Near-infrared spectroscopy is a type of molecular vibration spectroscopy. Temperature variations cause changes in molecular vibrations such as O-H and inter molecular forces such as hydrogen bonding， which lead to absorption spectral intensity and peaks changes， affecting the prediction accuracy of minor components such as blood glucose. To address the impact of temperature perturbation on spectral detection and modeling analysis， a method of temperature perturbation discrimination based on aquaphotomics and two-trace two-dimensional correlation spectroscopy （2T2D-COS） was proposed. The 2T2D-COS analysis was applied to diffuse reflectance spectra of simulated solutions under temperature perturbation and varying glucose concentrations. Spectral features induced by changes in temperature and glucose concentration were successfully extracted， revealing distinct water spectral patterns under different perturbations. Quantitative analysis shows that temperature changes of 0.1 °C is equivalent to glucose concentration changes of 45 mg/dL in terms of intensity. A temperature perturbation outliers discrimination model was further established based on raw spectra， water spectral features， and 2T2D-COS asynchronous spectra. The accuracy rates of the model based on 2T2D-COS asynchronous spectra is 95.83%. After removing outliers， the root mean square error of glucose concentration prediction is reduced by 51.89%. This work provides a foundation for improving the accuracy of in vivo blood glucose detection using near-infrared spectroscopy.