Study On Separation Of Mixing Partial Discharge Signals Produced By Multipel Insulation Defects Model In GIS

Gas isolated switchgear (GIS) with low failure rate and long inspection span serves the safe and stable operation of modern power grid. Partial discharge (PD) is a common precursor to the potential insulation defects in GIS cylinder and the identification of external ultra-high frequency (UHF) PD signal has developed to be an most immediate and effective approach to insulation defects inside GIS. It is possible that the special space architecture in GIS and PD mechanism together results in the multiple insulation defects in GIS cylinder. This paper gets rid of the unsound knowledge that the only single insulation defect excites single PD signal and considers the status on multiple insulation defects in GIS.It is sure that multiple insulation defects ignite mixing PD signals. Directly through these mixing signals to acquire the information on insulation defects in GIS is extremely of challenging and front. Blind source separation (BSS) theory disconsiders the information from source signals and mixing process to acquire the knowledge of source signals with the help of separating several mixing signals. The characteristic of"black box"indirectly for PD knowledge is extremely similar to that of obtaining the information (number and type) on multiple insulation defects in GIS from external UHF mixing PD signals. Therefore, this paper introduces the BSS theory into research on multiple insulation defects in GIS for the first time.Directly denoising PD signal probably makes the expected denoising effects unsatisfying due to over-denoising or under-denoising and inappropriate denoising method, and even causes the unexpected distortion of PD waveform and the loss of its characteristic information. Hence, this idea on firstly evaluating the interference degree of PD signal in the quantitative and then denoising it is proposed in this paper and the rule is built for the signal-to-noise ratio (SNR) estimation suitable for PD signal to quantitatitively evaluate the interference degree on actual PD signal on the basis of 2- order statistic SNR estimation theory.Due to uncertain positions of insulation defects under the status of multiple insulation defects in GIS, it is hard to build up an accurate mathematical model for mixing process of PD signals in GIS cylinder. For theoretical analysis to the separation of mixing PD signals, the simplified linear & instantaneous mixing model and linear & convolutive mixing model are employed, respectively. Furthermore, different BSS algorithms are described for different models. In experimental GIS model, four physical models for typical insulation defects are constructed for actual PD source signals. Two of them for a group, 4 groups of simulating mixing PD signals are designed. Similarly, 4 groups of models of multiple insulation defects are arranged in this GIS for actual mixing UHF PD signals. Both of them are for the separation and comparative analysis of mixing PD signals.2-oder statistic blind identification separation algorithm SOBI and its weights-adjusted algorithm WASOBI for linear & instantaneous mixing model are employed to separate these mixing PD signals including simulating and real ones. Their separation effects are quantitatively evaluated by several evaluation parameters so that the performances for SOBI and WASOBI are compared. On the other hand, critical influence factors on separation effect are discussed, including the types of insulation defects, the distance between two insulation defects and the mixing process in GIS.For linear & convolutive mixing model, actual mixing PD signals with long time and non- stationary character are partitioned into a series of short time and stationary mixing PD signals in frequency domain space by virtue of windowed Fourier transform. Correlation decomposition algorithm is used to separate these subsets of statinary PD signals. Taking advantage of the correlation of envelopes of these separated PD signals, they are accurately reconstructed and then are transformed into single PD signals in time domain space by inverse Fourier transform. Mixing PD signals from simulative mixing process and GIS model are employed to demosnstrate separation experiment and comparative analysises are done. The characteristic of linear & convolutive mixing model for actual PD signals are stuyed and then the hypothesis of convolutive mixing process on actual GIS inside is validted, which paves a way to the further research on multiple insulation defects in GIS.