多环芳烃(PAHs)作为一种芳香族化合物, 普遍存在于人们的生产生活中, 它具有强烈的致癌性, 威胁着人们的生命和健康。 所以, 对多环芳烃实施简洁、 高效、 精确的检测方法很有必要。 根据常见的多环芳烃类型, 选取多环芳烃萘(NAP)、 芴(FLU)、 苊(ANA)的固体粉末状物质作为实验样本。 取NAP, FLU和ANA粉末各1 g溶于少量的甲醇(光谱级)溶液, 然后转移到100 mL的去离子水溶液中, 配置PAHs标准溶液。 采用FS920荧光光谱仪, 实验中为避免荧光光谱仪本身产生的瑞利散射影响, 设置起始的发射波长滞后激发波长10 nm。 以标准溶液为基准, 获取ANA, NAP和FLU单质的水溶液的荧光光谱图。 在标准溶液的基础上, 配置0.1 mg·mL^-1的单质水溶液, 然后将ANA与NAP, FLU分别取不同的体积相互混合形成两种混合溶液, 各自形成16种不同浓度比例的混合溶液, 再取不同体积的三种溶液相互混合, 摇匀震荡, 最后一共形成48种不同体积比例的混合溶液。 最后将实验数据输入Matlab中得到苊萘、 苊芴、 苊芴萘混合溶液的荧光光谱, 发现混合溶液的激发波长在260320 nm、 发射波长300380 nm波长范围内, 最佳发射波长的位置相似, 荧光峰对应的激发波长有大部分重叠。 针对荧光光谱不能直接辨别混合物的种类的不足, 将基于遗传算法(GA)优化的支持向量机(SVM)应用于多环芳烃混合物种类的检测中, 将数据随机打乱, 并且将遗传算法的终止进化代数设为200、 训练数据和预测数据分别为36个和12个, 得到训练结果的准确率为95.42%。 将实验结果对比分析普通支持向量机和BP神经网络, 结果表明, 基于遗传算法优化的支持向量机分类误差较小, 能比较准确的分辨混合物的种类。

As an aromatic compound, Polycyclic aromatic hydrocarbons (PAHs) are ubiquitous in human production and life. They have strong carcinogenicity and threaten human lives and health. Therefore, it is necessary to implement a simple, efficient, universal and accurate method to detect polycyclic aromatic hydrocarbons. According to the common types of polycyclic aromatic hydrocarbons, solid powdery substances of polycyclic aromatic hydrocarbon naphthalene (NAP), fluorene (FLU) and acenaptene (ANA) were selected as experimental samples. Of all the samples NAP, FLU, ANA powder were carried out by 1 g and dissolved in a small amount of methanol (spectral grade) solution, then transferred them to 100 mL of deionized water solution, getting a configure PAHs standard solution. The experiment was carried out by the FS920 fluorescence spectrometer. In order to avoid the Rayleigh scattering effect generated by the fluorescence spectrometer itself, the initial emission wavelength was set to lag the excitation wavelength by 10 nm. It could obtain the fluorescence spectrum of the aqueous solution of ANA, NAP and FLU, on the basis of the standard solution, a 0.1 μg·mL^-1 aqueous solution of a simple substance was placed. Then, different volumes of ANA, NAP and FLU were mixed to form two mixed solutions, each of them formed a mixed solution of 16 different concentration ratios, and then took different volumes. The three solutions were mixed with each other, they were shaken and finally a total of 48 mixed solutions of different volume ratios were formed. Finally, the experimental data were input into Matlab to obtain the fluorescence spectrum of the mixed solution of naphthalene, anthracene and anthracene naphthalene. It was found that the excitation wavelength of the mixed solution was in the wavelength range of 260~320 nm and the emission wavelength was 300~380 nm, and the position of the optimal emission wavelength was similar. Most of the excitation wavelengths corresponding to the fluorescence peaks overlap. Support vector machine (SVM) based on genetic algorithm (GA) optimization was applied to the species detection of PAHs mixture, because the shortage of species in which the fluorescence spectrum cannot directly react with the mixture solution. The data were randomly scrambled, and the genetic evolutionary algorithm havd a termination evolution algebra of 200. Training data and prediction data are 36 and 12, respectively. Under the optimal conditions, the average accuracy of the training result was 95.42%. The experimental results were evaluated by comparing with traditional support vector machine and BP neural network. The results showed that SVM based on genetic algorithm optimization has potential for the smaller classification error and can distinguish the mixture more accurately.