一种基于偏序排名的高效的多目标演化算法  

Efficient Multi-Objective Evolutionary Algorithm Based on Partial Order Ranking

作　　者：刘敏忠[1] 邹秀芬[1] 康立山[1] 

机构地区：[1]武汉大学软件工程国家重点实验室,湖北武汉430072

出　　处：《小型微型计算机系统》2004年第12期2102-2106,共5页Journal of Chinese Computer Systems

基　　金：国家自然科学基金 ( 60 13 3 0 10 ;70 0 710 42 ;40 2 75 0 3 4)资助

摘　　要：利用 Better函数对多目标优化问题 (MOP)建立了一种偏序关系 ,并通过这种偏序关系对种群中的个体进行排名 ,设计了一种高效的多目标演化算法 (EMOEA) ,应用有限 Markov链的有关理论证明了此演化算法的收敛性 ,同时对四个 Benchmark函数进行了数值试验 ,测试结果表明 ,新算法在解集分布的均匀性、多样性、精确性以及快速收敛性均优于已有的 MOEA.将此算法推广用于求解经济领域或工程领域的一些难题 。This paper introduce a new and efficient multi-objective evolutionary algorithms (EMOEA) for solving unconstrained and constrained multi-objective optimal problems (MOPs).The novelty of our algorithm are: (1) defining a new partial-order relation by Better function,(2) the individuals are ranked by this partial-order relation,which is very different from the ranking described in ordinary MOEA,because the constrained conditions are merged into ranking,therefore when we evaluate an individual,we dont need to consider whether the individual is feasible,(3) adopting multi-parents crossover and dynamical mutate to keep the uniformity of search,(4) By using the theory of finite Markov Chain,the convergence properties of our algorithm are proved.We take several benchmark MO optimization problems to test our algorithm.The numerical experiments show that our algorithm is superior to other MOEAs in terms of the precision,the quantity,the distribution uniformity and diversity of solutions and the convergence rate of algorithm.

关 键 词：多目标优化问题 多目标演化算法 偏序关系 收敛性 多样性 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]