实用性。In the framework of multi-objective evolutionary algorithm based on decomposition （MOEA/D）, this paper intro- duced a dynamic multi-strategy differential evolution model （MOEA/D-DMDE）. The model chose three differential evolution strategies and each sub-population was corresponding to a differential evolution strategy based on the analysis of the characteri- stics of different strategies. In order to improve the performance of the algorithm, it adjusted the size of sub-population dynami- cally on the basis of a differential evolution strategy contribution for updated of neighborhood. It adopted each strategy to partic- ipate in coordination during the evolution process. Via the comparative analysis of different schemes of differential strategy, MOEA/D-DMDE also performed well. Comparing with NSGA-1I and MOEA/D on the LZ09 benchmarks, the experimental results indicate that MOEA/D-DMDE has a better performance in terms of convergence and diversity. To validate its perfor- mance on constraint multi-objective optimization problems, the proposed MOEA/D-DMDE is applied for solving the I -Beam. The uniformly distributed Pareto sets obtained by MOEA/D-DMDE show its practicability for engineering problems.

关 键 词：MOEA/D 多目标优化 多策略差分进化 动态子种群 I型梁设计 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程] TP301.6[自动化与计算机技术—控制科学与工程]