Hopfield型ニューラルネットワークを用いた薬品添加量の最適化 (<小特集>ファジィ/ニューラルネットのダイナミックスと制御への応用)

概要

論文の詳細を見る
The optical filter is composed of transparent plastics and plural chemicals which have various transmittances. The determination of the amount of each chemical to be used is important in obtaining the desired transmittance. Previously, an expert drew up an initial list of the required amounts of chemicals based on intuition and experience, and then produced a final list after repetition of trial manufacture, assessment and modification. In this paper, we propose a mathematical model of transmittance using Beer-Lambert's approximate equation and the Hopfield Neural Network, which was modified to apply to real values. We defined the energy function as the square of the difference between desired and calculated transmittances. Moreover, we avoided the influences of the local minima by adding an annealing method to the network, and achieved exceIlent results.
社団法人日本機械学会の論文
1992-06-25