利用近红外光谱分析技术对葡萄酒发酵液中总酸含量和pH值进行定量分析,并通过偏最小二乘法(PLS)建立定量分析模型,同时采用组合间隔偏最小二乘法(SiPLS)、移动窗口偏最小二乘法(MWPLS)并结合遗传算法(GA)对整个谱区进行光谱特征波长变量筛选.以决定系数(R2)、校正标准偏差(RMSEC)、预测标准偏差(RMSEP)、相对分析误差(RPD)以及最佳主因子数作为模型质量的评价指标.其中SiPLS-GA对模型的优化效果最佳,总酸含量和pH值优化模型的R2分别达到0.978 5和0.983 8,RMSEC为0.074 5和0.065 4,RMSEP为0.076 3和0.062 7,RPD为4.85和5.58.结果表明:进行变量筛选可提高模型准确度和稳定性,降低解析难度,达到优化模型的作用.

The total acid content and pH of wine fermented liquid were quantitatively analyzed using Near Infrared Spectroscopy,and the ment of the quantitative analysis model was established by partial least squares method,while the variables of spectral characteristic wavelength in the entire region of the spectrum were screened through SiPLS,MWPLS,and GA.R2,RMSEC,RMSEP,RPD,and the best main factors were used as index forevaluation of model quality.The results showed that variable selection could be optimized to obtain role model and improve its recognition accuracy and stability,thus reduce the analytical difficulty.The optimization effect of SiPLS-GA was the best.R2 of optimization mode for total acid content and pH value were 0.978 5 and 0.983 8,RMSEC were 0.074 5 and 0.065 4,RMSEP were 0.076 3 and 0.062 7,and RPD were 4.85 and 5.58.