Feature Extraction and State Identification of Transformer DC Bias Vibration Based on HHT-SVM

Abstract

Aiming at the problem of feature extraction and state identification of transformer DC bias vibration in HVDC transmission, the principle of transformer DC bias and the vibration mechanism of core and windings are studied and analyzed. Based on the dynamic simulation experiment platform, the vibration signals of transformer under different operating conditions are collected and carried out by the statistical calculation and Hilbert-Huang Transform (HHT). Then, the characteristic parameters and vibration law are obtained and explored. Taking the statistical parameters and HHT data of transformer as feature vectors, the DC bias state of transformer is identified by support vector machine (SVM) classification method, which provides a new idea and auxiliary decision-making means for DC bias fault diagnosis of transformer. The results showed that the ratio of windings vibration in 300 Hz frequency band increases under DC bias, while that of core vibration signal in 300 ~ 600 Hz frequency band increases. Through the SVM classification method, the mean identification rate of transformer DC bias state reaches 96.8%, which can accurately identify the problem of transformer DC bias.