ニューラルネットワークによるアルコール発酵プロセスの菌体濃度の同定

概要

論文の詳細を見る
An easy method for identification of the dynamic characteristics in fermentation processes was investigated by a neural network. Using the case of ethanol fermentation, simulations of cell density were attempted. The neural simulator was constructed according to the following procedures;1.The cell(X), ethanol(P), and glucose(S)time courses were approximated by polynomial equations, and these equations were standardized by dividing by the maximum values. 2.The space formed by time as the X-axis and the standardized values as the Y-axis was divided into many local areas. 3.A vector was constructed by deciding the position where a standardized value belonged, relative to a time band as a set of local areas.The vector consisted of some local area sets (a vector column);local areas with a standardized value were element 1 and several areas without standardized values were element 0.4.Three columns of vectors, X, P, and S, at a time band (t) were inputed to the first layer of "back-propagation" as a neural netword(NN). The proposed NN had three layers, the first layer was an input layer, the second was a hidden layer, and the third was the output layer.Here, the output values and teaching values in the porposed NN were set of local areas(vector column for X) as the next time band, t+1. After leaning, the time courses of X at various initial glucose conditions were satisfactorily simulated.
社団法人日本生物工学会の論文
1991-11-25