Study On Noise Suppression And Pattern Recognition Of Partial Discharge Of Transformer Based On EEMD And Probabilistic Neural Network

The power transformer as the key link of the power system plays the important role in the operation of power system.There are a lot of factors on the condition of power transformer,among of which insulation aging is the most common and primary.Generally,insulation aging is the gradual process so that it is difficult to be found.There are serial aging forms of insulation aging in power transformer.The discriminative effect on power transformer corresponds to the different aging form.The partial discharge(PD)is the external form of insulation aging.The partial discharge pattern is consistent with the aging form.The insulation state can be evaluated by the proper detection of partial discharge signal and pattern recognition of discharge mode.Therefore,the risk factor can be overcome to make the power transformer safe and stable.The PD noise suppression,feature parameter extraction and discharge pattern recognition are studied based on PD characteristics of power transformers in this dissertation.The main contents of the work are as follows:(1)A noise suppression algorithm,in which the ensemble average empirical mode decomposition(EEMD)and autocorrelation function are fused,is proposed to reduce the interference of noise in PD signals detection.In the algorithm,the noisy partial discharge signal is decomposed into multiple intrinsic mode functions(IMF)to overcome the modal aliasing.The self-correlation function is adopted to identify the boundary points of the IMF dominated by noise signal and PD signal.The partial discharge signal is extracted from dominated IMFs by wavelet-based soft threshold method.The partial discharge signal is reconstructed from the IMFs.(2)The needle plate electrode,column electrode and ball electrode are designed to implement the corona discharge,surface discharge and internal air gap discharge in solid in the view of transformer insulation structure characteristics.The characteristic-constructed method based on spectrogram is developed for the condition in which three different discharges signal coexist.In the method,the PD signal is detected by the pulse current.The Three-dimensional diagrams are generated from partial discharge phase distribution(PRPD)mode.The three-dimensional diagrams are mapped to the two-dimensional diagrams.The statistical characteristic parameters as the characteristic vector are extracted from the two-dimensional diagrams.(3)A partial discharge pattern recognition method based on probabilistic neural network(PNN)is proposed.The statistical characteristic parameters of two-dimensional histogram are used as the input eigenvalues in this algorithm.Then,the Gaussian function is adopted the basic function of neurons.The probability density of the discharge pattern is estimated by the Parzen window.The discharge is discriminated by the posteriori probability.The experiment shows that the proposed method can overcome the problems of long iterative time and coding sensitivity.It has higher recognition accuracy and faster recognition rate,compared with the back propagation neural network(BPNN)and the hidden Markov model(HMM)recognizer.