摘要
基于多模型(模型融合)建模的思想,开发了两种新的叠加多元校正分析算法:叠加PCR(PLS)多元校正分析和叠加移动窗口PCR(PLS)多元校正分析。与一般的多模型建模方法不同的是其通过赋予光谱数据中的不同部分不同权重叠加子多元校正模型。因此,其可以通过权重调节或选择变量。在消除光谱数据中常见的冗余信息的同时,避免信息遗漏的缺点,并最终提高模型的稳健性,简化了模型。对于这两个新的算法,尽管其具体步骤不同,但仍取得了相似的预测结果。本文通过两套近红外光谱文献数据计算验证了这两个新方法的优越性。
Two novel algorithms that employed the idea of ensemble models（ stacked generalization or stacked regression） ,stacked multivariate calibration（ PCR and PLS） and stacked moving-window multivariate calibration（ PCR and PLS） ,were reported. The proposed algorithms established parallel,conventional PCR or PLS models based on all intervals of a set of spectra to take advantage of the information from the whole spectrum. Unlike traditional methods,they stack or incorporate these parallel models in a way to emphasize intervals （ regions） highly related to the target property. These two stacking algorithms generate more parsimonious regression models with better predictive power than that of conventional PCR and PLS,and perform best when the spectral information is neither isolated to a single,small region,nor spread uniformly over the spectral data. The predictive performance of these two new algorithms is similar. Thus,two real NIR spectra were used here to show the improvement in predictive performance from these two new algorithms.
出处
《分析化学》
SCIE
EI
CAS
CSCD
北大核心
2010年第3期367-371,共5页
Chinese Journal of Analytical Chemistry
基金
国家留学基金委(No.留金字[2007]3020)资助项目
关键词
多模型
叠加
移动窗口
多元校正分析
Ensemble models （ model fusion）
Multivariate calibration
Stacked multivariate calibration
Stacked moving-window multivariate calibration
