Partial Discharge Type Recognition for Transformers based on Fisher Discriminant Method

Towards the problem of low rate of partial discharge (PD) recognition caused by lack of effective train samples, Fisher discriminant method is applied to improve recognition rate of PD for transformer. The discharge data produced by four PD models is collected, from which forty-four statistical characteristics are extracted. In order to solve the problem of singular matrix due to the high dimension, an effective dimension-reduced strategy is put forward. Forty-four characteristics are divided into seven low-dimensional subgroups, which become the input data for seven classifiers constructed by Fisher discriminant method. The PD type of the test samples is identified as that voted by results of seven classifiers. Results show that, in contrast to the back-propagation network method, the proposed method is more stable and possesses higher recognition rate under the condition of limited training samples, thus with good practical values.