基于化学计量学方法的黄水还原糖预测模型研究

Research on prediction model of reducing sugar in yellow water based on chemometric method

中文关键词:  黄水  还原糖  傅里叶近红外光谱法  间隔偏最小二乘法  连续投影算法

英文关键词:yellow water  reducing sugar  Fourier near-infrared spectroscopy  interval partial least squares  successive projections algorithm

中文摘要:

      目的 利用预处理对近红外光谱原始数据集进行降噪及非相关信息剔除后, 采用间隔偏最小二乘法(interval partial least squares, iPLS)与连续投影算法(successive projections algorithm, SPA)联用的特征波段筛选算法降低模型复杂度, 建立高精度低冗余度的黄水还原糖预测模型。方法 在最佳的3种预处理方法的基础上, 利用竞争性自适应重加权算法、间隔偏最小二乘回归法、连续投影算法对250个样品的光谱数据进行特征波段筛选, 采用光谱-理化值共生距离算法进行样品集的划分, 划分比例为3:1。结果 黄水还原糖预测模型经iPLS-SPA算法处理后, 得到了更高的精度与稳定性, 且预测可决系数较原始数据集提升7.28%, 为0.962; 预测均方根误差下降85.40%, 为0.220; 光谱变量数下降95.46%, 为100。结论 在预处理后加入iPLS-SPA特征波段筛选算法, 能够提升黄水还原糖预测模型精度, 极大减低冗余度。

英文摘要:

      Objective After denoising and removing irrelevant information from the original infrared spectrum data set by preprocessing, the feature band screening using interval partial least squares (iPLS) combined with successive projections algorithm (SPA) was adopted. The algorithm reduces the complexity of the model and establishes a high-precision and low-redundancy reducing sugar prediction model in yellow water. Methods On the basis of the best 3 kinds of preprocessing methods, the spectral data of 250 samples were screened by using the competitive adaptive re-weighting algorithm, interval partial least squares regression method and continuous projection algorithm. Spectral-physicochemical value co-occurrence distance algorithm was used to divide the sample set, and the division ratio was 3:1. Results After being processed by the iPLS-SPA algorithm, the yellow water reducing sugar content prediction model obtained higher accuracy and stability, and the prediction coefficient of determination increased by 7.28% compared with the original data set to 0.962; the prediction root mean square error decreased by 85.40% to 0.220; the number of spectral variables dropped by 95.46% to 100. Conclusion Adding the iPLS-SPA feature band screening algorithm after preprocessing can improve the accuracy of the yellow water reducing sugar content prediction model and greatly reduce the redundancy.