一种新的正交多目标最优化遗传算法  

A Novel Orthogonal Multiobjective Genetic Algorithms

作　　者：刘海林[1] 王宇平[2] 刘永清[1] 

机构地区：[1]华南理工大学自动控制工程系 [2]西安电子科技大学数学科学系,西安710071

出　　处：《计算机工程与应用》2002年第11期27-29,131,共4页Computer Engineering and Applications

基　　金：国家自然科学重点基金资助(编号:69934030)

摘　　要：文章首先构造了一个新的适应值函数,它以标准化后的目标函数的加权极大值作为个体的适应值,通过独特选取多组权重向量和正交试验方法,保持了种群的多样性,使得搜索方向在目标空间中均匀指向有效界面。该算法的显著特性是不管有效界面是否凸,都能找到足够多均匀分布的有效解。数值实验也表明了该算法非常有效。A new fitness function is constructed by maximization of the weighted normalized- objectives,where the weight vectors are carefully and reasonably designed via generalized sphere coordinate transformation and orthogonal design.As a result ,the population can keep the diversity,and the search uniformly moves towards Pareto frontier.The most important characterization of the proposed algorithm is that it can always find enough solutions distributed uniformly on Pareto fron tier no mater whether the Pareto frontier is convex or not.The computer simulat ions also show the proposed algorithm is effective.

关 键 词：遗传算法 多目标最优化 极大极小策略 正交试验设计法 

分 类 号：O224[理学—运筹学与控制论] O242.23[理学—数学]