The invention discloses an ultra-high voltage device insulation defect type judgment method. The method comprises the following steps: 1) collecting a partial discharge signal of an ultra-high voltage device; 2) carrying out wavelet packet transformation to obtain wavelet packet coefficients of an nth layer of nodes; 3) carrying out time-frequency analysis to extract time-frequency signal energy characteristics of the nth layer of nodes; 4) dividing the time-frequency signal energy characteristics of the nth layer of nodes into m groups through AP clustering; 5) selecting an optimal wavelet packet coefficient; 6) calculating minkowski distance from the time-frequency signal energy characteristic correspoinding to the optimal wavelet packet coefficient to a clustering center thereof; and 7) judging the type of insulation defect of the ultra-high voltage device according to the time-frequency signal energy characteristics of the node corresponding to the optimal wavelet packet coefficient and the minkowski distance. The invention also discloses a corresponding system. Compared with a conventional SVW classifier judgment method, the method has higher accuracy and stability.