构建卷烟感官香气风格特征的PLSR定量预测模型
Construction of quantitative PLSR prediction model of cigarette sensory aroma characteristics by variable selection optimization

采用气相色谱-三重四极杆质谱(GC-MS/MS)方法检测73种烟丝挥发性/半挥发性成分，结合其他12种烟丝/烟气成分检测含量，采用偏最小二乘回归 (PLSR)对80种国内市售成品卷烟建立定量预测模型，并利用变量筛选方法优化模型. 结果表明：模型对果香评吸指标具有最佳的拟合效果R2＞0.910，其所能解释的方差Q2＞0.850；增加对茴香醛质量浓度和降低乙基麦芽酚的质量浓度可以有效提高果香评吸的感官评价得分；模型对于测试集果香评吸指标也具有很好的预测能力，平均绝对误差MAE=0.1112（<0.500)，PLSR预测结果可靠，可作为一种客观预测卷烟感官品质的方法.
GC-MS/MS method was constructed for the determination of 73 kinds of volatile / semi-volatile components of tobacco，combined with the other 12 kinds of tobacco/smoke substances’ content. Thus，quantitative prediction model could be established using partial least squares regression (PLSR) on 80 kinds of finished products sold in domestic，meanwhile the variable selection method could be used for optimization model. The results showed that model had the best fitting results for fruity smoking indicators R2>0.910，whose explanation for the variance Q2>0.850. Increasing anisaldehyde concentration and reducing the concentration of ethyl maltol could improve fruity smoking sensory evaluation score. Model had also a good predictive ability for the fruit smoking indicators of test set，whose mean absolute error MAE=0.1112（<0.500)，PLSR had a reliable prediction result，which could be used as an objective method to predict the sensory quality of cigarette