Transient earth voltage (TEV) based partial discharge detection and analysis

Partial discharge (PD) detection is an effective way to evaluate the insulation condition of electrical equipment in power systems. The non-intrusive TEV-detecting method which detects transient earth voltage (TEV) signals from the external surface of equipment does not require interruptions of electrical operations and is thus preferred by more and more researchers, engineers and users. However, as a new technique, TEV based PD measurement is not well developed in many aspects, for example, the measuring system and the de-noising methods. Consequently, the research and development of the TEV based PD measurement has become an interesting topic in recent decades. This thesis presents an investigation on the sensing system and the de-noising methods of TEV based PD measurement system. First of all, the mechanism, popular measuring systems, noise types and existing de-noising methods of PDs are reviewed. Secondly, based on the characteristics of TEV signals, a TEV based PD measuring system was proposed and its effectiveness has been demonstrated by an experimental test. Next, the optimal settings of a popular de-noising method for non-impulsive noise, wavelet thresholding, are selected and its processing efficiency is enhanced by using parallelism algorithm in C environment. Furthermore, the wavelet entropy is proposed to classify PD pulses from impulsive noises. Finally, a noise reduction system using Fourier transform and time-frequency entropy is proposed to reject various kinds of noises. The non-intrusive PD measuring techniques have been more and more popular in recent years. In this thesis, a TEV based PD measuring system is proposed. The major parts: non-intrusive sensor and high-pass filter are designed according to the characteristics of TEV signals. The performance of proposed system is demonstrated by an experimental test where the PD signals are collected by both TEV and HFCT sensors. By considering the features of proposed system, the detected TEV signals are well simulated. Due to the external locations of TEV sensors, the performance of TEV based PD detection is limited by noises. To remove non-impulsive noises, wavelet thresholding is often used. As the de-noised results depend on the settings of algorithm, the optimal ones are selected according to the features of TEV signal and the proposed system. Further, the processing efficiency of wavelet thresholding with optimal settings is enhanced. As wavelet transform is good at time-frequency analysis of PD signals, its capability in rejecting impulsive noises is also explored. Therefore, wavelet entropy is proposed. By comparing with the traditional energy spectrum, the wavelet entropy is more stable to represent a single pulse. With the help of a trained ANN whose parameters are selected carefully, the PD pulses can be extracted with good percentages. The impulsive noise reduction based on features of single pulses is often ineffective when PD pulse and noise occur at the same time. Thus, a de-noising system is proposed to remove both impulsive and non-impulsive noises even if they occur at the same time. In this system, Fourier transform and time frequency entropy are employed. The de-noised results of two experimental signals and one field-collected signal show that the proposed noise-rejecting system is effective in extracting real PD pulses.DOCTOR OF PHILOSOPHY (EEE