Power transformer is an indispensable core component of power system. During the aging process of oil-paper insulation equipment, insulation oil or paper decomposes under the action of electricity or heat to produce various characteristic substances reflecting the aging state of insulation, such as furfural, acetone, methanol, CO, CO2, and dissolves in oil, which contains a lot of aging information of oil-paper insulation. In order to diagnose the aging stage of oil-paper insulation effectively, a large number of aging oil samples were obtained by accelerated thermal aging test. According to the aging days, the samples were divided into 12 categories, and 230 Raman spectra were obtained by Raman spectroscopy. The KNN algorithm is used to predict the class of test samples by Pearson correlation coefficient. Then the Euclidean distance is introduced into the model, and the KNN algorithm is improved. The prediction accuracy reaches 87.92%, and the class prediction bias is reduced.