In the present paper, the micro-NIR spectrometer with the splitter of linear variable filter was used to develop the recognition models of the West Lake Longjing tea and the ordinary flat tea of the year 2012 and 2013. The NIR spectral data of different years and different storage times were decomposed by PCA algorithm. The PLS-DA models were developed by the representative samples selected by the mathematical characteristics of PCA-scores' distribution in order to analyze the reason for the inadaptability of the models according to mathematical principles and find out the solution for its correction. Being examined by the external validation set, the adaptability of the authenticity identification model was enhanced effectively. The result of this research indicated that, for the West Lake Longjing tea and the ordinary flat tea, the correct recognition rate of the model developed by all different-year samples' NIR spectral data would be enhanced effectively. The model developed by the NIR spectral data of different storage time samples indicated that the physicochemical properties of the ordinary flat tea have changed remarkably after cryopreservation for 3 months, while the physicochemical properties of the West Lake Longjing tea are relatively stable. The model adaptabilities for different years and different storage times were studied according to the mathematical perspective of the principal component characteristics of spectral data. After the authenticity identification model of West Lake Longjing tea was developed, the prediction accuracy was enhanced effectively. This research would provide reference for not only the application of NIR spectroscopy in quality grading and safety of agricultural products, but also the enhancement of the prediction accuracy of the NIR grading models for agricultural products.