The Anomaly Vibration Detection Of Transformer Based On Novelty Detection Technology

Transformer is one of the key equipment of power grid,and its reliability is closely related to the efficient and stable operation of power grid.Analysis of transformer operation based on vibration method has gradually been the concern of many scholars.However,there is a problem that the actual data samples are incomplete,especially the fault samples are scarce and the failure modes are incomplete.Novelty detection achieves the unknown or anomalies identified through the study of the known state observation of the sample.Therefore,it can be applied to the research of transformer vibration fault diagnosis.Aiming at the lack of fault samples in vibration anomaly detection of transformer,a Gaussian kernel function support vector data description detection model(SVDD)is established.Firstly,novelty detection and its methods are analyzed and summarized.Two kinds of models of single support vector machine are studied emphatically,and the indexes to measure the performance of the model are analyzed.Secondly,the frequency analysis of the transformer vibration signals is analyzed.The results show that the surface vibration signals of different measuring points differ greatly.And feature vector of transformer vibration signals is extracted by using wavelet packet analysis technology and the surface vibration data set of the transformer is build.The influences of Gaussian bandwidth parameters and balance parameters on the SVDD detection model are discussed and analyzed,and the parameters are set.Experiments on the surface vibration data set of the transformer show that the SVDD detection model is suitable for the detection of transformer anomalies and the SVDD detection model must have continuous learning ability due to the change of vibration characteristics during the long-term operation of the transformer.Considering the fact that the transformer vibration data gradually become more and more involved,the paper proposes a Q-ISVDD algorithm based on fast convex hull algorithm.Experiments show that the Q-ISVDD algorithm has lower time complexity and better performance than the SVDD algorithm and the Online SVDD algorithm.And the Q-ISVDD algorithm makes full use of the historical and current samples to make the SVDD detection model possess continuous learning ability,which is more advantageous in on-line detection of transformer anomalies.