Research Of GIS Faults Classification And Location Methods Based On Biomimetic Pattern Recognition Technology

With the expansion of the power grid, Gas Insulated Switchgear-GIS internal faults increases as well. The Partial Discharge-PD signals are generally used to monitor GIS operation condition by Ultra High Frequency-UHF sensor. At present, PD signal feature extraction methods are concentrated in Wavelet Transform-WT, Fractal theory and so on, but these methods can only remove the noise but not extract detail characters of PD signals. Subsistent GIS fault classification methods are effective for single PD fault but not so well for multiple PD faults. Seriously disturbed by interference signals, GIS fault location technology can’t get detail information of PD signals accurately, as a result the location accuracy is poor, and there are anchor points could not be covered.According to the defect of current GIS fault classification and location technologies, following problems are studied based on4kinds of GIS PD model:Firstly, the improved rapid Matching Pursuit-MP algorithm is put forward to extract the features of GIS PD signals. Wavelets in wavelet dictionary are deposited in a matrix. Then this matrix is turned around and permanent deposited in matrix W. Numerous times of matching inner product operation are converted to a convolution computation of PD signal and wavelet in tight interval. This method improves the accuracy of time parameters of the optimal matching wavelet. The MATALB simulations show the superiority in speed of the improved MP feature extraction algorithm. Based on the improved MP algorithm the feature of multiple PD fault can be extracted.Secondly, the improved Biomimetic Pattern Recognition-BPR method is presented to classify4typical faults of GIS. In this method, the super high dimension fluid is built unrelated to the sequence of training samples. The concept of impression space is introduced into BPR method which is used to solve the problems of space overlap between species and space vacancy of some specie in super high dimension space. Comparing to the BPR methods based on super high dimension sausage and super high dimension simplex, the MATLAB simulations show that the improved BPR method displays a much better fault classification accuracy. The multiple PD fault is classified accurately by this improved BPR method.Finally, the time parameters extracted by the improved rapid MP algorithm are used to calculate the time difference in Time Difference of Arrival-DTOA location technology. The MATALB simulations show that this method gains a very good precision. The4test nodes of DTOA by orthogonal center axis hyperbola are enriched, which makes the locating points covering the whole detecting space.