実数領域適応型(ARRange)遺伝的アルゴリズムの開発

概要

論文の詳細を見る
In this paper, we propose new type of real genetic algorithms named adaptive real range (ARRange) genetic algorithms (GAs). In conventional real GAs, we simply set a minimum value and a maximum value for each design variable and divide the range into some specific divisions after decoding binary strings to integer. However, in such cases, we need to divide given ranges enormously in order acheive sufficient accuracy. Moreover, initially, we usually do not have any information on the minimum and maximum values. Thus, we have to set them ignoring the accuracy of the real values. In the proposed method, a range of real numbers will move adaptively in each generation by using the mean value and standard deviation of the previous generation. By using ARRange GAs, we do not have to consider the setting of minimum and maximum real value or number of bits for accuracy of real numbers. However, in ARRange GAs, we need four additional GA parameters which influence on the performance of GAs. In particular, two of these parameters greatly influence the convergence. We also present additional options that make the designer not to consider bothering of these presetting. In this study, we demonstrate the proposed method by simple numerical examples and demonstrate its effectiveness and characteristics.
一般社団法人日本機械学会の論文
1997-12-25