Skip to Main Content
This paper presents a new recognition method for ultra-high-frequency (UHF) signal radiated by partial discharge (PD) occurred in transformers, based on wavelet transform and fractal theory. Wavelet transform provides an effective way to decompose a signal on unlimited scales. During this process, the wavelet coefficients corresponding to each scale are obtained, which could be considered as an accurate expression of the decomposed signal. However, the wavelet coefficients as signal features are inavailable for UHF signal recognition because of massive data generated by UHF detection. Fortunately, based on fractal theory, fractal dimensions of the wavelet coefficients, as the compressed features of UHF signal, are calculated by differential boxing-count estimation (DBC) to recognize different PD activities. To testify the effectiveness of our method, an experiment was carried out in laboratory. In experiment, four types of artificial defects models are constructed to generate UHF PD sample data. And A 3rd Hilbert fractal antenna with compact size, which performs well in the properties of radiating pattern and VSWR, is designed to detect signals. The features extracted from the PD data are classified by a radial basis function neural network (RBFNN). The recognition results convince that the recognition method, combining the knowledge of wavelet transform and fractal dimension estimation, is qualified to apply in the field of UHF detection pattern recognition.