A semi-supervised transfer learning based dynamic multi-objective evolutionary algorithm for dynamic multi-objective optimization
LIU Kan-rong
LI Yan
TAN Shu-bin
LIU Yuan-chao
LIU Jian-chang
Abstract：In dynamic multi-objective optimization problems,multiple conflicting objectives vary over time,which will lead to the change of Pareto optimal front.In most dynamic multi-objective optimization problems,there exists the correlation between different environments,in other words,the algorithm can use the information from the previous environments to track the dynamically changing Pareto optimal front timely.In order to make full use of environmental information to track the dynamically changing Pareto optimal front,a semi-supervised transfer learning based dynamic multi-objective evolutionary algorithm(SSTL-DMOEA)is proposed in this paper.SSTL-DMOEA consists of two core components.First,it introduces a semi-supervised transfer mechanism to transfer favorable information from the historical environments to the current environment.Thus,the algorithm can generate a good initial population for improving the search efficiency in the current environment.Secondly,a series of sample points are created in the target domain by using the center point of the Pareto optimal solution set from the historical environments and the evolutionary information of the new environment.These points can help the algorithm build a more accurate prediction model.Compared with the four state-of-the-art dynamic multi-objective optimization algorithms,SSTL-DMOEA is competitive in dealing with dynamic multi-objective optimization problems.
Keywords：dynamic multi-objective optimizationevolutionary algorithmknowledge transfer
Publication Date：2025-01-27
Online Publishing Date：2025-08-15(First online date of this platform, not the publication date of the document)
Pages：12( 1-12 )