多态偏最小二乘法模型

为了更合理的确定偏最小二乘法的主成分数,提出了一种多态偏最小二乘法的建模方式。介绍了建模和预测具体实现过程。给出了预测时样品相似度计算的两种方式:直接距离法和性质得分距离法。以玉米样品近红外光谱数据为例,分别采用多态偏最小二乘法与传统偏最小二乘法建模对蛋白质指标进行了检测。结果表明:多态偏最小二乘法预测结果优于传统偏最小二乘法预测结果,有更强的适应性和兼容性。

Polymorphic Partial Least Squares Model

In order to determine the principal component number of Partial least squares（PLS）, we propose a polymorphic PLS modeling approach. Describes the specific process of building model and predictive, two methods of sample similarity calculation is gived, while we predict samples： direct distance and property score distance. The article takes the near infrared spectroscopy（NIR） data of corn Samples as example, and detects the protein by polymorphic PLS model and traditional PLS model. The results indicated that： prediction results of the polymorphic PLS is better than prediction results of traditional PLS, it has greater flexibility and compatibility.