A dynamic multiobjective evolutionary algorithm based on a dynamic evolutionary environment model
A dynamic multiobjective evolutionary algorithm based on a dynamic evolutionary environment model
复制标题
基于动态进化环境模型的动态多目标进化算法
DOI：
10.1016/j.swevo.2018.03.010
复制
发表时间：
2019-02-01
影响因子：
10
通讯作者：
Pei, Tingrui
中科院分区：
文献类型：
--
作者：
Zou, Juan;Li, Qingya;Pei, Tingrui
传统的动态多目标进化算法通常模拟自然界的进化，通过不同的策略保持种群的多样性，使种群在环境变化后有效地跟踪到帕累托最优解集。然而，这些算法忽略了动态环境在进化中的作用，导致缺乏主动引导搜索。提出了一种基于动态进化环境模型的动态多目标进化算法(DIE-DMOEA)。该算法在环境不变的情况下，利用进化环境记录进化过程中产生的知识和信息，进而指导搜索。当检测到变化时，该算法通过建立动态进化环境模型来帮助种群适应新的环境，通过引导式方法增强种群的多样性，使环境和种群同步进化。此外，本文还介绍了动态进化环境模型的算法实现。用环境面积和单位面积来表示演化环境。在此基础上，提出了约束策略、促进策略和引导策略。与其他三种最新策略相比，对于设计变量之间具有线性或非线性相关性的一系列测试问题，该算法对于处理动态多目标问题是有效的。
Traditional dynamic multiobjective evolutionary algorithms usually imitate the evolution of nature, maintaining diversity of population through different strategies and making the population track the Pareto optimal solution set efficiently after the environmental change. However, these algorithms neglect the role of the dynamic environment in evolution, leading to the lacking of active guieded search. In this paper, a dynamic multiobjective evolutionary algorithm based on a dynamic evolutionary environment model is proposed (DEE-DMOEA). When the environment has not changed, this algorithm makes use of the evolutionary environment to record the knowledge and information generated in evolution, and in turn, the knowledge and information guide the search. When a change is detected, the algorithm helps the population adapt to the new environment through building a dynamic evolutionary environment model, which enhances the diversity of the population by the guided method, and makes the environment and population evolve simultaneously. In addition, an implementation of the algorithm about the dynamic evolutionary environment model is introduced in this paper. The environment area and the unit area are employed to express the evolutionary environment. Furthermore, the strategies of constraint, facilitation and guidance for the evolution are proposed. Compared with three other state-of-the-art strategies on a series of test problems with linear or nonlinear correlation between design variables, the algorithm has shown its effectiveness for dealing with the dynamic multiobjective problems.