Fault Diagnosis Method Of Transformer Winding Based On Multi-Level Features Of FRA

Transformers are widely used.It is a key equipment for power transmission and distribution,and also a traction power supply system hub.Transformers are easily impacted by short-circuit currents.The winding is prone to deformation failures such as displacement and buckling,under the effect of electric power accumulation,which affects the reliability of operation.Frequency response analysis method has been widely used and researched for its repeatability and sensitivity.Transformer winding detection based on frequency response method is mainly aimed at identifying fault types and fault severity for common winding transformers.However,there are many transformers with special structures.Different transformer structures and winding structures have different frequency response laws.Therefore,it is necessary to propose an interpretation method that can reflect the small changes in the frequency response curve and enrich the application scenarios,which is helpful to improve the fault detection of the transformer winding.A frequency response interpretation method based on multi-decomposition and image features is proposed in this paper.The main research line is fault sample acquisition,multi-decomposition research,image feature research and intelligent diagnosis.Firstly,the axial displacement,short circuit and series capacitance variation are simulated in model transformer.Secondly,the discrete wavelet transform multi-decomposition is used to enrich the characteristic information.The mother function and decomposition level are determined,the amplitude-frequency and phase-frequency curves are processed.It is applied in the faults of split winding and ordinary winding.Furthermore,Polar plot can simultaneously use the information of amplitude-frequency and phase-frequency curves.Twelve image features and FCD were extracted and applied to split windings and common windings to study the correlation between features and faults.Finally,based on 15 feature groups,the fuzzy C-means clustering algorithm is used to obtain the cluster centers and membership degrees of the data samples.Combining the specific conditions of fault identification,a binary tree topology is constructed to identifying the fault of transformer.