基于连续投影算法的黄酒成分检测模型

Quantitative models research of the willet wine compositions based on successive projection algorithm

为减少建模过程中的计算量,提高模型的稳健性以及预测精度,将连续投影算法(SPA)应用于黄酒非糖固形物、酒精度、总酸和氨基酸态氮的近红外检测模型的建立。首先采用K-S法选择具有代表性的校正样本100个,然后运用SPA算法对全光谱进行敏感波点提取,4个指标经SPA筛选分别得到20、23、19、13个特征波长点,与其特征分子结构的吸收峰位置相对应。进而分别建立SPA-PLS、SPA-MLR以及全光谱-PLS模型。结果表明:SPA-MLR模型效果最好,各指标R2分别达到0.881、0.998、0.983和0.924,且RMSEP也较全波段建模减小了10%,有效地加快模型运算速度的同时,提高了模型的稳定性与准确度,为黄酒品质快速检测提供了参考。

In order to reduce the amount of calculation and improve the robustness and prediction accuracy of the models in the process of modeling,Successive Projection Algorithm( SPA) is applied into millet wine. Based on the parameters of non sugar solids,alcohol,total acid and amino acid nitrogen,the near infrared detection model is established. Firstly 100 representative samples of calibration are selected by the method of K-S. Then SPA algorithm is used to extract the sensitive wave points from full spectrum. The four components respectively get 20,23,19,13 characteristic wavelength points which correspond to their absorption peak position of characteristic molecular structure. At last,SPA-PLS,SPA-MLR and full spectrum-PLS model are established separately. Results show that SPAMLR model have the best effect that R2 reachs 0. 881,0. 998,0. 983,0. 924 respectively,and the RMSEP also reduces by 10% compared with that in full wave modeling. It speeds up the calculation of the model effectively and improves its stability and accuracy,thus provides a reference for the rapid test of millet wine.