Acoustic Signal Diagnosis Technology For Typical Anomalous State Of Power Transformer

As an important equipment in the power grid,the operation state of power transformer directly affects the reliability of power system.The acoustic signal of the transformer can reflect the internal mechanical anomalous state,realize the early warning and diagnosis of transformer in the early stage of typical anomalous state,and then avoid major accidents.It is of great significance to maintain the safe and stable operation of power grid.Aiming at the problems existing in acoustic signal diagnosis such as acoustic sensor layout,ambient acoustic interference denoising,anomalous state acoustic eigenvalues and anomalous state diagnosis algorithm,this paper studied the methods oftransformer near acoustic field reconstruction and acoustic signal monitoring measurement position optimization,proposed the methods of transformer ambient noise identification and acoustic signal denoising,and constructs the acoustic signal eigenvalue group of typical mechanical anomalous state of transformer,The transformer anomalous state acoustic signal diagnosis method based on eigenvalue group threshold is formed.Finally,the transformer anomalous state acoustic signal online monitoring system platform is developed and is applied.The main research contents are as follows:The selection method of measuring positions for transformer acoustic signal monitoring is studied.The surface acoustic signals of all parts of the transformer are collected as the equivalent source of the acoustic field.Combined with the measured data and the finite element method,the near-field reconstruction of the transformer is realized,and the goodness of fit between the measured spectrum results and the measured spectrum results is between 0.85～0.92.It is verified that the reconstruction results have reference value for the optimization of nearfield frequency spectrum measurement positions;Based on the reconstruction results of near acoustic field,the Pearson equilibrium correlation coefficient is constructed,it is used to measure the correlation degree between the 120 alternative measurement positions in the space and the spectrum of the equivalent surface acoustic source.The results show that the correlation degree of positions near the ground and wall is low,and the optimal acoustic signal monitoring position should be set at a certain distance from the ground and the reflecting wall,and in the vertical middle position.The methods of distinguishing ambient noise and acoustic signal denoising of transformer are studied.The mixed cepstrum constructed by Mel Frequency Cepstral Coefficient,Gammatone Filter Cepstral Coefficient,Power-Normalized Cepstral Coefficient is combined with lightweight convolution neural network to quickly distinguish the unknown type of acoustic signals.The unknown type of acoustic signals can be divided into four categories:no interference,transient interference,steady strong interference and steady weak interference;In view of the difficult denoising of transient interference and steady strong interference,this paper proposes the blind source separation method based on similarity matrix and the amplitude phase angle fluctuation method respectively to realize the denoising of the ambient noise around the transformer.The acoustic signal eigenvalue group of typical mechanical anomalous state of transformer is constructed.The theoretical analysis and simulation experiments of the transformer are carried out around the DC bias state,loose state of internal components and short-circuit impact state of the transformer.The research shows that the DC bias state of the transformer will lead to the decrease of the proportion of acoustic signal fundamental frequency,the increase of high/low frequency ratio and the increase of 50Hz odd/even frequency multiplication ratio.The loosening of transformer components will lead to the increase of acoustic signal high-low frequency ratio and the decrease of autocorrelation mean value,The degree of winding deformation during shortcircuit impact of transformer is directly proportional to the proportion timefrequency spectrum entropy,and the accumulation of winding deformation is directly proportional to the Euclidean distance of proportion time-frequency spectrum components.In addition,it is pointed out that the laboratory model can study the variation law of acoustic signal in fault state,but the numerical variation range of actual eigenvalue is quite different from that of actual transformer,so a large number of actual transformer acoustic signal samples need to be collected to determine the early warning threshold of eigenvalue group,which will be used in actual transformer acoustic signal monitoring.The acoustic signal diagnosis methods are studied,and diagnosis platform of transformer anomalous state are developed.Acoustic signal diagnosis method is formed based on the manually designed acoustic signal eigenvalue threshold and has been verified in an actual case.According to the acoustic signal data of 162 normal transformers in operation,the clear early warning threshold of acoustic signal eigenvalue group is given.The lower early warning threshold of fundamental frequency proportion is 0.025,the upper early warning threshold of high/low frequency ratio is 0.1366,the upper early warning threshold of 50Hz odd/even frequency multiplication ratio is 0.852,the lower early warning threshold of autocorrelation mean value is 0.9775,the upper early warning threshold of proportion time-frequency spectrum entropy is 0.066,the upper early warning threshold of Euclidean distance of proportion time-frequency spectrum components is 1.568;Finally,the online monitoring system platform of transformer acoustic signal is developed,which can realize the online acquisition and diagnosis function of transformer acoustic signal,and has carried out practical application.