基于THz光谱检测大豆油中反式脂肪酸含量建模方法研究

发布时间：2018-05-17 03:37

本文选题：大豆油 + 反式脂肪酸　；参考：《哈尔滨商业大学》2017年硕士论文

【摘要】：食用油是人类膳食的重要组成部分,其质量安全对人体健康至关重要。近年来,食用油脂中反式脂肪酸含量超标问题尤为严重,已引起社会各界的广泛关注。目前,我国针对反式脂肪酸含量的测定方法普遍存在消耗化学试剂多、检测速度慢、测定程序复杂、需要对样品进行复杂的前期处理等问题,难以满足现代社会对食用油质量快速、准确、简便、现场化的测定要求。太赫兹光谱分析技术可以解决传统方法的诸多问题,更适用于油脂生产过程中的品质监控,因此本文提出基于太赫兹光谱技术检测食用油中反式脂肪酸含量的方法,以我国第一大食用油——大豆油为例,重点对反式脂肪酸含量太赫兹光谱分析中的数据处理以及建模方法进行深入研究。首先,制备出不同反式脂肪酸含量的大豆油脂样品34个,利用气相色谱仪精确测定了其反式脂肪酸含量,同步采集了样品的太赫兹时域谱并利用傅里叶变换将其转换为频域谱,然后通过光学参数计算得出太赫兹吸收谱以及折射谱。在分析了大豆油脂太赫兹光谱吸收特点的基础上,剔除异常样品1个。将33个样品进行了分集,按反式脂肪酸含量的多少进行排序,从中选取28个作为训练集样品建立校正模型,剩下的5个作为预测集样品对模型进行验证;为了寻找最佳模型,分别利用偏最小二乘(PLS)、支持向量机回归(SVR)和最小二乘支持向量机(LS-SVR)三种方法建模并对比分析,最终确定建模效果最好的是LS-SVR,预测误差均方根RMSEP为0.3246,决定系数R2为0.9792,相对标准差RSD为3.81%,可以满足实际检测要求;为了进一步提高模型的预测精度,分别采用网格搜索法、粒子群算法(PSO)、遗传算法(GA)对LS-SVR模型参数进行优化,并对优化结果进行对比分析,发现粒子群(PSO)算法对LS-SVR模型参数优化的效果更佳、更稳定,预测误差均方根(RMSEP)、决定系数(R2)和预测相对标准偏差(RSD)分别达到0.0763、0.9989和0.90%,模型的预测精度得到了显著提高。本文研究证明了利用太赫兹光谱检测油脂中反式脂肪酸含量的可行性,为开发专用油脂太赫兹光谱分析仪器及实现在线检测奠定了理论基础。
[Abstract]:Edible oil is an important part of human diet, its quality and safety is very important to human health. In recent years, the problem of trans fatty acids in edible oils is especially serious, which has attracted wide attention from all walks of life. At present, there are many problems in the determination of trans fatty acids in China, such as the consumption of chemical reagents, the slow detection speed, the complexity of the determination procedures, and the need for complex pre-treatment of the samples. It is difficult to meet the requirements of fast, accurate, simple and field determination of edible oil in modern society. Terahertz spectroscopy can solve many problems of traditional methods and is more suitable for quality control in oil production. Therefore, a method based on terahertz spectroscopy to detect trans fatty acid content in edible oil is proposed in this paper. Taking soybean oil, the largest edible oil in China, as an example, the data processing and modeling method in THz spectrum analysis of trans fatty acid content were studied in detail. First of all, 34 soybean oil samples with different trans fatty acid contents were prepared, and their trans fatty acids were accurately determined by gas chromatograph. The terahertz time-domain spectra of the samples were simultaneously collected and converted into frequency domain spectra by Fourier transform. Then the terahertz absorption spectrum and refraction spectrum are calculated by optical parameters. On the basis of analyzing the absorption characteristics of terahertz spectrum of soybean oil, one abnormal sample was eliminated. In order to find the best model, 33 samples were sorted according to the content of trans fatty acids, 28 samples were selected as training set samples to establish calibration model, and the remaining 5 samples were used as predictive set samples to verify the model. Three methods, partial least squares (PLS), support vector machine regression (SVR) and least squares support vector machine (LS-SVR), are used to model and analyze the model. Finally, LS-SVR is the best model, the root mean square (RMSEP) of prediction error is 0.3246, the coefficient of determination (R2) is 0.9792, and the relative standard deviation (RSD) is 3.81. in order to further improve the prediction accuracy of the model, the grid search method is used. Particle Swarm Optimization (PSO), genetic algorithm (GA), is used to optimize the parameters of LS-SVR model, and the results are compared and analyzed. It is found that PSO (Particle Swarm Swarm Optimization) algorithm is more effective and stable in the optimization of LS-SVR model parameters. The mean square error (RMS), the determination coefficient (R2) and the relative standard deviation (RSD) of prediction are 0.0763 ~ 0.9989 and 0.90, respectively. The prediction accuracy of the model has been improved significantly. In this paper, the feasibility of using terahertz spectrum to detect the content of trans fatty acids in oils has been proved, which lays a theoretical foundation for the development of a special THz spectrometer for oil analysis and the realization of on-line detection.
【学位授予单位】：哈尔滨商业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TS227;TP18

【参考文献】