油脂酸价近红外光谱检测PLS建模方法研究

Research on PLS modeling for detecting oil acid value by near infrared spectrum

将目前广泛流行的近红外光谱分析技术应用于食用油脂酸价的检测。由于近红外光谱分析是一种间接检测方法,需要先利用校正集样本建立统计模型,然后再利用模型来预测未知样品性质,因此建立准确可靠的模型是近红外光谱分析的关键。详细介绍了偏最小二乘(PLS)回归的基本思想和建模方法。为使建立的校正模型更稳健,还研究了光谱波段选择。通过间隔偏最小二乘回归波段选择法进行特征波段提取,对提取的特征波段和全谱分别进行偏最小二乘回归建模,对比分析以说明波段选择的必要性。

This paper applied the popular near infrared spectrum analysis technology into detecting oil acid value. Because the near infrared spectrum analysis is a kind of indirect detection technologies, the statistical model must be constructed first using the calibrating set samples, then it is used to forecast the properties of unknown samples thereafter. Therefore the construction of accurate and reliable model is the key technology in near infrared spectrum analysis. This paper introduced the basic idea and modeling method of PLS regression in detail. In order to make the model steadier, this paper also studied the spectrum waveband selection, extracted feature waveband through interval PLS regression, then compared and analyzed two models constructed by using extracted feature waveband and total spectrum respectively to demonstrate necessity of waveband selection.