摘要
用模拟数据和实验数据研究了人工神经网络（ＡＮＮ）方法用于解析色谱重叠峰的可能性，以二甲苯异构体及丙酮－异丙醇重叠峰为例对网络结构进行了优化，提出了两种简单易行的采集数据的新方法和模拟非正态峰的数学模型．结果表明，对不易分开的组分，只要训练集及测试集条件一致，大部分实验数据预测结果满意，用这种方法对色谱峰进行定量，结果可靠。
The deconvolution of overlapping peaks by using artificial neural networks was reported in this paper. The separations of acetone～isopropanol and p xylene～o xylene were used as examples. Five points act as input nodes to represent overlapping peaks. The concentrations of acetone in act the system of acetone～isopropanol are 5%～95% and the concentrations of p xylene are 5%～50% in o xylene～p xylene mixture. Simulated data were used to train the network so as to gain the suitable learning Epochs and the size of learning set. With the network structure of 5 4 1 and a certain learning Epoch, the predicted results are satisfied. This is a new way to predict the concentration of components in chromatography.
出处
《兰州大学学报（自然科学版）》
CAS
CSCD
北大核心
1999年第2期99-104,共6页
Journal of Lanzhou University(Natural Sciences)
关键词
气相色谱
神经网络
反向传播模型
重迭峰
色谱
gas chromatography
artificial neural network
back propagation
overlapping peaks
