A diversity introduction strategy based on change intensity for evolutionary dynamic multiobjective optimization

作     者：Liu, Ruochen Peng, Luyao Liu, Jiangdi Liu, Jing 

作者机构：Xidian Univ Lab Intelligent Percept & Image Understanding Minist Educ Xian 710071 Peoples R China 

出 版 物：《SOFT COMPUTING》 (Soft Comput.)

年 卷 期：2020年第24卷第17期

页      面：12789-12799页

核心收录：

学科分类：08[工学] 0812[工学-计算机科学与技术（可授工学、理学学位）] 

基　　金：National Natural Science Foundation of China [61876141, 61373111] Provincial Natural Science Foundation of Shanxi of China [2019JZ-26] 

主　　题：Dynamic multiobjective optimization Diversity introduction strategy Inverse model Evolutionary algorithm 

摘      要：Many real-world problems can be modeled as dynamic multiobjective optimization ones with several competing objectives, which requires an optimization algorithm to track the movement of Pareto front over time. This paper proposes a novel dynamic diversity introduction strategy based on change intensity to improve the performance of dynamic multiobjective optimization based on evolutionary algorithm (DMOEA). In this proposed method, the information generated during evolution is recorded in preparation for evaluating the change intensity. Then, by comparing the evaluated intensity with the inherent intensity, the introduction ratio can be determined by that greater one. Two diversity introduction strategies are utilized to keep the balance between convergence and diversity when environmental change is detected. An improved inverse modeling is used for those drastic changes, while partial solutions random initialization is utilized for these mild ones. We compare the proposed algorithm with four existing DMOEAs on a variety of test instances. The experimental results show that the proposed algorithm exhibits better search performance.