Research On Online Anomaly Vibration Detection Method For Power Transformer

Power transformer is one of the most important equipment in the power system,and its operating state is related to the safety of the entire power grid.Analysis of transformer operation status based on vibration method is an effective method for transformer online monitoring.In the process of transformer vibration online monitoring,the huge amount of vibration monitoring data leads to inconvenient data storage and processing,especially due to the incomplete data samples and the scarcity of failure modes,making it difficult to detect abnormal transformer vibration.Aiming at the problem of transformer vibration data feature extraction,an unsupervised feature selection algorithm based on feature information gain is proposed.The mutual information theory is used to analyze the correlation and redundancy between features,and the feature importance is proposed and applied as the basis for evaluating the features.The paper also presents a feature sorting algorithm based on associated information entropy,and conducts experimental simulations on standard data sets and transformer vibration data to analyze and verify their feasibility and effectiveness.About the problem of incomplete data samples,especially the lack of fault samples,during the online monitoring of the transformer’s operating status,a new heterogeneous detection technology described by support vector data was used to establish a SVDD model using a large number of normal samples to achieve the detection of abnormal or unknown samples Detection,The feasibility and effectiveness of the detection model is verified on the standard data sets and the measured data set of the transformer.Aiming at the problem of updating the transformer vibration anomaly detection model,a method based on a fast convex hull algorithm combined with a fixed buffer is used to realize online learning and updating of the transformer vibration anomaly detection model.Experimental analysis is performed on standard data sets and measured data arithmetic of transformers to verify the effectiveness and feasibility of the algorithm.