Transformer anomaly detection based on multimeasuring point vibration feature fusion

The vibration signal on the transformer surface contains abundant information about the transformer operation state, so it is difficult to describe the transformer operation state comprehensively by the vibration signal characteristics of a single measuring point. In this paper, a transformer anomaly detection method based on multi-measuring point vibration feature fusion is proposed. This method uses consistent data fusion algorithm to obtain the weight of measuring points, uses wavelet packet decomposition to extract the frequency band energy, and combines kurtosis, low frequency ratio, frequency doubling THD and vibration energy entropy to construct feature vectors, which are input into the one-dimensional convolutional neural network optimized by Osprey optimization algorithm for anomaly detection. The simulation analysis of the measured signal of transformer vibration shows that the proposed method can fuse the vibration characteristic information of multiple measuring points, and the accuracy of abnormal detection results can reach over 99%, which provides reference for abnormal detection of transformer vibration in operation.