Research On Partial Discharge Signal Process Based On Blind Source Separation Algorithm

The partial discharge is a phenomenon of high-voltage electrical equipment in the process of strong electric field, the degree of partial discharge of the device reflects the process and cons of the insulation structure,The generation of partial discharge will have an effect on insulating medium, It can also be bring about Insulation aging, and then eventually leading to insulation breakdown, which will cause to accident. In a word, one of the main reasons for the high-voltage electrical equipment the final insulation breakdown is caused by the partial discharge During the process of line monitoring, Through the study of partial discharge signal, we can determine the discharge location, type and the intensity of the discharge, reflecting the insulation condition and the development trends timely and accurately, avoiding the insulation deterioration caused by Insulation aging as far as possible. However, during the process of line monitoring, will generate a lot of interference signals, which will drown the partial discharge signal, and will have an bad impact on sensitivity and reliability of line monitoring. In order to extract the partial discharge signal from the interference signals completely, we propose the blind source separation algorithm.This thesis proposed the basic model of blind source separation algorithm verify the feasibility of the model is applied to the PD signal extraction. The thesis also analyzes the mathematical model and the principle of blind source algorithms-the three basic principles (based on statistics, information theory, sparse theory),In this thesis, all the algorithms are built on the basis of the three principles above. Because this model is briefly discussed based on the number of source signals and observed signals equal(positive definite case),Therefore, the paper briefly discussed underdetermined blind source algorithms based on sparse theory, and Focuses on the Fast-ICA algorithm(based on the information criterion of the algorithm),and AMUSE algorithm based on the statistic criteria, these two algorithms have a strong representation.The fourth section gives the simulation model and the measured waveform, which combined with performance indicators(Mean Square Error and The Similarity Coefficient),compared the separation performance of the above two algorithms by simulation experiments, the final choice of the Fast-ICA algorithm as the optimal algorithm, and applied to the separation of the measured signal.Because of the periodic narrowband interference signal and white noise signal is the largest source of interference during the process of Discharge Monitoring,This thesis primarily aim to isolate from the narrowband interference signal and white noise signal out-put signal. In the white noise simulation, we found the isolated PD signal has a small amount of a glitch, so we combination of the wavelet analysis method with the Fast-ICA algorithm isolated signal of wavelet threshold denoising method to remove noise, the results show that the combination of the two methods can be better Isolated from the partial discharge signals.This thesis selects MATLAB as simulation environment to build models. This thesis first combine the wavelet analysis method with the Fast-ICA algorithm, And used the algorithms in the extraction of partial discharge signals, and first introduced into the observation model of the multi-channel single-channel observation model (EMD), and applied to the PD signal measured signal separation.