Fault Diagnosis Of Autotransformer Winding In High-speed Railway Based On Attribute Selection And Support Vector Machine

China's HSR generally adopts AT power supply mode,among which autotransformer is one of the core equipments in AT power supply system.Short-circuit accidents caused by direct lightning strikes and contact system foreign bodies may lead to deformation failures such as displacement and buckling of the autotransformer windings,which greatly affects the reliability of its operation.Frequency response analysis is a important method for transformer winding fault diagnosis.At present,the intelligent diagnosis method of transformer winding based on frequency response analysis is mostly for double-winding transformer,whose high and low voltage windings are independent of each other;while HSR autotransformer often adopts split winding structure,in which electrical connection exists between windings,it's frequency response curve is quite different from that of the ordinary electric transformer,and the corresponding fault characteristics are also different.Therefore,it is necessary to propose an intelligent diagnosis method for high-speed rail autotransformer windings.In this thesis,the single-phase autotransformer windings for HSR power supply are taken as the research object.The fault sample acquisition—feature space establishment—fault modeling and pattern recognition is the main research line.In view of the shortage of autotransformer winding fault samples,the equivalent circuit model of the autotransformer winding is established.The finite element analysis is used to solve the equivalent circuit parameters and the frequency response curve is obtained.Based on the simulation model,the frequency response simulation of different degrees of fault is carried out.At the same time,the transformer test platform which can simulate various winding faults is designed.The simulation tests of different fault types are carried out,which provides a large number of fault samples for subsequent intelligent diagnosis.In order to obtain more fault information from the frequency response curve,this paper extracts the mathematical statistical features reflecting the difference between the frequency response curves,the zero-pole characteristics reflecting the changes of the internal structural parameters of the windings and the digital image features reflecting the changes of the polar plot.During the zero-pole feature extraction,the fast relaxation vector fitting is used to realize the high-precision identification of the transfer function.During the digital image feature extaction,the gray-gradient co-occurrence matrix features and the gray-scale difference statistical features of the polar graph are calculated.In order to eliminate the redundant and conflict features in the above features,the neighborhood rough set and gray relational analysis are combined to attribute the above three types of fault features,and the reduced features are used as input features of the intelligent algorithm.According to the special winding structure and hierarchical diagnosis requirements of autotransformer,the fuzzy C-means clustering algorithm and the class-based separation measure method based on class distribution characteristics are integrated in the design of binary tree topology.A hierarchical diagnosis model of binary tree support vector machine that can locate fault windings and identify fault types is established.In the design of each sub-classifier of binary tree,the effects of different kernel functions,different parameter optimization algorithms and different input features on prediction accuracy are compared.