摘要
Kohonen与BP人工神经网络结合用于解析钨和钼的吸收光谱,讨论了Kohonen网络输出层的拓扑结构,并利用确定的结构对钨和钼的重叠光谱进行波长选择,在全光谱中选择最能代表光谱特征的不同类波长。所选波长处的吸光度作为三层BP-ANN网络的输入集,分光光度法同时测定了钨和钼。利用Koho-nen网络选择全谱特征波长,优化了BP-ANN的输入层。与常规的波长选择方法进行比较,分析结果表明,经K-ANN方法进行波长选择后,提高了BP-ANN的预测能力。确立了Kohonen网络作为选择最优波长集的一种工具。
BP artificial neural networks combined with Kohonen networks were used for the simultaneous determination of tungsten and molybdenum by spectrophotometry. First, the topology structure of Kohonen artificial neural networks was studied, and applied to the wavelength selection of the overlapped spectra of tungsten and molybdenum. Then the most informative wavelengths were selected from the full spectra, and absorbance values were used as the optimal input sets of the three-layer BP neural networks. Compared with the routine method of wavelength selection, the results prove that using Kohonen networks to select the most informative wavelengths can optimize the input layer of BP-ANN, and the prediction ability of BP-ANN is improved. So Kohonen networks can be used as a tool for wavelength selection.
出处
《光谱学与光谱分析》
SCIE
EI
CAS
CSCD
北大核心
2006年第12期2319-2322,共4页
Spectroscopy and Spectral Analysis
基金
国家自然科学基金项目(29971029)
莱阳农学院博士启动基金项目(630408)资助
关键词
人工神经网络
化学计量学
分光光度法
钨
钼
Artificial neural network
Chemometries
Spectrophotometry
Tungsten
Molybdenum
