Automatic outlier detection based on PLS-MMD model

As a chemical detection technology for analysis of in-situ, real time and multiple materials, Laser Raman Spectroscopy is competent to take quantitative research for ingredients in ocean environment, and achieve long-term in-situ supervision of acid radical ionic concentration in seawater. Due to the inevitable interference in the actual detection environment, the spectral data often contains some outliers, the existence of which has a significant impact on the performance of the quantitative analysis model. In order to remove the outliers accurately in the in-situ detection and online analysis, the MMD algorithm was combined with the PLS quantitative analysis model (PLS-MMD model) to detect and remove the outliers in the calibration set. It was demonstrated that the PLS-MMD method can effectively eliminate the abnormal spectra in the calibration data. After removing outliers the accuracy of the measured seawater sample concentration was improved with the relative error decreasing from 1.468% to 1.160%. Correspondingly, the prediction stability of in-situ seawater is also improved.