随着中药制剂存储时间的延长, 其有效成分含量逐渐降低。 化学检测手段损耗样品、 检测时间长、 成本高, 利用近红外光谱对不同年份的经典名方安胎丸进行年份鉴别。 为探讨这种无损、 快速质量控制方式的可行性, 采集了三年的105粒样本在1 000～1 799 nm波段近红外光谱吸光度数据, 随机选择80个作为训练集, 25个作为测试集。 首先采用连续投影算法（SPA）, 消除原始光谱数据中的冗余信息, 对输入全光谱进行优化降维, 根据测试集的内部交叉验证均方根误差值, 从输入的800个波长中提取出11个特征波长, 分别是: (1 692, 1 714, 1 405, 1 001, 1 114, 1 478, 1 514, 1 788, 1 202, 1 014, 1 164) nm; 然后建立支持向量机(SVM)分类模型, 由于SVM模型中的参数选取对分类正确率影响很大, 利用粒子群优化（PSO）算法, 对SVM模型中惩罚参数C和核函数参数进行寻优, 形成PSOSVM分类模型; 最后将SPA降维后的特征波长输入到PSOSVM分类算法中。 用Matlab软件进行仿真测试, 分别构建SVM, SPA-SVM和本文的SPA-PSOSVM三种方法分类模型, 分类测试正确率分别达到了76%, 92%和100%。 从仿真结果可以看出, SPA波长优选可有效地降低光谱信息中存在的冗余信息, 减少建模所需的时间, 结合PSOSVM分类模型降低了模型的复杂度, 提高分类精度。 结果证实, 依照所建立的利用近红外算法, 可以准确无损区分中药制剂安胎丸生产的年份, 该研究可为中药制剂年份间差异评价提供一种思路。

With the increase of the storage time of traditional Chinese medicine, the content of its effective components gradually decreases. Chemical detection means to consume the samples, with a long period and a high cost. In this paper, near infrared spectroscopy was used to identify the years of Antai pills of the classical prescriptions with different years. In order to explore the feasibility of this nondestructive and rapid quality control method, the absorbance data of 105 samples in 1 000～1 799 nm band near infrared spectroscopy of three years were collected, 80 samples were randomly selected as training sets and 25 samples as test sets. Firstly, the Successive Projection Algorithm (SPA) was adopted to eliminate the redundant information in the original spectral data, and the full input spectrum was optimized and the dimensionality was reduced. According to the internal of the test sets, the error value of the root mean square was cross-verified, 11 characteristic wavelengths were extracted from 800 wavelengths, respectively: (1 692, 1 714, 1 405, 1 001, 1 114, 1 478, 1 514, 1 788, 1 202, 1 014, 1 164) nm. Then the Support Vector Machine (SVM) classification model was established. Since the selection of the parameters in SVM model has a great influence on the classification accuracy, the Particle Swarm Optimization (PSO) algorithm was used to optimize the penalty parameter C and the kernel function parametersin SVM model to form PSOSVM classification model. Finally, after SPA dimension reduction, the characteristic wavelength was input into PSOSVM classification algorithm. Matlab software was used in the simulation test, and SVM, SPA-SVM classification models and SPA-PSOSVM classification model in this paper were respectively constructed. The classification test accuracy reached 76%, 92% and 100% respectively. From the simulation results, it can be seen that the SPA wavelength optimization could effectively reduce the redundant spectral information and reduce the time required for modeling. The PSO-SVM classification model, the complexity of the model was reduced and the classification accuracy was improved. The results show that the near infrared algorithm established in this paper could accurately and nondestructively distinguish the production years of the traditional Chinese medicine Antai pills, and this study could provide a way of thinking for the evaluation of the differences of the years of traditional Chinese medicine.