紀要論文

分解ベース進化的多目的最適化アルゴリズムの並列化に関する検討

Parallelization of decomposition-based evolutionary multi-objective optimization algorithms

jpn

24321192

https://doi.org/10.15002/00025261

法政大学大学院情報科学研究科

法政大学大学院紀要. 情報科学研究科編

17

1

6

2022-03-24

Version of Record

This paper focus on parallelization of Multi-objective Evolutionary Algorithm Based on Decomposition (MOEA/D), and addresses the decrease of precision and the increase of execution time in Parallel MOEA/D. This paper proposes several grouping methods aim to parallelize MOEA/D while solve the problems such as the sparse boundary and Pareto front dislocation in Parallel MOEA/D, thereby improving the accuracy of the Parallel MOEA/D. The proposed method is also based on the idea of setting overlap zones, but using grouping method that uses inter-individual distances, etc., as grouping methods to reduce additional calculations, instead of setting additional overlap zones. We Implemented and evaluated the grouping methods based on K-means, Mean Shift, Random and Improved Random (Smoother-random). Based on the testing result, the parallel MOEA/D achieved a significant acceleration when the individual size reached a certain size. In addition, these clustering methods also obtain a better accuracy performance in massively parallel environments compared with the grouping method based on neighbor weight vector.

Article

AA12746425