To tackle the problem of fault diagnosis of a high voltage shunt reactor, a new method based on Empirical Wavelet Transform (EWT) and a multi-scale mathematical morphological spectrum for feature extraction is proposed. Kernel K-means clustering is used to diagnose fault pattern recognition. First, the vibration signal of the reactor under the three conditions is decomposed by EWT to obtain several modal components. Then, the correlation coefficients of each modal component and the original signal are calculated separately and arranged in descending order of the coefficient size. The first four modes are taken. The state components constitute the effective component vector; the multi-scale morphological spectrum is used to analyze and calculate the effective component vector to form a four-dimensional eigenvector. Finally, Kernel K-means clustering is used to classify and identify the sample feature set. The experiment proves that the method can effectively extract the characteristic quantity of the vibration signal of the reactor and can correctly identify the different working conditions of the reactor. This work is supported by National Natural Science Foundation of China (No. 51577050) and Key Science and Technology Project of State Grid Jiangsu Electric Power Company in 2018 (No. J2018014).