Difficulties of Evolutionary Many-Objective Optimization

DOI IR Web Site Open Access

Bibliographic Information

Other Title
  • 多数目的最適化における進化的探索の問題点
  • タスウ モクテキ サイテキカ ニオケル シンカテキ タンサク ノ モンダイテン

Search this article

Abstract

Well-known Evolutionary Multi-objective Optimization (EMO) algorithms, such as NSGA-II and SPEA2, show rapid degradation of accuracy with increasing number of objectives. To solve this problem, EMO algorithms have been modified by strengthening selection pressure, limitation of search area in the objective space, and use of indicator functions, etc. Here, we describe the difficulties of the search in many-objective space by examining the search of some modified EMO algorithms. The difficulties can be divided into two classes. The first is the difficulty of convergence toward the Pareto-optimal front, which was confirmed to be due to weak selection pressure and disproportion between the extent of search area and the number of solutions. The second is the difficulty of diversity maintenance; it was confirmed that the solutions lost their diversity even if they converged toward the Pareto-optimal front by strengthening the selection pressure. For these difficulties, we examined the search of a preference-based algorithm as an example of a strategy limiting the search area. We demonstrated a trade-off relation between accuracy and diversity through computational experiments.

Journal

Related Projects

See more

Details 詳細情報について

Report a problem

Back to top