基于Raman光谱和支持向量机回归的古井贡酒年份鉴别方法

Study on the Recognition of Liquor Age of Gujing Based on Raman Spectra and Support Vector Regression

白酒年份的快速准确鉴定是白酒品质分析的重点和难点问题之一。实现白酒年份酒的快速、准确的鉴别,对促进白酒行业的健康发展、维护消费者的合法权益具有重要意义。光谱分析法结合模式识别技术是实现白酒品质快速鉴别的首选方法之一,而Raman光谱由于其受水的影响很小且很少或不需要样品前处理,在白酒分析中具有广阔的发展空间。因此,采用Raman光谱和支持向量机回归(SVR)建立数据分析模型,用于不同年份白酒的年份鉴定和同一年份不同贮存时间的白酒年份鉴定。该研究创新之处主要包括如下三个方面:(1)应用Raman光谱对白酒品质进行分析,在分析方法的应用上具有一定的创新之处。(2)研究白酒的年份鉴定问题,在研究对象的选择上,具有一定的创新之处。(3)建立基于回归框架的白酒年份与年份指数对应关系,实现白酒年份识别及预测,不仅可以有效鉴别白酒年份,同时可用于鉴别白酒贮存时间,因此,在分析方法的确定和应用上,具有一定的创新之处。实验中采用古井贡5年、8年、16年及26年系列年份酒进行了实证分析,数据分析实验结果表明,所建立的基于Raman光谱和SVR的白酒年份鉴别分析流程和方法,对鉴别不同年份的白酒,以及同一年份不同贮存时间的白酒样品(包括对数据库内已有样本年份的鉴别,以及对数据库内没有的盲样的年份预测),均取得较好的应用效果,相比于其他常用回归分析方法具有明显的优越性,可以为白酒年份酒分析提供一定的技术支持。

It is an important and difficult research point to recognize the age of Chinese liquor rapidly and exactly in the field of liquor analyzing,which is also of great significance to the healthy development of the liquor industry and protection of the legitimate rights and interests of consumers.Spectroscopy together with the pattern recognition technology is a preferred method of achieving rapid identification of wine quality,in which the Raman Spectroscopy is promising because of its little affection of water and little or free of sample pretreatment.So,in this paper,Raman spectra and support vector regression（SVR）are used to recognize different ages and different storing time of the liquor of the same age.The innovation of this paper is mainly reflected in the following three aspects.First,the application of Raman in the area of liquor analysis is rarely reported till now.Second,the concentration of studying the recognition of wine age,while most studies focus on studying specific components of liquor and studies together with the pattern recognition method focus more on the identification of brands or different types of base wine.The third one is the application of regression analysis framework,which cannot be only used to identify different years of liquor,but also can be used to analyze different storing time,which has theoretical and practical significance to the research and quality control of liquor.Three kinds of experiments are conducted in this paper.Firstly,SVR is used to recognize different ages of 5,8,16 and 26years of the Gujing Liquor;secondly,SVR is also used to classify the storing time of the8-years liquor;thirdly,certain group of train data is deleted form the train set and put into the test set to simulate the actual situation of liquor age recognition.Results show that the SVR model has good train and predict performance in these experiments,and it has better performance than other non-liner regression method such as the Partial Least Squares Regression（PLS）method,and can also be applied in the practice of liquor analysis.