以硫氰化钾(KSCN)为内标物,利用主成分分析(PCA)降维,利用支持向量机(SVM)算法建立定量分析模型——支持向量回归(SVR),并结合网格搜索(GS)、遗传算法(GA)和粒子群优化算法(PSO)三种参数优化方法,实现了芘、菲单一溶液和混合溶液的定量分析。研究结果表明:以KSCN为内标物,提高了定量分析结果的准确性;利用PCA降维提高了建模速度;三种优化模型对芘预测的平均相对误差(ARE)在7.6%以内,对菲预测的ARE在11.3%以内;三种参数优化方法对同一物质的预测结果相近,但GS的运算速度最快;综合考虑误差和分析速度后,采用GS-SVR模型获得了菲、芘混合溶液的最佳结果。表面增强拉曼光谱(SERS)技术结合SVM算法有望实现多环芳烃的定量分析。

Potassium thiocyanate (KSCN) is used as the internal standard. And principal component analysis (PCA) is utilized to reduce the dimension. Quantitative analysis model, that is, support vector regression (SVR), is established by support vector machine (SVM) algorithm. Meanwhile, three parameter optimization methods, that is grid search (GS), genetic algorithm (GA) and particle swarm optimization (PSO), are used to fulfill quantitative analysis of single and mixed solutions of pyrene and phenanthrene. The research results show that the use of KSCN as the internal standard improves the accuracy of the quantitative mensuration results. The modeling speed is improved by PCA dimensionality reduction. The average relative errors (AREs) of pyrene solution predicted by three optimized models are within 7.6%. The AREs of phenanthrene solution prediction are within 11.3%. The three parameter optimization methods have similar prediction results for the same sample, but the operating rate of GS is the fastest. Considering the errors and analysis speed, the best results of phenanthrene and anthracene mixed solution are obtained by GS-SVR model. Surface-enhanced Raman spectroscopy (SERS) technology combined with SVM algorithm is expected to actualize quantitative analysis of polycyclic aromatic hydrocarbons.