人工神经元网络辅助丙烷氨氧化催化剂设计

ARTIFICIAL NEURAL NETWORK AIDED DESIGN OF CATALYSTS FOR PROPANE AMMOXIDATION

报道了利用人工神经元网络辅助丙烷氨氧化制丙烯腈催化剂设计的研究进展结果．通过筛选得到了可以拟合该反应体系的神经元网络模型，从而使丙烷转化率和丙烯腈选择性可以表示为催化剂组成的函数，利用培训后的神经元网络模型所得到的权值文件，结合作者自编的优化程序，该网络可用于预测最优的催化剂组成．实验得到丙烯腈收率和选择性可分别达到４３．０％和５６．２４％．

An artificial neural network(BP network) is applied to the design of catalyst for ammoxidation of propane to acrylonitrile. The conversion of propane ( X (C 3)) and the selectivity for acrylonitrile ( S (ACN)) can be calculated as functions of the catalyst components by training with the BP network. Thus, a weighted data file containing the weights of every unit in the network can be got. Taking the conversion of propane and the selectivity for acrylonitrile as the two sub objectives, a model for this reaction can be given as Max( y 1= X (C 3)) (0≤ y 1≤1 0) Max( y 2= S (ACN)) (0≤ y 2≤1 0) =F(, ) Using this model, the yield of acrylonitrile of 43 0%, which is higher than those reported in the literatures, has been obtained.