Chinese liquors with different years have become the focus of the enterprise development as the high-end products in the industry.But the standards about the years of the product have a greater randomness.Therefore,to establish a technical standard of liquor's year has become desperately needed in order to regulate the industry and the market.Based on three-dimensional fluorescence spectra of the original degree Chinese liquors with different years,which belongs to a well-known series in China,this article establishes a year forecast model of Chinese liquors.The research contents and innovations are as follows: firstly,from the analysis of correlation between the fluorescence spectra and the liquor's year,it is found that the correlation coefficient of three-dimensional fluorescence spectra between 0.5 year and others reaches to 0.811 4.The year information in the original spectra are mainly distributed in the area of the excitation wavelength of 200~230 and 250~320 nm and the emission wavelength in the band of 400~500 nm.The year information of the derivative spectrum distributes widely and its dispersion is high,in which the two order derivative spectra of the year information distribute more discrete.Secondly,the correlations of the two-dimensional fluorescence spectra with the excitation wavelength of 300 nm is studied.The results show that there is a high multicollinearity in the original fluorescence spectra.The value of the correlation coefficient is close to 1,in the whole range of the emission wavelength of 400~600 nm.The resolution capability can be improved after derivation,and the multicollinearityalso can be reduced at this time.Two order derivative has a better effect on the suppression of multicolinearity while most of the correlation coefficient are less than 0.6.Finally,the year forecast model of the Chinese liquors is established with the excitation wavelength of 300 nm using quantum genetic algorithm and wavelet neural network.The concept of spectral modeling information density is proposed.It is found that the error of the root mean square of original spectra in the prediction reaches to 5.4 years.The modeling effect is the worst.The main reason for this is that the original spectrum has a serious of multicollinearity,which leads to the correlation between the spectra and the year are not significant and the changes are slowly.The derivation spectra have a higher information density and a better effect of modeling than the original spectra.The validation set correlation coefficient of the second derivative spectra can reach to 0.999 8,and the error of the year forecast is 0.79 years.The research results provide not only a convenient optical method for the year calibration of the liquor,but also an important reference for the fluorescence spectra research of the multi-component gradient system.%年份白酒现已成为企业开发重点,但年份标准有较大的随意性,建立年份标准已成为规范行业和市场的迫切需要.基于某品牌原浆白酒的三维荧光光谱,对白酒年份预测模型进行了研究.研究内容和创新工作如下: 首先,研究了荧光光谱与白酒年份的相关性.研究发现: 0.5年与其他年份白酒的三维荧光光谱之间的相关系数达0.811 4;原始光谱中年份信息主要分布在激发波长为200~230和250~320 nm、发射波长为400~500 nm的光谱区;导数光谱的年份信息分布区域广且离散性高.其次,研究了荧光光谱之间的相关性.研究表明: 原始光谱具有严重的多重共线性,在400~600 nm的区间内,相关系数接近1;求导能提高光谱分辨能力并降低多重共线性,二阶导数具有更好的抑制多重共线性的作用,相关系数大部分小于0.6.最后,基于量子遗传算法-小波神经网络研究了激发波长为300 nm的白酒年份预测模型,并提出了光谱建模信息密度的概念.研究发现: 原始光谱年份预测误差达5.4年,效果最差,其原因是原始光谱具有严重的多重共线性以及光谱与年份的相关性不显著;导数光谱具有更高的信息密度和更好的建模效果,二阶导数光谱预测集的相关系数达0.999 8,年份预测误差达0.79年.研究成果将为白酒年份标定提供一种便捷的光学手段,同时也为多组分渐变体系的荧光光谱研究提供重要的参考.
展开▼
