多目标优化的进化环境模型及实现

An Evolutionary Environment Model of Multiobjective Optimization and Its Realization

传统多目标进化算法主要是模仿生物自身的进化过程,没有考虑环境对进化的作用,缺乏能动的、指导性的搜索.提出一种基于进化环境的多目标进化模型,利用进化环境记录群体进化过程中产生的知识信息,并反过来指导群体搜索,实现环境与群体的共同进化.此外,给出基于进化环境的多目标进化模型的一种算法实现,利用环境域和单元域表示进化环境,设置了一组环境规则,从而实现进化环境对进化群体的约束、促进和导向作用.通过与5个代表性经典多目标进化算法,对12个具有不同特征和不同求解难度的测试函数,在Generational Distance、Hypervolume和Inverted Generational Distance三项性能指标上进行比较实验,验证了文中所提出的算法具有良好的收敛性和综合性能.

Traditional multi-objective evolutionary algorithms （MOEAs） usually imitate the biological evolution of their own, without considering the role of environment in evolution, thus lacking of active and instructional search. In this paper, a multi-objective evolutionary model based on evolutionary environment is proposed （EEMOEM）. This model makes use of the evolu- tionary environment to record the knowledge and information generated in evolution, and in turn the knowledge and information guide the search, which makes the simultaneous evolution of the environment and population. In addition, an algorithm named EEMOEA implementation of EEMOEM is introduced in this paper. The environment area and the unit area are employed to express the evolutionary environment. Also, a group of environment rules is set to realize the function of constraint, promotion and leading for the evolution population. The results of an extensive comparative study on three metrics of Generational Distance, Hypervolume and Inverted Generational Distance, in solving twelve distinct test problems of varied difficulty, show that the EEMOEA outperforms the other five states-of-the-art MOEAs in terms of convergence and comprehensive performance.