样品温度对植物油的近红外定性分析模型的影响  

Effect of Sample Temperature on Near-Infrared Qualitative Analysis Models of Vegetable Oil

作　　者：涂斌[1] 宋志强[1] 郑晓[1] 曾路路[1] 尹成[1] 何东平[2] 亓培实 

机构地区：[1]武汉轻工大学机械工程学院,武汉430023 [2]武汉轻工大学食品科学与工程学院,武汉430023 [3]武汉百信环保能源科技有限公司,武汉430023

出　　处：《中国粮油学报》2016年第4期133-137,共5页Journal of the Chinese Cereals and Oils Association

基　　金："十一五"国家科技支撑计划(2009BADB9B08);武汉市科技攻关计划(2013010501010147);武汉工业学院食品营养与安全重大项目培育专项(2011Z06)

摘　　要：主要研究不同的样品温度对基于激光近红外食用植物油分类模型预测能力的影响。选择样品温度分别为30、40、50、60℃作为研究对象,利用激光近红外光谱仪采集4种温度下的合格食用油样品的光谱数据,用标准正态变量变换(SNV)对光谱数据进行预处理,应用支持向量机分类(SVC)方法建立独立温度分类模型和混合温度分类模型,然后采用遗传算法(GA)对模型参数组合(C,g)进行寻优,确定最佳参数组合,利用建立的8个模型对4种不同温度下的预测集样品分别进行预测。试验结果表明:某个样品温度下的独立模型对于该温度下的样品的预测准确率较高,但是对于其他温度下的样品的预测准确率不够理想;混合模型对不同温度的样品预测能力相对较好,具有更好的预测稳定性和温度适应性。研究表明:样品温度对模型的预测能力有很大的影响,是建立食用植物油分类模型过程中需要考虑的重要变量。The paper has mainly emphasized that different sample temperature has different effect on the predictive ability based on the classification model of laser near infrared edible vegetable oil. First, three sample tempera- tures have been selected as 30, 40, 50, 60 ℃, respectively; the spectral data of qualified edible oil samples were collected by laser near infrared spectrometer. The spectral data were preprocessed through Standard Normal Variate （ SNV）, and classification model of independent temperature and classification model of mixing temperature were es- tablished by Support Vector Machine （SVM）. Further, after the model parameters （ C, g） being optimized by application of Genetic Algorithm （ GA）, the optimal parameters have been finally defined. The the prediction samples with the four different temperatures were predicted by exploiting 8 established mathematical models respectively. Accord- ing to the analysis, at a certain temperature, independent model had high predicting accuracy in the temperature, while it was far from ideal for the samples in the other temperature. Hybrid model had the better predicting stability and thermal adaptability on the ability of predicting samples at different temperatures. The results showed that sample temperatures had great effect on the predictive ability of classification model, which could be a very important variable in establishment of classification model of edible vegetable oil.

关 键 词：油脂 激光近红外 样品温度 模型 遗传算法 支持向量机 

分 类 号：O657.3[理学—分析化学]