基于ICP-MS截尾数据与支持向量机优化模型鉴别蜂蜜植物源

Identification of the Botanical Source of Honey Based on Optimized SVM Model with Censored Data of ICP-MS

该文开展了一种电感耦合等离子体质谱(ICP-MS)截尾数据和支持向量机(SVM)分类模型识别蜂蜜植物源的研究。实验选取荆条蜜、洋槐蜜、葵花蜜、油菜蜜4种不同植物源的蜂蜜共97例,经微波消解等预处理后,采用ICP-MS分别测得蜂蜜样品中16种金属元素的含量,并研究13种具有显著性差异的金属元素,以含截尾数据和不含截尾数据的元素作为输入变量分别建立基于高斯径向基函数的SVM分类模型,并通过网格搜索法(GS)、遗传算法(GA)、粒子群优化(PSO)算法对SVM模型中的惩罚参数c和核函数参数g进行优化。结果表明:Al、Ti、Cr、Ni、As、Se、Cd、Ba、Pb 9种金属元素存在截尾数据;方差分析结果表明,4种不同植物源蜂蜜之间,Na、Mg、Al、K、Ca、Mn、Ni、Cu、Zn、Se、Ba、Pb 12种金属元素在95%置信区间差异极显著,As元素在95%置信区间差异显著,Ti、Cr和Cd在95%置信区间无显著性差异,使用替换法将截尾数据按二分之一检出限值处理并作为输入变量时所建立的SVM模型分类效果更优;使用截尾数据所建立模型的判别正确率为91.8%,而不含截尾数据建立模型的判别正确率仅为82.5%。使用网格搜索法、遗传算法、粒子群优化算法对分类模型中惩罚参数c和核函数参数g作进一步优化,通过PSO算法寻优获得惩罚参数c为62.8,核函数参数g为1.26的条件下所建立的分类模型最优,其综合判别正确率为96.9%。由此可见,利用替换法按二分之一检出限值处理截尾数据作蜂蜜植物源鉴别分析是可行的,同时表明基于ICP-MS截尾数据结合SVM优化模型能提高模型判别正确率并可有效鉴别不同植物源蜂蜜。

In this study the censored data of inductively coupled plasma mass spectrometry with support vector machine were employed in order to identify honeys according to their botanical source.97 samples were collected for this study,including four kinds of honeys such as vitex honey samples,acacia honey samples,sunflower honey samples and rape honey samples.After pretreated by microwave digestion,the 16 kinds of metal elements in honey samples were measured by inductively coupled plasma mass spectrometry and 13 kinds of metal elements with significant differences were studied.The support vector machine classification model based on Gaussian radial basis function was established by using the metal elements with and without censored data as input variables.Then,the penalty parameter c and the kernel function parameter g of the support vector machine model were optimized by three optimization algorithms:grid search,genetic algorithm and particle swarm optimization.The result showed that there are 9 kinds of metal elements have censored data,namely Al,Ti,Cr,Ni,As,Se,Cd,Ba,Pb.The analysis of variance results showed that 12 kinds of metal elements such as Na,Mg,Al,K,Ca,Mn,Ni,Cu,Zn,Se,Ba and Pb have extremely significant differences in 95%confidence interval(p<0.01),the element of As has significant differences in 95%confidence interval(p<0.05)and the elements of Ti,Cr and Cd have no significant differences in 95%confidence interval(p>0.05)among four different botanical source honeys.The censored data was processed to the one-half of the detection limit value by using the substitution method and the support vector machine model which established by censored data of metal elements as input variables has better results than the support vector machine model which without the censored data.The accuracy rate of the model established with censored data is 91.8%,while the accuracy rate of the model established without censored data is only 82.5%.Further optimization of penalty parameter c and kernel function parameter g in classification model by using grid search,genetic algorithm and particle swarm optimization,the support vector machine model with the penalty parameter c of 62.8 and the kernel function parameter g of 1.26 was the best by using particle swarm optimization.The correct rate of comprehensive discrimination of the best support vector machine classification model is 96.9%.It is concluded that it is feasible to identify honey botanical source through the substitution method which made the censored data to the one-half of the detection limit value and also shows that the optimized support vector machine model with censored data of inductively coupled plasma mass spectrometry can improve the accuracy of model discrimination and identify effectively honey samples from different botanical sources.