近红外光谱分析中异常值的判别与定量模型优化

Outlier Diagnosis and Calibration Model Optimization for Near Infrared Spectroscopy Analysis

介绍了利用马氏距离、Cook距离、光谱特征异常值、光谱残差比、化学值绝对误差等指标结合数理统计检验来判断光谱和化学值的异常 ,并利用这些方法进行近红外光谱定量分析中模型优化 。

Outlier diagnosis is a very important step in building near infrared calibration model. Data outlier includes spectral outlier and chemical value outlier. Mahalanobis' distance, ratio of spectral residual and spectral variable leverage test were used to evaluate sample spectral outlier. Cook's distance and the ratio of sample square error of chemical value and predict value to the mean square error of calibration set were used to test chemical value outlier. Three calibration models of protein content of 50 wheat samples, protein content of 90 corn samples and cyclohexane content of four compounds mixture were investigated. It is demonstrated that outlier test is very helpful for optimizing near infrared calibration model.