Research on Tap-changer Fault Diagnosis of Transformer Based on an Improved Binary Tree SVM

According to the on-load transformer tap-changer malfunction current characteristics, the paper designs a fault diagnosis method based on improved binary tree SVM. First calculate the center of gravity of all classes, and calculate the distance between the center of gravity of each classes to overcome the effects of the outliers. Then constructs a separability measure with weights by the distance of center of gravity between different classes and the hyper-sphere radius of class, therefore, the class which has the large distance between the classes and the wide spatial distribution within class is seperated prior. Compared with different algorithms, it has verified the improved binary tree SVM superiority. This paper takes transformer tap-changer as the research object, extracts current feature under the three conditions : normal running. disconnection and poor contact. It compares with many types of different algorithms to verify the superiority of the improved algorithm, and achieves an accurate diagnosis of common faults.