Diagnosis Method Of Transformer Vibration State Based On Gaussian Mixture Model

As power system equipment runs longer,increases in number and maintenance costs are greatly increased,higher requirements are put forward for equipment status diagnosis.The vibration-based transformer condition monitoring method has gained attention due to its real-time,accuracy and other advantages,but the practical application of the vibration diagnosis method has not yet been resolved.For this reason.this paper proposes a transformer abnormal diagnosis method based on statistical ideas and prior knowledge to realize real-time state monitoring of transformers.Combined with the vibration model,this paper analyzes the characteristics of the windings vibration and the iron vibration under normal and abnormal conditions are analyzed.Based on the mechanism of vibration generation,the performance of the characteristic under normal and abnormal conditions are given.And the corresponding characteristic descriptions are given,namely frequency complexity,current correlation,odd and even harmonic amplitude ratio,and signal energy distribution.This paper designs a high-precision sensing and acquisition system that can monitor the transformer vibration data in real time and upload it to the server through the network.This paper focuses on the selection of vibration probes as the main monitoring component.Through comparative analysis of different types of vibration sensors and different power supply methods,the IEPE-A26D500 piezoelectric type of vibration sensor with constant current power supply is finally selected.The equipment was installed on a 110kV transformer substation in a municipality directly under the Central Government for online data monitoring,and its vibration and electrical signals were collected.The analysis of these signal shows that the vibration state of the transformer in this substation is ideal.the frequency spectrum distribution is concentrated around 100 Hz,and the mechanical state is good.This paper uses the principal component analysis method to reduce the dimensionality of the data.At the same time,considering some factors that are not easy to consider,a transformer vibration monitoring method based on the Gaussian mixture model is proposed.The order of the Gaussian mixture model is determined by Bayesian information criteria,and the measured data is used for parameter fitting.Finally,the vibration data of the aging transformer with a long running time is used as an abnormal sample to test the proposed algorithm,which proves the effectiveness of the algorithm.