Hiper-Heurística utilizando a técnica Upper Confidence Bound para otimização multiobjetivo baseada em decomposição
Many real optimization problems have been formulated with more than one objective. Recently, several multi-objective evolutionary algorithms were proposed and particularly those based on decomposition, such as the MOEA/D-DRA, have been successfully applied to solve these problems. However, the liter...
| Autor principal: | Prestes, Lucas |
|---|---|
| Formato: | Dissertação |
| Idioma: | Português |
| Publicado em: |
Universidade Tecnológica Federal do Paraná
2018
|
| Assuntos: | |
| Acesso em linha: |
http://repositorio.utfpr.edu.br/jspui/handle/1/3293 |
| Tags: |
Adicionar Tag
Sem tags, seja o primeiro a adicionar uma tag!
|
| Resumo: |
Many real optimization problems have been formulated with more than one objective. Recently, several multi-objective evolutionary algorithms were proposed and particularly those based on decomposition, such as the MOEA/D-DRA, have been successfully applied to solve these problems. However, the literature demonstrates that the performance of this class of algorithms is severely affected by its parameter settings. Automatic parameter configuration methods emerge as an alternative as they have presented good results in the literature. Therefore, based on a Hyper-Heuristic for automatic parameter configuration we aim at improving the performance of an algorithm based on decomposition developed to solve multi-objective problems. The Hyper-heuristic parameter setting is performed in both offline and online modes. The irace algorithm is used at the offline stage to set the parameters used during the algorithm warm-up while the Upper Confidence Bound (UCB) is used at the online stage which occurs after the warm-up. The majority of the works related to the use of UCB to improve the performance of the MOEA/D and its variants adjust only the mutation strategy, whereas the proposed approach adjusts, besides the mutation strategy, several other parameters like crossover rate, mutation factor, probability of choosing the global or local neighborhood and the maximum number of updates in the neighborhood. The experiments are divided into stages to find out an efficient structure. The best variant is then compared with state-of-the-art methods (NSGA-II, IBEA and MOEA / D-DRA). The results indicate that the proposed approach is very competitive with the literature. The performed statistical analysis shows that the proposed approach is equal or superior to the state-of-the-art methods in 38 of 51 instances of benchmarks CEC2009, GLT, LZ, MOP, DTLZ and WFG. |
|---|