Online Anomaly Detection of Transformer Vibration Based on SVDD Incremental Learning

Abstract

The problems in theoretical model analysis and the lack of prior knowledge bring difficulties and challenges to the analysis of transformer by the vibration method. In this paper, an incremental learning algorithm based on support vector data description was proposed for on-line anomaly detection of transformer vibration. The anomaly detection model was established by using one-class support vector data description, the incremental data were reduced based on the Quickhull algorithm to achieve quick incremental learning. The simulation results of measured transformer vibration showed that the proposed method could quickly learn new samples and accurately capture transformer vibration changes.