Vibration Anomaly Detection of Transformer in Operation Based on Dynamic Response Analysis

In this paper, the abnormal vibration detection method of transformer in operation based on dynamic response analysis is proposed, and the identification model of dynamic response relationship between transformer working condition and surface vibration is established, and its vibration health state is judged according to the identification accuracy analysis. For the imbalance problem of vibration samples under different working conditions, a two-stage learning framework is proposed by combining decoupling learning and conditional generation confrontation network. In the characterization learning stage, the encoder is trained based on the frequency domain characteristics of the original data to fully perceive and obtain the vibration characteristics of different working conditions. In the classification learning stage, the conditional generation confrontation network is used to enhance the small sample vibration data, and the retraining strategy is adopted to train the classifier and fine-tune and optimize the classification boundary. The simulation analysis of the measured vibration signal of the transformer in operation shows that the proposed method can detect the vibration anomaly based on whether the dynamic response relationship of the transformer in operation changes or not.