In electrical equipment

different types of discharge have different damage to oil-paper insulation be-cause of the difference of discharge power and energy

which means different detriment to insulation. Determiningthe partial discharge mode and type are the basis of discharge risk assessment. Many feature parameters includingthe basic parameters and statistical parameters are extracted from partial discharge spectrogram under different de-fects in this paper. In an attempt to reduce the dimension of identification parameters

the characteristic vector sepa-rability assessment criteria are defined and using the floating forward search algorithm

9 group feature parameterswith optimal separability are selected. The feature vector is reduced to 2-dimention adopting the principal componentanalysis

linear discriminant analysis

kernel principal component analysis and generalized discriminant analysismethods. The results show that the optimal separability of feature vector can be obtained using the general separabili-ty analysis method. Then

the pattern recognition classifier which combines many kinds of algorithm is extracted.Comparing the recognition results of minimum distance method

artificial neural network and support vector ma-chine（SVM）methods

the results present that the classifier recognition accuracy reaches 95.8%. Finally

the adopt-ed pattern recognition method is applied to recognize the defects type of partial discharge of actual electrical equip-ment. The results reveal the accuracy of the adopted recognition method.