基于荧光光谱—模拟退火法年份白酒中乙酸浓度预测研究

近年来年份白酒市场中行业规范有所缺失，因此年份白酒的研究具有深远意义和市场价值。白酒中单体物质的浓度会随着白酒的年份改变，检测白酒中单体浓度可用来鉴定白酒质量及其年份。基于国内某品牌年份原桨白酒的三维荧光光谱，对其中乙酸浓度进行了建模研究。对原始光谱进行了小波分解和求导预处理。研究发现小波分解第一层和第二层呈噪声特征，浓度信息主要分布在第三层和第四层信号中。不同激发波长的荧光发射光谱强度分布不同，如何选择合适的激发波长目前还没有一个统一的方法。根据小波分解信号引入有效信号强度概念并获得了合适的建模激发波长（200 nm）；导数光谱的细节特征比原始光谱丰富，光谱求导可以提高光谱的分辨率。研究了乙酸浓度与荧光光谱的相关性，原始荧光光谱与乙酸浓度之间相关性较小，小波分解光谱和导数光谱与浓度的相关性达0.8以上，且呈现出更多离散化的相关性特征峰。因此，小波分解光谱和导数光谱中包含更多乙酸浓度信息且分布比原始光谱更广。基于荧光光谱和模拟退火法研究了乙酸浓度偏最小二乘法（PLS）多元回归模型。研究发现原始光谱的乙酸浓度预测集的均方根误差高达70.03 mg·L-1，模型效果较差；小波分解光谱和导数光谱由于光谱之间多重相关性降低且分辨率提高的特点，模型预测效果更好，其中二阶导数光谱的乙酸浓度预测集的均方根误差和相关系数分别为20.32 mg·L-1和0.999 8，建模效果最好。基于1 000次循环执行模拟退火算法建模得到的光谱信息密度曲线显示出二阶导数光谱比原始光谱包含更多的乙酸浓度信息。以乙酸为例，为年份白酒中物质浓度预测提供了一种简易的光学方法，研究方法对研究多组分渐变体系浓度预测具有一定的参考价值。

Prediction of Acetic Acid Concentration in Chinese Liquors Based on Fluorescence Spectrumand Simulated Annealing Algorithm

In recent years, the industry of vintage liquor market is not standardized. It is of deep significance and market value to study year liquor. The concentration of monomer in liquor will change with liquor age, so the detection of monomer concentration in liquor can be used to identify liquor quality and age. In this paper, based on the three-dimensional fluorescence spectrum of a certain domestic puree liquor brand, the concentration prediction model of acetic acid is studied. The main contents and innovations are as follows: Firstly, wavelet decomposition and derivative preprocessing are performed on the original spectrum. It is found that the first layer and the second layer of the wavelet mainly present the characteristics of noise, the concentration information is mainly distributed in the third and fourth layer signals. The intensity distribution of fluorescence emission spectra with different excitation wavelengths is different. At present, there is no unified method to select the appropriate excitation wavelength. According to wavelet decomposition signal, this article introduced effective signal strength and obtained the proper modeling excitation wavelength (200 nm). The derivative spectrum has more detailed features than the original spectrum, which can improve the spectral resolution. Secondly, the correlation between acetic acid concentration and fluorescence spectrum was studied. In general, the correlation between the original fluorescence spectrum and the concentration of acetic acid is not high. The correlation between the wavelet decomposition spectrum and derivative spectrum and the concentration is more than 0.8 and shows more discrete correlation peaks. Therefore, the wavelet decomposition spectrum and derivative spectrum contain more information about the acetic acid concentration, which has a wider distribution than the original spectrum’s. Finally, the partial least squares (PLS) multiple regression model of acetic acid concentration was studied based on fluorescence spectra and simulated annealing. The results show that the root means square error of the prediction set of acetic acid concentration in the original spectrum is as high as 70.03 mg·L-1, so its model’s effect is poor. Wavelet decomposition spectrum and derivative spectrum have better prediction effect because the multiple correlations between the spectra is reduced, and the resolution is improved. The second derivative spectral modeling is the best. The root mean square error of the prediction set is 20.32 mg·L-1, and the correlation coefficient is 0.9998. The spectral information density curve based on 1000 simulated annealing algorithms shows that the second derivative spectrum contains more acetic acid concentration information than the original spectrum. This study provides a simple optical method for predicting the concentration of substances in the year liquor. The research methods have a certain reference value for studying the concentration prediction of multi-component gradual change system.