Skip to content
Conference Open access

Research on Data-Driven Signal Processing Methods for Wind Turbine Circuit Breakers

Sep 2026 · 2026 IEEE 7th International Conference on Advanced Electrical and Energy Systems (AEES) · pp. 849-854 · 0 citations · 5 references

Abstract

The opening and closing characteristic curves of wind turbine circuit breakers are susceptible to electromagnetic interference and sensor noise during field measurement, while fault samples are extremely scarce in actual operation, posing significant challenges for data-driven fault diagnosis modeling. To address these issues, this paper proposes a joint processing method integrating signal denoising, feature selection, and data augmentation. First, four denoising methods-moving average, Butterworth low-pass filtering, wavelet threshold denoising, and EMD-wavelet combined denoising-are compared on the original displacement signal, and wavelet threshold denoising with the Bayes criterion is selected as the preprocessing method for its optimal signal-to-noise ratio. Second, three feature selection approaches-Pearson correlation coefficient, maximal information coefficient, and RFECV-are jointly applied to screen the candidate features, reducing the feature dimension from 12 to 4 while shortening the training time by 86% with no significant loss in classification accuracy. Finally, a cWGAN-GP generative adversarial network is adopted to augment the minority-class fault samples, overcoming the limitation of traditional SMOTE oversampling, wherein generated samples are confined to the convex hull of the original data due to linear interpolation. Experimental results demonstrate that wavelet threshold denoising within the proposed joint method achieves a signal-to-noise ratio of 46.87 dB. After cWGAN-GP data augmentation, the classification accuracy improves from 93.46% on the original imbalanced dataset to over 99%, and the generated fault samples outperform those produced by SMOTE in terms of both KL divergence and the preservation of inter-feature correlation structure.

Read PDF

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.