It is difficult to classify different types of partial discharge pulses with similar waveform characteristics by using equivalent time-frequency characteristic. In this study

on the premise of ensuring sufficient samples in each grid for statistic analysis

the equivalent time-frequency spectrum of partial discharge pulse is divided into some grids. Six statistical parameters extracted from the samples in each grid along with average equivalent time and equivalent frequency form the feature vector. All pulses in one grid are regarded as same type of discharge.The cluster analysis is used to classify these grids based on the feature vectors from every grid. This method is used to classify the mixed partial discharge pulses induced by two types of simulated defects

and the comprehensive classification result is superior to that of the method directly using equivalent time-frequency features. Additionally

the classification accuracy of both pulse types is more balanced. It is concluded that the proposed method can improve the classification accuracy to some extent and achieve effective separation among different types of discharge pulses

and it can also separate interference pulse from discharge pulse.