近红外光谱结合PSO-SVM算法快速定量分析鲨肝醇片

Rapid Quantitative Analysis of Batilol Tablets by NIRS Combined with PSO-SVM Algorithm

目的:借助近红外漫反射光谱(NIRS)分析技术和支持向量机(SVM)结合粒子群算法(PSO)建立鲨肝醇片快速定量分析方法。方法:以3个厂家生产的143批鲨肝醇片为分析对象,首先采集样品的NIRS,同时以硫代硫酸钠滴定的方法测定样品含量,然后在主成分分析的基础上,采用PSO-SVM算法建立鲨肝醇片定量模型。结果:主成分分析方法结果显示,选取前20个主成分的累积贡献率为95.43%。采用外部验证方法建立定量模型,120批样品用来建立模型,建模谱段为90004250 cm^(-1),预处理方法为一阶导加矢量归一化,最终优化得到的核函数径宽σ和惩罚因子c分别为0.15和2.25。23批样品作为验证集,预测结果与真值的相关系数达到了0.9854,预测均方根误差为0.5788,模型预测结果与实测结果很好吻合,其平均偏差为0.10%,最大偏差为0.99%;平均相对偏差为0.42%,最大相对偏差为4.88%。结论:所建立的快速分析定量模型可对鲨肝醇片进行准确、快速的定量分析,可用于药品快速分析和检验。

Objective:To construct a quantitative analysis model for batilol tablets by using NIRS combined with PSO-SVM algorithm(particle swarm optimization and support vector machine)to achieve the goal of quick online determination.Methods:Totally 143 batches of batilol tablets from 3 manufacturers were chosen as the samples for analysis.Firstly,the NIR spectra were collected and the contents of the samples were determined by titration with sodium thiosulfate.And then,parameters optimization of the SVM model was achieved by using PSO algorithm.Finally,120 batches of batilol tablets samples were used as training set for quantitative analysis modeling.Results:Principal components analysis showed that the first 20 main components could explain the original spectroscopy data very well with the accumulative contribution factor of 95.43%.The quantitative model was established by an external validation with 120 bathes of samples.The spectra ranges were 9000-4250 cm^(-1).The pretreatment method was one derivation and vector normalization;the optimal kernel widthσand penalty factor c were 0.15 and 2.25,respectively.The regression correlation coefficient was 0.9854 and the RMSEP was 0.5788.The average deviation was 0.10%,the maximum deviation was 0.99%,the average relative deviation was 0.42%,and the maximum relative deviation was 4.88%.Conclusion:The objective quantitative analysis model of batilol tablets constructed by using PSO-SVM algorithm has the advantages of high accuracy and efficiency,which can be used for the quick online determination.