基于表面增强拉曼光谱技术和GWO-SVR算法快速实现水中总氮总磷含量检测

提出了一种将表面增强拉曼光谱技术（SERS）和基于灰狼优化（GWO）算法的支持向量回归（SVR）相结合快速定量检测水中总氮(TN)、总磷(TP)含量的定量分析方法。传统的TN、TP检测方法不但过程繁杂，实验环境要求高，而且耗时较长，不能实现快速检测。而SERS技术操作简单，耗时短，将其与GWO-SVR算法相结合可以实现快速精确检测。以实验室配制的银溶胶作为拉曼光谱增强基底，不同浓度梯度TN、TP溶液为研究对象，分别配制TN、TP样本溶液26组和23组，其中TN溶液选取8组作为测试集，TP溶液选取7组作为测试集，剩余样本溶液作为训练集。根据待测溶液与银溶胶不同体积配比确定最佳实验方案，将TN、TP分别与银溶胶进行1∶1，1∶2，1∶3，2∶1和3∶1的体积比混合，结果表明当待测溶液与银溶胶以2∶1比例混合时增强效果最佳。采集光谱信息并对特征峰进行归属，然后采用暗电流扣除、背景扣除（基线校正）和平滑处理对原始光谱数据进行预处理。经光谱分析结果可知，由于不同浓度溶液官能团浓度差异，光谱特征峰强度随溶液浓度变化而变化。以训练集样本溶液光谱特征峰强度和溶液浓度值作为回归预测模型的输入值和输出值，建立GWO-SVR定量分析模型。通过测试集样本溶液的相关系数(r)和均方误差(MSE)对模型的预测能力进行分析，并将GWO-SVR模型和其他两种模型进行对比。结果表明，GWO-SVR模型对TN溶液预测的相关系数为0.999 5，均方误差为0.005 8，高于人工蜂群算法优化支持向量回归（ABC-SVR）和粒子群算法优化神经网络（PSO-BP）的0.993 8，0.052 7和0.998 3，0.022 7。对TP溶液预测的相关系数为0.998 5，均方误差为0.037 6，也均高于另外两种模型。而且与ABC-SVR和PSO-BP模型相比，GWO-SVR定量分析输入参数更少，收敛速度更快，更容易找到全局最优解。因此，该方法可以实现对水中TN、TP含量的快速准确检测，为水质检测提供了新方法。

Rapidly Detection of Total Nitrogen and Phosphorus Content in Water by Surface Enhanced Raman Spectroscopy and GWO-SVR Algorithm

A new quantitative analysis method was proposed, which combined surface-enhanced Raman spectroscopy (SERS) and support vector regression (SVR) based on Grey Wolf Optimization (GWO) algorithm to quickly and quantitatively detect the total nitrogen (TN) and total phosphorus (TP) content in water. The traditional TN and TP detection methods are complicated in process and time-consuming in the experimental environment. Therefore, rapid detection cannot be realized. However, SERS technology is easy to operate and time consuming, so combining it with the GWO-SVR algorithm can realize fast and accurate detection. With laboratory silver sol as the Raman enhanced substrate and TN ,TP solutions with different concentration gradients as the research objects.TN and TP sample solutions were allocated to 26 and 23 groups respectively, in which 8 groups were selected as the test set for TN solution, 7 groups as the test set for TP solution, and the remaining sample solutions as the training set. The optimal experimental scheme was determined according to the different volume ratios of the tested solution and the silver sol. TN ,TP were mixed with silver sol for 1∶1, 1∶2, 1∶3, 2∶1, 3∶1, respectively. The results showed that the enhancement effect was the best when the solution and the silver sol were mixed at a ratio of 2∶1. Spectral information was collected, and characteristic peaks were assigned. The original spectral data were preprocessed by dark current deduction, background deduction (baseline correction) and smoothing processing. The spectral analysis results show that the intensity of characteristic spectral peak varies with the concentration of solution due to the difference of functional group concentration in different concentrations of solution. The GWO-SVR quantitative analysis model was established by taking the spectral characteristic peak strength and solution concentration of the training set sample as the input and output values of the regression prediction model. Themodel’s prediction ability was analyzed by correlation coefficient (r) and mean square error (MSE) of the sample solution of the test set, and the GWO-SVR model was compared with the other two models. The results showed that the GWO-SVR model predicted the TN solution with a correlation coefficient of 0.9995 and a mean square error of 0.005 8, which were higher than the 0.993 8, 0.052 7 and 0.998 3, 0.022 7 of the artificial bee colony algorithm optimization support vector regression (ABC-SVR) and particle swarm optimization neural network (PSO-BP).The correlation coefficient of TP solution prediction was 0.998 5, and the mean square error was 0.037 6, which was also higher than the other two models. Moreover, compared with ABC-SVR and PSO-BP models, GWO-SVR has fewer input parameters, faster convergence speed, and easier to find the optimal global solution. Therefore, this method can realize the rapid and accurate detection of TN and TP content in water and provides a new method for water quality detection.