Investigating the Effects of Initial Population Quality on Evolutionary Algorithms

カテゴリ: 部門大会

論文No: TC4-1-6

グループ名: 【C】2023年電気学会電子・情報・システム部門大会

発行日: 2023/08/23

タイトル(英語): Investigating the Effects of Initial Population Quality on Evolutionary Algorithms

著者名: Zheng Jiaming（室蘭工業大学）,渡邉 真也（室蘭工業大学）,Han Jiayi（室蘭工業大学）

著者名(英語): Jiamiing Zheng (Muroran Institute of Technology),Shinya Watanabe (Muroran Institute of Technology),Jiayi Han (Muroran Institute of Technology)

キーワード: Evolutionary algorithm|Initial population|Chaos entropy

要約(日本語): Evolutionary algorithms have been widely used in optimization problems, where the quality of the initial population has a significant impact on the algorithm's performance. This study aims to evaluate the impact of initial population quality on evolutionary algorithms and proposes an improved initialization method using Latin hypercube sampling.
The proposed method generates multiple candidate populations and selects the best population based on complex indexes such as variance, the entropy of chaos, etc. The experimental results, using the differential evolution algorithm, demonstrate that the proposed initialization method outperforms the normal initialization approach on several test functions. The method also shows better effectiveness and convergence of the search.