2段階GA ``Solid EMO'' によるレンズ系設計 Lens System Design by A Two Stage GA ``Solid EMO''

この論文にアクセスする

著者

    • 田中 雅晴 Tanaka Masaharu
    • 東京工業大学 大学院総合理工学研究科 知能システム科学専攻 Department of Intelligent Systems Science, Interdisciplinary Graduate School of Science and Engineering, Tokyo Institute of Technology
    • 秋本 洋平 Akimoto Yohei
    • 東京工業大学 大学院総合理工学研究科 知能システム科学専攻 Department of Intelligent Systems Science, Interdisciplinary Graduate School of Science and Engineering, Tokyo Institute of Technology
    • 佐久間 淳 Sakuma Jun
    • 東京工業大学 大学院総合理工学研究科 知能システム科学専攻 Department of Intelligent Systems Science, Interdisciplinary Graduate School of Science and Engineering, Tokyo Institute of Technology
    • 小野 功 Ono Isao
    • 東京工業大学 大学院総合理工学研究科 知能システム科学専攻 Department of Intelligent Systems Science, Interdisciplinary Graduate School of Science and Engineering, Tokyo Institute of Technology
    • 小林 重信 Kobayashi Shigenobu
    • 東京工業大学 大学院総合理工学研究科 知能システム科学専攻 Department of Intelligent Systems Science, Interdisciplinary Graduate School of Science and Engineering, Tokyo Institute of Technology

抄録

This paper discusses evolutionary multi-objective optimization (EMO) method for lens system design problems that have properties of global and local multimodality, epistasis among parameters and ill-scaledness. Applying NSGA-II-like EMO to them, it faces some difficulties. To solve them, we present a two stage GA called Solid EMO that consists of a repeated ESO (Evolutionary Single-objective Optimization) and an augmented EMO. The repeated ESO searches seeds of Pareto optimal solutions through solving weighted sum minimization problems repeatedly by a real-coded GA using ISM that deals with global multi-modality well. The augmented EMO, that behaves like a kind of local search by k-nearest neighbor limitation in reproduction and crossover with an ability of explorative search, refines and expands the seeds found by the first stage GA. Solid EMO was applied to three and four element lens system design problems. As a result, the proposed method succeeded in finding highly precise solution sets that consist of well-known types, triplet-type and Lee-type lens systems, in the three-element and four-element lens system design problems, respectively.

収録刊行物

  • 人工知能学会論文誌

    人工知能学会論文誌 23(3), 193-204, 2008

    一般社団法人 人工知能学会

各種コード

  • NII論文ID(NAID)
    130000098070
  • 本文言語コード
    JPN
  • データ提供元
    J-STAGE 
ページトップへ