近红外光谱技术结合粒子群最小二乘支持向量机算法在山茱萸药材质量控制中的应用研究

Application of near infrared spectroscopy combined with particle swarm optimization based least square support vactor machine to rapid quantitative analysis of Corni Fructus

本文结合近红外光谱(NIR)定量分析技术以及粒子群最小二乘支持向量机(PSO-LS-SVM)方法,发展了一种方便、快速的用于山茱萸药材的多指标质量控制方法。实验以水分、浸出物、马钱苷、莫诺苷为质控指标,利用粒子群算法对最小二乘支持向量机算法进行参数优化,并建立定量校正模型,发现模型中各个指标校正和预测性能都优于偏最小二乘回归(PLSR)和神经网络(BP-ANN),其中校正集相关系数均大于0.942。对于未知样本的预测,PSO-LS-SVM模型的RMSEP和RSEP值分别小于1.176和15.5%,较其余两个模型更低。本文建立的PSO-LS-SVM模型具有模型性能好、预测精度高的优点。近红外光谱技术结合化学计量学方法在山茱萸药材质量控制中具有潜在的应用价值。

A novel method was developed for the rapid determination of multi-indicators in cornifructus by means of near infrared （NIR） spectroscopy. Particle swarm optimization （PSO） based least squares support vector machine was investigated to increase the levels of quality control. The calibration models of moisture, extractum, morroniside and loganin were established using the PSO-LS-SVM algorithm. The performance of PSO-LS-SVM models was compared with partial least squares regression （PLSR） and back propagation artificial neural network （BP-ANN）. The calibration and validation results of PSO-LS-SVM were superior to both PLS and BP-ANN. For PSO-LS-SVM models, the correlation coefficients （r） of calibrations were all above 0.942. The optimal prediction results were also achieved by PSO-LS-SVM models with the RMSEP （root mean square error of prediction） and RSEP （relative standard errors of prediction） less than 1.176 and 15.5% respectively. The results suggest that PSO-LS-SVM algorithm has a good model performance and high prediction accuracy. NIR has a potential value for rapid determination of multi-indicators in Corni Fructus.