Abstract: Near infrared spectroscopy (NIRS) was used to determine the content of quality components (water,crude protein,acid detergent fiber,neutral detergent fiber) of Leymus chinensis. In order to effectively reduce the redundant unknown variables and to increase the prediction accuracy and stability of the model,this study used the method of unknown variable elimination (UVE),random frog algorithm (RF) and partial least squares (PLS) to establish the quality prediction model of Leymus chinensis. The results showed that UVE-PLS screening model was superior to both full-spectrum PLS and RF-PLS screening model. UVE-PLS model when using for predicting water content,crude protein content,acid detergent fiber content and neutral detergent fiber content of Leymus chinensis significantly reduced cross-validation root mean square error (RMSECV) and prediction root mean square error (RMSEP),and increased the determination coefficient of calibration set determination coefficient (R_C²),cross-validation determination coefficient (R_CV²) and prediction set. (R_P²),indicating that UVE-PLS near infrared spectroscopy model is feasible to predict the contents of quality components (moisture,crude protein,acid detergent fiber,neutral detergent fiber) of Leymus chinensis,with a high prediction accuracy,ranging from 95% to 98%.

Key words: Leymus chinensis, Quality components, Near infrared spectroscopy, Unknown variable elimination, Random frog

摘要： 本研究采用近红外光谱法快速测定羊草（Leymus chinensis）中的常规营养成分，利用无信息变量消除法（unknown variable elimination，UVE）、随机蛙算法（random frog algorithm，RF）结合偏最小二乘法（partial least squares，PLS）建立了羊草品质测定模型，有效降低了冗余无信息变量，提高了模型的测量精度和稳定性。研究发现利用UVE-PLS筛选建立的羊草品质测定模型优于全光谱PLS和RF-PLS筛选建立的模型；UVE-PLS模型显著降低了交叉验证均方根误差和预测均方根误差，提高了校正集决定系数、交叉验证决定系数及预测集决定系数。研究表明UVE-PLS模型在测定羊草中的水分、粗蛋白、酸性洗涤纤维和中性洗涤纤维是可行的，校正集决定系数和预测集决定系数95%~98%。

关键词: 羊草, 常规营养成分, 近红外特征波长, 无信息变量消除法, 随机蛙跳