基于分解技术的动态多目标引力搜索算法  

Dynamic multi-objective gravitational searching algorithm based on decomposition

作　　者：刁鹏飞 毕晓君[2] 王艳娇[3] DIAO Pengfei1,2,BI Xiaojun2,WANG Yanjiao3(1. College of Engineering and Technology, Northeast Forestry University, Harbin 150000, China; 2. College of Information and Communication Engineering, Harbin Engineering University, Harbin 150001, China; 3. College of Information Engineering, Northeast Dianli University, Jilin 132012, Chin)

机构地区：[1]东北林业大学工程技术学院,哈尔滨150000 [2]哈尔滨工程大学信息与通信工程学院,哈尔滨150001 [3]东北电力大学信息工程学院,吉林132012

出　　处：《系统工程理论与实践》2018年第5期1300-1309,共10页Systems Engineering-Theory & Practice

基　　金：国际科技合作专项(KY10800150002);国家自然科学基金(61501107)~~

摘　　要：为有效求解动态多目标问题,提出一种基于分解技术的动态多目标引力搜索算法.首先为在环境变化前,得到解集分布性和收敛性都较好的非支配解集,采用基于分解技术的静态多目标引力搜索算法求解环境变化前的静态多目标问题;当环境变化后,根据相邻子种群最优解的相似性与同一权重向量对应子种群最优解的相似性,提出一种新的对最优解的预测模型,以缩小环境变化后各子问题的搜索空间,提高算法的求解效率.最后与目前较先进的静态多目标算法和预测策略在四个测试问题上进行比较,实验结果表明,当待优化问题随时间变化时,本文方法能够取得收敛精度更高、解集分布性更好的最优解集.In order to solve dynamic multi-objective optimization problems, a new dynamic multi-objective gravitational searching algorithm which is based on decomposition technique is proposed in this paper.Firstly an environmental monitoring strategy based on the changes of each objective function optimal solution is adapted to monitor the environment. If the environment doesn＇t change, we will use the static multi-objective gravitational searching algorithm to solve the problem. If the environment changes, we will use a hybrid prediction model response to changes in the environment. The hybrid prediction model is based on the similarity of the optimal solution of the adjacent sub population and the optimal solution of the same weight vector corresponding to the sub population. Finally, compared with the advanced static multi-objective algorithm and the forecasting method are compared on four test problems. Experimental results suggest that the proposed algorithm has obvious effectiveness and advantages which shows great capability of evolving convergent, diverse and uniformly distributed Pareto fronts.

关 键 词：动态多目标 预测模型 基于分解技术的多目标算法 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]