A Transformer Winding Deformation Fault Diagnosis Method Based on Improved Directed Acyclic Graph

Transformer winding is an important component of transformer. The faults caused by mechanical deformation of windings is about 15% of the total transformer.This paper presents a transformer winding deformation fault diagnosis method based on improved directed acyclic graph support vector machine (DAG-SVM). The most easily distinguished types are arranged in the upper layer of digraph to reduce the cumulative error and make up for the fault tolerance ability of digraph. In this paper, three-phase transformer is selected to carry out tests to verify the effectiveness of this method. Firstly, the vibration signals of windings under normal and fault conditions are obtained through experiments. Secondly, feature extraction of vibration signal is carried out by energy entropy decomposition of wavelet packet. Thirdly, the improved DAG-SVM is constructed by using the distance measure between classes. Finally, the improved DAG-SVM is used for transformer winding deformation fault diagnosis. The results show that this method is superior to traditional methods in accuracy and time-consuming diagnosis.