菜籽油在加工及贮藏过程中, 易受氧气、 温度、 光照等因素的影响, 产生氧化酸败现象。 为准确判断油脂氧化程度, 实现不同氧化模式下菜籽油品质的快速判别, 采用三维同步荧光光谱技术结合平行因子分析法及BP神经网络法建立菜籽油氧化状态的智能评价模型。 以冷榨菜籽油为原料, 将样品分别置于常温、 Schaal烘箱、 高温模式中氧化处理, 期间采集菜籽油的三维同步荧光光谱数据及理化指标, 当理化指标超出国标限定范围时, 停止采集数据。 结果表明, 菜籽油中荧光物质在不同氧化模式中的演变规律呈显著差异, 氧化温度对菜籽油荧光光谱有明显影响。 常温氧化350 d与第1 d相比, 菜籽油的特征荧光峰位置无变化, 仅在激发波长Ex为620和660 nm附近荧光峰强度发生微弱变化; Schaal烘箱氧化26 d后, 在激发波长Ex为620和660 nm附近荧光峰强度显著减弱, 且在激发波长Ex为350~450 nm之间有新的荧光峰生成; 高温氧化48 h后, Ex为620和660 nm处荧光峰消失, 在Ex为400~550 nm处产生显著荧光峰, 相对Schaal烘箱氧化, 荧光波长发生一定程度红移, 这是由于高温氧化过程中油脂氧化生成的物质稳定性较差引起的。 利用平行因子分析法对三维同步荧光光谱数据进行分解获取有效的二维荧光光谱数据, 当组分数为6, Δ λ=60 nm时激发波长的载荷值最大, 不同样品间差异最显著。 选定Δ λ=60 nm波段的二维荧光光谱数据用于智能评价, 作为BP神经网络模型的输入值, 以极性组分作为模型输出值, 分别对菜籽油三种氧化模式数据建模训练。 实验结果表明, 三种氧化模式对应的训练集、 验证集、 测试集模型相关系数 r均能达到0.9以上, 其中常温氧化模式中验证集及测试集模型的相关系数 r为1, 输出值与目标值较接近, 模型的预测效果较好; 综合三种氧化模式数据建模, 对应训练集、 验证集、 测试集模型的相关系数分别为0.999, 0.913和0.988, 均方误差均较小, 说明该模型能准确判断菜籽油的不同氧化状态。 因此, 三维同步荧光光谱技术结合平行因子分析法、 BP神经网络法建立快速检测模型能实现菜籽油不同氧化状态的判别, 为菜籽油的氧化程度的评价提供新方法, 同时为其他食用油的品质评价提供参考。

Rapeseed oil in the process of processing and storage are vulnerable to oxygen, temperature, light and other factors, resulting in oxidative rancidity phenomenon. In order to judge the oxidation degree of oil accurately and realize the intelligent evaluation of the quality of rapeseed oil under different oxidation modes, the intelligent evaluation model of rapeseed oil oxidation state was established based on the three dimensional synchronous fluorescence spectrometry combined with parallel factor analysis and BP neural network method. With cold pickled oil as raw materials, the samples were treated in the normal temperature, Schaal oven and high temperature oxidation mode respectively. During the period, the three dimensional synchronous fluorescence spectrum data and physical and chemical indexes of the rapeseed oil were collected. When the physical and chemical indexes exceeded the limits of the national standard, the data were stopped.The results of three dimensional synchronous fluorescence spectra showed that there were significant differences in the evolution of fluorescent substances in rapeseed oil in different oxidation modes. Oxidation mechanisms of rapeseed oil changed significantly with the temperature. The characteristic fluorescence peak position of rapeseed oil had no significant changes at normal temperature between 1 day and 350 days, only with a slight change of fluorescence peak near Ex 620 and 660 nm. After oxidation of 26 days in Schaal oven, the fluorescence peak near 620 and 660 nm decreased significantly, and a new fluorescence peak was formed between Ex 350 and 450 nm. The fluorescence peak of Ex at 620 and 660 nm disappeared after 48 h of high temperature oxidation, and a significant fluorescence peak produced at Ex 400~550 nm. Compared with the oxidation of Schaal oven, the fluorescence wavelength shifted to a certain extent, which was caused by the poor stability of the substance produced by the oxidation of oil in the high temperature mode. The parallel factor analysis method was used to decompose the three-dimensional synchronous fluorescence spectra. When the number of components was 6, the load value of excitation wavelength was the largest when Δ λ=60 nm, and the difference between the different samples was the most significant.The two-dimensional fluorescence spectra of Δ λ=60 nm band were selected for intelligent evaluation, which were used as the input values of the BP neural network model. The polar components were used as the output values to model the three kinds of oxidation mode data respectively. The experimental results show that the correlation coefficient R of the training set, the verification set and the test set model corresponding to the three oxidation modes can all reach above 0.9. The correlation coefficient R of the validation set and the test set model in the normal temperature oxidation mode is 1, showing that the output value and the target values are close and the prediction effect of the model is better. The correlation coefficients of the three training models, i.e. the corresponding training set, the validation set and the test set model, are 0.999, 0.913 and 0.988 respectively, and the root mean square error is small, which shows that the model can accurately determine the different oxidation status of rapeseed oil. Therefore, three-dimensional synchronous fluorescence spectroscopy combined with parallel factor analysis, BP neural network method to establish rapid detection model can achieve different oxidation state discrimination of rapeseed oil, which provides a new method for the evaluation of rapeseed oil oxidation degree, and also provides a new method for evaluating the quality of other edible oils.