光谱变换技术结合连续投影算法在模型简化中的应用.doc

光谱变换技术结合连续投影算法在模型简化中的应用 专业：测控技术与仪器 学号：20070310110104 学生姓名：曾卫琴 指导老师：郝勇 摘要 小波变换（Wavelet Transformer，WT）技术结合连续投影算法（Successive Projection Algorithm，SPA）用于近红外光谱（Near Infrared Spectroscopy，Vis/NIR）和拉曼光谱（Raman Spectroscopy，RS）有效建模波长的选择，并采用选择后的变量建立偏最小二乘回归（Partial least squares Regression，PLSR）模型。分别采用近红外透射光谱和RS技术对葡萄酒酒精度和药片活性成分含量进行定量分析。将107个葡萄酒样品的近红外光谱按照近似2:1的比例进行校正集和预测集样本的选择，因此，71个样本作为校正集，36个样本作为预测集，将WT-SPA-PLSR方法用于葡萄酒酒精度分析，结果表明，采用WT-SPA-PLSR方法后，酒精度模型的预测均方根误差（root mean square error of prediction，RMSEP）由12.594减小为12.548，预测相关系数（Correlation coefficient，R）没有变化仍为0.951，AIC值由5040.199减小为152.694，建模变量由69个变为67个；将120个药片的RS按照2:1的比例进行样品集划分，80个样本作为校正集，40个样本作为预测集，将WT-SPA-PLSR方法用于药片活性成分含量分析，结果表明，采用WT-SPA-PLSR方法后，药片活性成分含量的RMSEP由7.561变为7.689，R没有变化仍为0.929，AIC值由7406.874减小为201.484，建模变量都是78个。 因此，采用WT-SPA-PLSR方法不仅可以提高Vis/NIR分析模型的精度，而且可以有效简化模型；采用WT-SPA-PLSR方法虽然没有提高RS分析模型的精度，但是误差相差不大，更为重要的是，采用WT-SPA-PLSR方法可以有效简化RS分析模型。 关键词：小波变换；连续投影算法；近红外光谱；拉曼光谱；偏最小二乘法 By using Spectrum change technology combined with Successive Projection Algorithm to simplified model Abstract Wavelet Transformer(WT) combined with Successive Projection Algorithm (SPA) were used to select effective wavelengths from Near Infrared Spectroscopy (Vis/NIR) and Raman Spectroscopy(RS).The selected effective wavelengths were as inputs of Partial least squares Regression (PLSR) model. Using Vis/NIR technique to analyze the alcohol content in wine, and using RM technique to analyze the active substance contents of pharmaceutical tables. Separate Vis/NIR of wine samples into 71 samples for calibration and 36 samples for prediction according to the proportion of approximate two for one. Using the method of WT-SPA-PLSR to analyze the alcohol content in wine. Results show that after adopting the method of WT-SPA-PLSR, root mean square error of prediction (RMSEP) of the model of wine is from 12.594 reduced to 12.548, Correlation coefficient(R) remain unchanged, is still 0.951,value of AIC is from 5