关键词: 变压器, 故障诊断, 完全自适应噪声集合经验模态分解, 递归分析, 卷积神经网络

Abstract: Aiming at the problem of difficult to extract the features of transformer vibration signal and low utilization of time information under noise interference, a fault diagnosis method based on the combination of multiple noise reduction and Recurrence Plot (RP) is proposed. First, the vibration signal is subjected to multiple noise reduction by complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) combined with adaptive wavelet thresholding; Second, recursive analysis is used to construct the time-series signals as recurrence maps, preserving the time-dependent and nonlinear dynamics features; Finally, the advantages of CNN in image processing are utilized to extract spatial features using recursive graphs as input for fault diagnosis of transformers. The results show that the method proposed in this paper has a low false alarm rate and a diagnostic accuracy of 98.125%, which proves that the method can be effectively applied to transformer vibration signal fault diagnosis.

Key words: Transformer, Fault diagnosis, CEEMDAN, Recursive analysis, Convolutional neural network