Accurate assessment of transformer oil-paper insulating thermal aging serves as an important part to ensure the safe operation of power equipment. The successful application of Near Infrared Spectroscopy in petrochemicals and other fields provides new ideas for electrical insulation testing. The accelerated thermal aging test has experimented in a vacuum environment of 130 ℃. Fourteen groups of samples with different aging time are prepared. The spectrum of the aged insulating oil was collected by the Near Infrared Spectroscopy, and the furfural content in transformer oil was detected by high performance liquid chromatography(HPLC). There are obvious absorption peaks at 8 373, 8 264, 7 181, 7 076, 6 981, 5 855, 5 799, and 5 678 cm^-1 in the original spectrum. This study specifically analyzes the attribution of each absorption peak. The original spectrum was preprocessed using a five-point cubic polynomial Savitzky-Golay convolution smoothing algorithm. The characteristic spectral regions for aging time are selected as 11 209~10 364, 9 087~7 818, 7 390~4 424 cm^-1, with a total of 1 320 wavelength points. At the same time, the spectral information of the characteristic region is extracted by PCA, which indicates that the cumulative contribution rate of the first seven principal components is 99.78%. On the basis of the above, a PCR, PLSR, PCA-BP-ANN prediction model for aging time was established. It is shown that the PCA-BP-ANN aging time prediction model with conjugate gradient algorithm is the best, with RMSEP of 18.67 and R² of 0.997 3. The characteristic spectral region of the furfural content in the oil is selected from 9 107 to 4 424 cm^-1 for a total of 1210 wavelength points. At the same time, the spectral information of the characteristic region is extracted by PCA, which indicates that the cumulative contribution rate of the first four principal components is 99.96%. On the basis of the above, a PCR, PLSR, PCA-BP-ANN prediction model for the content of furfural in oil was established. It is shown that the PCA-BP-ANN furfural content prediction model with conjugate gradient algorithm performs best, with RMSEP of 0.134 4 and R² of 0.987 7. It is feasible to evaluate the thermal aging time and the furfural content based on near-infrared spectroscopy of insulating oil.