Analysis On Noise Characteristics And Evaluation System Of Power Transformer

To meet the rapid improvement of urban power grid and the development of urbanization,it’s important for the manufacturers and power supply enterprises to search a feasible noise reduction measure for transformer.In addition,the noise characteristics of transformers are various under different working conditions.Thus,it’s significant for the safe operation of transformers to research an effective way of fault detection based on its noise signals.Therefore,the modeling and calculation of transformer noises,the noise measurement and the voiceprint recognition of transformer noises are studied as followed.To analyze the vibration and noise distribution of core,a finite element model of transformer core is built and its vibration distribution on the surface are calculated.Then the noise field distribution of core is calculated using boundary element method.Results indicate that the vibration is higher on the upper side of core,and the distribution of transformer core noise isn’t even.The front direction of the core is the main propagation direction of noise and the up side of transformer core has the lowest noise volume.The noise may increase when the core is loosened.Various core slackness conditions are set to some 10 kV dry type transformer,and no-load test has been done while the noise signals have been collected.The results have shown that 100 Hz and its integral multiples are the main components of the noise spectrum,mainly on low frequency range.When the core is loose,the noise may turn louder and the higher harmonic components may increase.Noise signals of different measuring points are similar.Noise volume increases with voltage level while its spectrum pattern remains the same.To accuracily analyse noise signal of transformer,MFCC is optimized by the weighting process and principal component analysis(PCA)based on the analysis of existing MFCC algorithm.The effectiveness is evaluated by F ratio parameter.The results indicate that,compared with the exsited MFCC,the main characteristics of the transformer noise can be extracted more correctly by the optimized MFCC within less time.To validate the validity of voiceprint recognition model on transformer noise,VQ models of the transformer noise are established by LBG for recognition.The results show that the transformer noises can be recognized effectively by the voiceprint recognition models.Compared with other feature extraction algorithms,the combination of the optimized MFCC and the VQ model has higher accuracy and efficiency.Meanwhile,the effect of the frame length is also analyzed,and the result has shown that 500 ms is suitable for the transformer noise.To analyze the transformer noise in real time,a noise evaluation system of transformer is developed based on NI PXI data acquisition system.This noise evaluation system can acquire data,preserve data and analyse data as required.It can extract feature vectors of noise signals and process voiceprint recognition in real time.The research could provide new evaluation of the transformer noise and important basis of noise debasing treatments and structure optimization.