Models established by spectral data collected by different spectroscopy instruments have model failure problems, leading to the low utilization rate, which is not conducive to the development of the spectroscopy industry. This study attempted to explore the model transfer in the two indicators of edible oil acid values and peroxide values by using migration learning analysis. The experimental samples were 50 edible oil samples. The experimental instruments were VERTEX70 Fourier infrared spectrometer and Antaris II Fourier near-infrared spectrometer (including fiber optic probe and transmission probe). Three sets of experiments were designed with different instrument combinations. A near-infrared spectroscopy quantitative analysis model was established based on the partial least squares method, and the model transfer study was carried out. Taking the first set of experiments as an example, the correction coefficient of the acid values and peroxide values model were established after migration learning, which increased from 0.068419 and -0.371980 to 0.730980 and 0.819040. The corrected root mean square error coefficient decreased from 0.358180 and 0.090110 to 0.192480 and 0.032720. Experiments showed that migration learning could effectively alleviate the model failure problem and improve the generalization ability of the model. This research provides a new idea for solving the wide application of the spectral analysis model.