基于联赛评价和知识提取的交互式遗传算法

Interactive genetic algorithms with tournament evaluation and evolutionary knowledge extraction

交互式遗传算法基于用户评价获得进化个体适应值,是解决性能指标难以(无法)显式描述的复杂优化问题的有效方法.为有效解决交互式遗传算法的用户疲劳问题,提高算法的整体性能,提出了一种基于有向图提取进化知识的高性能交互式遗传算法.首先,基于进化种群构造联赛评价对,并确定进化个体的占优关系;然后,建立有向图,利用有向图节点的出度和入度计算进化个体适应值,并确定优势个体和建筑块;最后,基于建筑块生成新个体,参与种群后续进化.在服装进化设计系统中的应用结果表明,本文算法可有效减轻用户疲劳,提高算法的搜索能力.

Interactive genetic algorithms, whose individual＇s fitness is assigned by a user, are effective methods to solve a complicated optimization problem with its indices being hard or even impossible to be explicitly described. In order to alleviate user fatigue and improve the algorithm＇ s performance, we presented an emeient interactive genetic algorithm with extracting evolution- ary knowledge based on a directed graph. First, some pairs of tournament evaluated evolutionary individuals were constructed according to the evolutionary population, and the dominance relations of these individuals were obtained. Then a directed graph was built, an individual＇ s fitness was calculated by using the in-degree and out-degree of its corresponding vertex of the directed graph, and some superior individuals as well as building blocks were obtained. Finally, some new individuals were generated based on these building blocks and involved in the subsequent evolutions. The proposed algorithm was applied in a fashion evolu- tionary design system and the results showed the algorithm＇ s advantage in alleviating user fatigue and improving search performance.