Evolution Characteristics and Stage Recognition Method of Surface Discharge Defects of Oil-Paper Insulation in Transformer

Surface discharge is the most common partial discharge form for transformer oil-paper insulation. In this paper, the constant voltage method was used. By analyzing discharge characteristic, discharge pattern and micro-characteristics in scanning electron microscopy (SEM), it was proposed that the surface discharge process can be divided into three typical stages, i.e. initial stage (S1), development stage (S2), and pre-breakdown stage (S3). Results show that at stage S1 discharge phase is mainly distributed in the interval of 30°~90° and 210°~270°, and discharge capacity does not exceed 100pC; at stage S2, the discharge phase is mainly distributed in 0°~160° and 180°~330°, the capacity can reach 3000pC; but at stage S3, the discharge phase covers almost the second and fourth quadrants, the discharge is distributed in 0°~360°, and the discharge capacity slightly increased to 3500pC. From stage S1 to S3, results of micro-morphology show that with deepening of discharge, the insulated cardboard fibers are gradually broken and polymerized, and their arrangement is more disordered, and the surface carbonized traces extend to the electrode's outer edge and the width becomes larger. As for Hn(φ) atlas, the changing trends of skewness sk + (sk -) and steepness ku + (ku -) in the spectrum both appear as irregular U-shape, and peak numbers npeaks + (npeaks -) increase gradually. By using genetic optimization support vector machine algorithm with feature selection, 27-dimension Phase-resolved partial discharge (PRPD) statistical data of different discharge stages are analyzed. Results show that this algorithm can accurately identify different surface discharge stages and is more accurate than fuzzy k-nearest classifier (FkNC) algorithm and Back propagation neural network (BPNN) algorithm. © 2018, Power System Technology Press. All right reserved.