提出了一种基于拉曼光谱和改进人工蜂群算法优化支持向量机回归(IABC-SVR)算法快速定量检测山羊血清蛋白含量的方法。 传统人工蜂群算法在数据区域规模较大时, 收敛速度逐渐减慢, 出现效率低、 精准度下降、 局部最优解概率高等问题。 所提出的算法解决了这些问题, 使算法在进化前期避免陷入局部最优解, 在进化中后期能够保持解的全局搜索能力。 常规测定血清蛋白总量的方法通常采用凯氏定氮法、 双缩脲法等, 但存在时效慢、 污染样本等缺点。 采用拉曼光谱法进行检测, 具有快速、 无损的优点。 以山羊血清为分析对象, 按一定体积比配置35组待测样本, 用拉曼光谱仪采集拉曼光谱, 光谱采集范围为300~1 300 cm^-1, 采用基线矫正去除荧光背景, 使用Savitzky-Golay光谱平滑法对原始光谱进行平滑处理, 归一化处理光谱数据, 并对拉曼光谱特征峰进行归属。 实验结果表明, 拉曼光谱能够表征血清中主要化学集团的信息, 且由于官能团浓度差异, 光谱特征峰强度随浓度变化明显, 因此基于特征峰信息可以测定血清蛋白总量。 实验中, 以购买的山羊血清蛋白含量为基准, 通过配置样本的体积比得到各组待测血清样本的蛋白含量, 配置的单个液体样本体积为3 mL, 随机选取8组实验样本作为模型测试集, 剩余27组作为模型训练集。 以经过处理的光谱特征峰强度和对应的血清蛋白含量分别作为模型的输入值及输出值, 建立IABC-SVR, ABC-SVR和BP三种算法的定量模型, 对测试集血清蛋白总量进行预测。 最后通过均方差(MSE), 相关系数(r)与建模时间分别进行对比, 结果表明通过IABC-SVR建立的山羊血清蛋白定量矫正模型效果最佳, 模型的相关系数为0.990 27, 均方误差为0.244 3, 建模时间为1.9 s, 预测值方差均小于0.001 g·mL^-1, 预测准确率为99.8%。 实验结果表明, 应用激光拉曼光谱技术结合IABC-SVR算法, 对快速定量检测山羊血清蛋白含量, 具有较高的准确率和稳定性。

A method based on Raman spectroscopy and improved artificial bee colony algorithm to optimize the support vector machine regression (IABC-SVR) algorithm for rapid quantitative detection of goat serum protein content was proposed. The traditional artificial bee colony algorithm gradually slows down the convergence rate when the data area is large in size, which causes problems such as low efficiency, decreased accuracy, and high optimal local solution probability. The proposed algorithm solves these problems so that the algorithm avoids falling into the local optimal solution in the early stage of evolution, and can maintain the global search ability of the solution in the middle and late stages of evolution. Conventional methods for measuring total serum protein usually use Kjeldahl method, biuret method, etc., which has the disadvantages of slow aging and contaminated samples. This paper uses Raman spectroscopy for detection, which has the advantages of fast and non-destructive. Using goat serum as the analysis object, configure 35 groups of samples to be measured according to a certain volume ratio. Raman spectra were collected using a Raman spectrometer with a spectral collection range of 300~1 300 cm^-1. Baseline correction was used to remove the fluorescent background, and Savitzky-Golay spectra were used to smooth this method smoothes the original spectrum, normalizes the spectral data, and assigns the characteristic peaks of the Raman spectrum. The experimental results show that Raman spectroscopy can characterize the information of the main chemical groups in the serum, and due to the difference in functional group concentration, the spectral characteristic peak intensity changes significantly with the concentration, so the total serum protein can be determined based on the characteristic peak information. In the experiment, based on the purchased goat serum protein content, the protein content of each group of serum samples was obtained by configuring the volume ratio of the samples. The volume of a single liquid sample was 3 mL. Eight groups of experimental samples were randomly selected as the model test set. The remaining 27 groups are used as model training sets. The processed spectral characteristic peak intensity and the corresponding serum protein content were used as the input and output values of the model to establish a quantitative model of IABC-SVR, ABC-SVR, and BP algorithms to predict the total serum protein in the test set. Finally, the mean square error (MSE), correlation coefficient (r) was compared with the modeling time, and the results showed that the goat serum protein quantitative correction model established by IABC-SVR had the best effect. The correlation coefficient of the model was 0.990 27, and the mean square error was 0.244 3, the modeling time is 1.9 s, the variance of the predicted values are less than 0.001 g·mL^-1, and the prediction accuracy is 99.8%. The experimental results show that the laser Raman spectroscopy technology combined with the IABC-SVR algorithm has high accuracy and stability for the rapid quantitative detection of goat serum protein content.