2216 DFP法を用いたニューラルネットワークによる圧電複合円板の変位制御 : 学習の安定化について(S14-3 材料の傾斜機能性と圧電特性への取り組み,S14 先進材料の材料特性における弾性数理解析の新たな取り組み)

概要

論文の詳細を見る
This paper discusses an effect of stabilization of learning process for DFP-based hierarchical neural network on a result of approximate optimization for controlling deformation of a piezoelectric composite disk. Though the DFP formula enables to enhance a learning speed of the neural network, a learning result of the DFP-based neural network is heavily dependent on a set of initial value for weighing coefficients. This will cause a fatal problem in the use of approximate optimization. The proposed stabilization algorithm is applied to the neural network-based approximate optimization technique. From the numerical result, it is shown that the proposed stabilization algorithm improves accuracy of an approximated optimum result using the DFP-based neural network.
2007-09-07