1102 大域的近似のための選択的3次多項式による応答曲面法(OS22 最適化計算法)

概要

論文の詳細を見る
This paper discusses global approximation in the context of multidisciplinary design optimization and proposes a response surface method with third-order polynomial with optimal selection of interaction terms. After the necessity and role of global approximation are reviewed, the indices of approximation model, computational cost, width of design region and fidelity of approximation model, are described. The discussion brings a potential of high-order polynomial approximation among their tradeoffs. Following them, a method for establishing third-order polynomial regression model is formulated. The method consists of design of experiments, Chebyshev's orthogonal polynomial, optimal selection of interactive terms with a genetic algorithm under the criterion of 95% confidence limit. Its application to some test functions demonstrates the effectiveness of high-order and interaction terms.
2005-08-02