Zobrazit v EDS

Decomposition-based multi-objective evolutionary algorithm with multi strategies for portfolio optimization problems.

Uloženo v:
Podrobná bibliografie
Název: Decomposition-based multi-objective evolutionary algorithm with multi strategies for portfolio optimization problems.
Autoři: Song, Chunlin1 (AUTHOR) 2023410060@sdtbu.edu.cn, Zhang, Bin1 (AUTHOR) 2023410079@sdtbu.edu.cn, Zhao, Gaoyang2 (AUTHOR) zgy17220260@163.com, Song, Yingjie1 (AUTHOR) songyj@sdtbu.edu.cn
Zdroj: Journal of Supercomputing. Feb2026, Vol. 82 Issue 3, p1-36. 36p.
Abstrakt: Decomposed-based multi-objective optimization algorithm (MOEA/D) has demonstrated remarkable optimization capabilities in solving multi-objective optimization problems (MOPs). However, the MOEA/D neglects the correlation between weight vectors and individuals during neighborhood construction, reduces the optimization performance by using single evolutionary operator and fixed control parameters. To solve these problems, an enhanced MOEA/D based on multi-population strategy, the combined differential strategy and parameter adaptive strategy, namely MOEA/D-MAC is proposed to further balance the convergence and diversity. In the MOEA/D-MAC, a new multi-population strategy with neighborhood adaptive reconstruction has been developed. It aims to tackle the loss of population diversity resulting from random matching of initial solutions and unique matching of subproblem solutions. Additionally, a combined differential strategy with bidirectional search has been developed. This strategy incorporates various crossover and mutation strategies tailored to different stages, catering to the exploration and exploitation requirements across various evolutionary periods. Furthermore, Gaussian distribution is employed to perform bidirectional local exploration to further expand the currently known search space. Finally, a parameter adaptive strategy is proposed by dynamically adjusting control parameters during the search process. ZDT and DTLZ benchmark test suites and an investment portfolio optimization problem are selected to prove the effectiveness of the MOEA/D-MAC. The experiment results show that the MOEA/D-MAC achieves better optimization performance by comparing with NSGA-II, MOEA/D-DE, MOEA/D-STM, MOBCA and MOEA/D-FRRMAB. [ABSTRACT FROM AUTHOR]
Databáze: Academic Search Index
Popis
Abstrakt:Decomposed-based multi-objective optimization algorithm (MOEA/D) has demonstrated remarkable optimization capabilities in solving multi-objective optimization problems (MOPs). However, the MOEA/D neglects the correlation between weight vectors and individuals during neighborhood construction, reduces the optimization performance by using single evolutionary operator and fixed control parameters. To solve these problems, an enhanced MOEA/D based on multi-population strategy, the combined differential strategy and parameter adaptive strategy, namely MOEA/D-MAC is proposed to further balance the convergence and diversity. In the MOEA/D-MAC, a new multi-population strategy with neighborhood adaptive reconstruction has been developed. It aims to tackle the loss of population diversity resulting from random matching of initial solutions and unique matching of subproblem solutions. Additionally, a combined differential strategy with bidirectional search has been developed. This strategy incorporates various crossover and mutation strategies tailored to different stages, catering to the exploration and exploitation requirements across various evolutionary periods. Furthermore, Gaussian distribution is employed to perform bidirectional local exploration to further expand the currently known search space. Finally, a parameter adaptive strategy is proposed by dynamically adjusting control parameters during the search process. ZDT and DTLZ benchmark test suites and an investment portfolio optimization problem are selected to prove the effectiveness of the MOEA/D-MAC. The experiment results show that the MOEA/D-MAC achieves better optimization performance by comparing with NSGA-II, MOEA/D-DE, MOEA/D-STM, MOBCA and MOEA/D-FRRMAB. [ABSTRACT FROM AUTHOR]
ISSN:09208542
DOI:10.1007/s11227-026-08302-1