Reconsideration of Adaptive Range Genetic Algorithms
スポンサーリンク
概要
- 論文の詳細を見る
Application of genetic algorithms to mixed design variable optimization has received wide recognition. However, there are some problems in expressing continuous number in GAs, and according to that, applications to mixed variables has not been successful as it recognized. We have developed Adaptive Range Genetic Algorithms and overcome most of the difficulties in expressing continuous and discrete numbers in Genetic Algorithms and obtain the best results in some simple bench mark problems. Key success lies in the adaptation of searching range according to the situation of generation and existence of gene. And this conclusion is different from what they said in Evolution Algorithms (EAs). In EAs, they use continuous number directly to chromosome and made up some rules in crossing over. They ignore the existence of gene to treat design variables. If their keys in success lie in the expression of continuous numbers, we thought existence of gene in ARange GAs will influence badly in its convergence. From this stand point view, we will examine the necessity of gene in GAs and compare the results that are obtained from one of the EAs method called BLX<method. From these comparisons, even though we do not use the information of population in the generation, we have obtained better convergence, because gene will give some deterministic value in each generation and combination of parents. In that sense, BLX<method is something like random search and it only happened to find good solution. Thus, we can say that gene will introduce some sorts of deterministic characteristics in genetic search and it will distinguish GAs from just random search technique. From this conclusion we intend to say the needs of gene in GAs.
- 一般社団法人日本機械学会の論文
- 2006-12-12