Abstract: The fault evolution law will behave differently pertinent to different kinds of insulation defects, which also make different influences on the aging process of insulation materials. Thus, performing pattern recognition to identify the types of insulation defects so as to make fault diagnosis will be beneficial to long-term stable operation of high voltage direct current gas-insulated transmission line (HVDC GIL). By means of building a 220 kV experimental platform and manufacturing four kinds of defects, including multiple locations and sizes, 180828 pulse current signals comprising 540 samples, are successfully measured using stepwise voltage ramp-up method. By virtue of extracting the discharge time and apparent discharge quantity of each partial discharge (PD) signal, spatial gray level dependency matrix is utilized to obtain 432 kinds of PD features. Then all kinds of insulation defects can be distinguished based on generalized discriminant component analysis, a kind of improved supervised subspace projection technology, as well as its kernelization forms, thus solving the dilemma that phase-resolved partial discharge mode cannot be applied to DC situations.