【摘要】 对葡萄酒酒精度偏最小二乘(Partial least squares,PLS)回归模型进行优化研究。使用近红外光谱仪采集葡萄酒样本的光谱数据,用于建立酒精度定量模型,实现在线快速检测。通过蒙特卡罗无信息变量消除(Monte Carlo uninformative variable elimination,MC-UVE)和遗传算法(Genetic algorithm,GA)进行变量选择,基于被选择的变量分别进行PLS和因子分析(Factor analysis,FA),建立回归模型。结果表明,MC-UVE-GA-FAR模型预测集相关系数(R2)为0.946,预测均方根误差(Root mean square error of prediction,RMSEP)为0.215,效果优于MC-UVE-GA-PLS模型。与基于全范围光谱所建PLS回归模型相比,模型效果有所提升,而且模型所选变量个数仅为6,极大地简化了模型。MC-UVE和GA算法与FA分析结合可以实现模型的优化。更多还原

【Abstract】 The optimization of the PLS regression model of wine alcohol content was studied. The near-infrared spectroscopy was used to collect the spectral data of the wine samples and the data were used to establish the quantitative model of alcohol to achieve rapid on-line detection. PLS regression model and FA model were established based on the selected variables,chosen by MC-UVE and GA. The results show that the MC-UVE-GA-FAR model,which yielded R2 of 0. 946 and RMSEP of 0. 215,is superior to the MV-UVE-GA-PLS model. In comparison of the performance of the full-spectra PLS regression model,the model based on the selected wave numbers is much better,and 6 variables in total are selected,which greatly simplifies the model. The study indicates the MC-UVE,GA and FA can optimize the model.更多还原