基于支持向量机的食醋总酸近红外光谱建模

摘要

Using 95 vinegar samples with different types and from different areas as raw materials, spectrum analysis for total acid in vinegar was carried out with application of least squares support vector machine(LS-SVM) based on statistics. The pretreatment spectra was conducted principal component analysis(PCA) to get principal component signals as the input variables to establish the near-infrared spectral model of total acid in vinegar, and comparison with partial least squares(PLS) model and backward interval partial least squares(biPLS) model was also studied. The results showed that correlation coefficient (rc) of calibration set and cross-validation root mean square error were 0.9614 and 0.2192, respectively, and correlation coefficient (rp) of prediction set and cross-validation root mean square error were 0.919 and 0.3226, respectively. Correlations of near infrared spectrum and contents of acid in was non linear. The model application of LS-SVM had accuracy than that of PLS and biPLS.%为了得到稳定可靠的食醋总酸光谱模型,以不同产地、不同种类的95个食醋样品为研究对象,应用基于统计学原理的最小二乘支持向量机(LS-SVM)对食醋总酸含量进行光谱分析.对预处理后的光谱进行主成分分析(PCA),以主成分信号作为输入变量建立食醋总酸含量的近红外光谱模型,并与偏最小二乘法(PLS)和向后区间偏最小二乘法(biPLS)模型进行比较.结果表明,LS-SVM模型中的校正集中的相关系数(rc)和交互验证均方根误差(RMSECV)分别达到0.9614和0.2192,预测集相关系数(rp)和预测均方根误差(RMSEP)分别达到和0.919和0.3226,均优于PLS和biPLS模型.研究表明,近红外光谱与食醋总酸含量呈非线性关系,采用LS-SVM建立的模型预测性能更好,精度更高.