超立方体交叉手法を用いた Differential Evolution の提案

概要

論文の詳細を見る
Differential Evolution (DE) is a global optimizer for solving real parameter optimization. DE is effective for many problems and has a few control parameters to be set. The child vectors can exist in a range wider than the range of the parent vectors, as the parent vectors are generated by differential operations. The search range becomes narrower as the range of population of individuals becomes narrower. So DE can search globally at an earlier stage of the search and can search locally at a stage in which the range of population of individuals is narrow. In this paper, we propose the Hypercube Crossover Method (HCM), which does not depend on U-valley direction. The proposed method sets positions of the parents to the hypercube diagonal position, and a child is generated on the vertex position of the hypercube. This will enable the DE algorithm to search in the perpendicular direction to the U-valley. The performance of the proposed method is evaluated by numerical experiments, and the experimental results are considered from the viewpoint of the variety of the individuals population. Finally, future work is discussed.
2013-03-30