Research On Mixing Characteristic And Blind Separation Of GIS Partial Discharge Signal

With the widespread use of GIS in power system, the online monitoring and diagnosing of partial discharge(PD) have attracted much attention of the scholars both at home and abroad. And a lot of research works are carried out concerning the fault location, propagation properties and pattern recognition of PD signal, and put forward many techniques and method, but few technique and method relate to mixed PD signals have been published. In practice, however, the pattern recognition of single PD signal is limited because of the existence feasibility of mixed PD signal generated by multiple PD sources. Compared with single PD signal, mixed PD signal is a better description of the fault signal. And the mixed PD signal recognition not only can be used to recognize the type of single PD signal, but also can be employed in recognizing mixed PD signal. On the other hands, it can be tried as the upgrade and alternatives of single PD signal recognition and its research and application foreground are very clear. For this, this paper study the pattern recognition strategy of mixed UHF PD signal, based on the research of UHF PD signal’ mixing mode and the blind source separation. And the efficient variant of algorithm FastICA (fast independent component analysis) has been implemented in this paper to acquire the single signals from the mixed UHF PD signal and its separation performance has been tested. The main contributions of this thesis are as follows.On the basis of extensive literatures, this paper summarizes and analyses the research results of the scholars both at home and abroad with regard to online monitoring and recognition of UHF PD signals. According to the distinctiveness of mixed UHF PD signal, this thesis analyzes the pattern recognition strategy of mixed UHF PD signal from two aspects. That is research directions, data-processing algorithms. And eventually comes to the conclusion that the viable scheme is to transform mixed signal into single signal based on blind source separation (BSS) algorithms.In order to explore the statistics and mixing feature of UHF PD signal, experimental signals and the XFDTD electromagnetic simulation software have been employed for research. Both of the theoretical analysis and simulation results show that UHF PD signal is kind of non-Gaussian signals which tending to Gaussian distribution. And convolution is the basic relationship between PD current signal and the single UHF PD signal; the mixed UHF PD signals are the linear and instantaneous mixture of single UHF PD signals, and are the linear and convolution mixture of PD current signals.According to the mixing feature of UHF PD signal and the using conditional of BSS methods, the pattern recognition strategy of mixed UHF PD signal is analyzed and summarized here, and the feasible way to identify mixed UHF PD signal are pointed out as follows. Firstly, implement blind de-convolution to get PD current signals from mixed UHF PD signals, and then identify the mode of PD current signals. Secondly, process the mixed UHF PD signal with independent component analysis (ICA) algorithms to obtain UHF PD signals, and then identify the mode of mixed signal by identifying the single signal. On the other hand, the recognition of UHF PD signals has been studied by researchers at home and abroad, as well as a large number of research achievements is obtained. So the second pattern recognition strategy has better application prospects.Finally, EFICA algorithm has been employed in this thesis to estimate UHF PD signals from the mixed signals. And the performance of EFICA in this separation of PD signals has been tested in this paper. The test result shows that the EFICA method has good separating effect, low noise susceptibility and strong robustness. In another words, EFICA method has a very optimistic prospect of application.