关键词: 变压器, 故障诊断, 深度学习, 多重降噪, 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:

To address the challenges of extracting features from transformer vibration signals under noise interference conditions and the low utilization of temporal information，a fault diagnosis method combining multi-level denoising and recurrence plots is proposed. Multi-level denoising of the vibration signals was achieved through complete ensemble empirical mode decomposition with adaptive noise（CEEMDAN）combined with adaptive wavelet thresholding. Recurrence analysis was then used to transform the time-series signals into recurrence plots，preserving temporal dependencies and nonlinear dynamic features. A convolutional neural network（CNN）was employed to leverage its strengths in image processing，using the recurrence plots as inputs to extract spatial features，thereby enabling transformer fault diagnosis. The results show that the proposed method achieves a low false alarm rate and a diagnostic accuracy of 98.125%，demonstrating its effectiveness for diagnosing transformer faults based on vibration signals.

Key words: transformer, fault diagnosis, deep learning, multi-level denoising, complete ensemble empirical mode decomposition with adaptive noise, recurrence analysis, convolutional neural network