The use of moment features for recognition of partial discharges in generator stator winding models

This contribution focuses on applying moment method on feature extraction and partial discharge (PD) discrimination. Four kinds of model bars are used to simulate typical partial discharges in generator stator winding. Moment features are calculated from the 3-d φ-q-n pattern charts. Back-propagation network (BP) is used to perform the recognition. The input vectors of BP network are formed in four ways: tabulated data, surface fitting parameters, moments and central moments. The effectivity of PD recognition with different kinds of input vectors is compared. The investigation shows that central moments have satisfactory ability in discriminating some typical types of PD in generator and in compressing the dimension of input characteristic vectors