Application of hilbert-huang transform in pattern recognition for partial discharge of transformers

Time-frequency analysis is an efficient method in feature extraction for transformer partial discharge. The shortcomings of wavelet transform were discussed firstly as approaches to analyze non-stationary signals; then the theoretical framework of Hilbert-Huang transform was introduced; finally, the time-frequency spectrums were obtained through the two methods mentioned above and time-frequency entropy vectors were extracted from them as the features for pattern recognition. The result of fuzzy clustering shows that the clustering characteristic of HHT entropy vectors are better than those of wavelet transform, so HHT is a more efficient way to extract features of partial discharge pulses for transformers.