摘要
提出一种解析分光光度同时测定数据的小波包分析-减法聚类-RBF网络新方法。该方法用小波包分析处理吸收光谱数据,滤除信号中的噪声;采用基于山峰函数的减法聚类算法,按照自适应聚类的结果构成预测各未知样的校正集,实现校准模型的优化;由此,使RBF网络在对未知样预测时能更有效地提取光谱数据中的特征信息,提高预测结果的准确性。把该算法用于模拟汽油中铁、锰合成样预测,计算表明,该方法可以显著降低测定结果的相对误差,预测结果令人满意。
A new method for resolving the spectrophotometric spectrum is proposed. The noise in the visible spectrum was eliminated by wavelet packets analysis and the calibration model was optimized by peak-subtraction clusler. The veracity of the predicled results was hnproved. The proposed method has been applied lo predict the contents of Fe and Mn in simulated gasoline simultaneously by using the wavelet packets analysissubtractive cluster- RBF（radial basis function） networks（WPA - subclust - RBF）. The results showed that the method could reduce the relative error of the resuhs si.gnificantly. The predicted results were satisfactory.
出处
《分析测试学报》
CAS
CSCD
北大核心
2005年第5期30-34,共5页
Journal of Instrumental Analysis
基金
四川省教育厅重点科研项目(2004A107)
关键词
小波包分析
减法聚类
RBF网络
分光光度法
Wavelet packets analysis
Subtractive cluster
Radial basis function networks
Spectrophotometry
