计及源数估计的盲源分离技术在GIS局部放电监测中的应用

Application of Blind Source Separation Technology with Estimate of Source Number in Partial Discharge of GIS

为解决气体绝缘开关设备可能同时存在多个局放源并且源个数未知的实际问题,文中提出在对气体绝缘开关设备中局放源进行源数估计的前提下再对混合信号进行盲源分离的方法。对不考虑传感器噪声和考虑传感器噪声的两类情形,分别利用零空间法与特征值分解法对其进行源数估计,采用估计结果,运用快速独立分量法对信号进行盲源分离。分析研究信噪比对盲源个数估计以及盲源分离效果的影响,为避开盲源分离存在反相与幅值不定的缺陷,运用相关系数法检验分离效果,分离结果说明了盲源个数估计及盲源分离方法在对多局放源激发的混合特高频信号的分离中具有一定的实用性及可行性。

To solve the problem that there can be several partial discharge（PD） sources and the number of sources is unknown in gas insulated switchgear, the paper proposed a method to separate the mixed signals based on the estimate of source number. The zero-space method is used to estimate the source number when there is no sensor noise and eigenvalue-decomposition method is used to estimate the source number when there is sensor noise. Use the FASTICA to separate the blind sources after the source number is estimated. Research the influence of SNR to the results of blind source separation. Correlation coefficients are used to test separating results in order to avoid the presence of blind source separation. The results of the separation indicate that estimate of source number and blind source separation are useful and practical in ultra high frequency（UHF） signals separation.