Study On Ultra-High-Frequency Method For Monitoring Partial Discharge In Power Transformers

Insulation deterioration, which is mainly caused by partial discharge (PD) occurring inside power transformers, is one of the prime reasons to cause transformer faults. In order to prevent accident faults and insure stable performance of power system, it is valuable to judge accurately the condition of transformers through on-line monitoring PD activities of transformers.Ultra-high-frequency (UHF) monitoring approach is focused on recent years because of its effectiveness to avoid low-frequency noises. However, there are still unsolved problems to obstacle the on-site application of UHF on-line monitoring system for PD in transformers. In this paper, UHF on-line monitoring approach for PD in transformers is studied on the basis of concluding and analyzing the research situation of on-line monitoring for PD activities. This paper concentrates on three aspects: the optimized design of UHF sensor (UHF antenna), interference suppression of UHF signal and the recognition of UHF signal, all of which are shown below.â—‹1 The basic principles of Hilbert fractal antenna is introduced and the optimization and design approaches of Hilbert fractal antenna are presented for UHF on-line monitoring for PD in transformers based on fractal and antenna magnetic theories. The performance of Hilbert fractal antenna is discussed and the influence of geometry parameters to the performance of the antenna is studied based on the magnetic simulation software Ansoft Designer; a 3rd Hilbert fractal antenna is designed taking the structure of transformers into account.â—‹2 An improved wavelet denoising method is presented to suppress the white noise mixed within the UHF signal generated by PD activities. The optimal basic wavelet is calculated in each scale through analyzing the influence of basic wavelet to signal energy decomposed in each scale. The method is applied to denoise UHF signals generated by four types of classic artificial insulation defects and the denoising results derived from different thresholds and threshold methods are also compared.â—‹3 The difference box-counting method for fractal dimension is studied to extract fractal features from wavelet coefficients of UHF signal generated by PD. According to the extracted feature, radial basis function artificial neuron network and probability artificial neuron network, as pattern classifier, respectively, is used to recognize four types of UHF signals derived from artificial defects and the recognition results of two ANNs are compared.