一般化ランダム・トンネリング・アルゴリズムによる大域的最適化 : 第5報,RBFネットワークを利用した近似最適化(機械力学,計測,自動制御)

概要

論文の詳細を見る
In practical applications, it is important to reduce the function evaluations in the simulation, and obtain the approximate optimum with high accuracy. To achieve these objectives, the integrative optimization system using the RBF Network (RBFN) and the Generalized Random Tunneling Algorithm (CRTA) is proposed in this paper. This system consists of three parts. (1) Construction of the response surface, (2) Optimization by the GRTA, and (3) Adding the sampling points. The RBFN is used to construct the response surface. The radius on RBFN, which affects the accuracy of response surface, is an important parameter. Firstly new equation for the radius is proposed, based on the examination of existing equation. Secondly a simple sampling strategy to obtain an optimum with high accuracy is also proposed. In general, the objective function and the constraints are approximated, separately. However, the optimum of response surface will often violate the constraints. To avoid such situations, the augmented objective function is utilized in this paper. Then the proposed sampling strategy is applied. Through typical benchmark problems, the validity and effectiveness are examined.
2007-05-25