The Failure Recognition Research Of GIS Partial Discharge Signal Measured By The Pulse Current Method

By measuring the Partial Discharge of the gas-insulated substation, the inherent defects of GIS can be discovered and the major accident can be avoided. At present GIS requires to do the PD test in the type test,factory test and field test. According to the measured PD signal, the type of the discharge defects can be identified and it has great significance to the production,running and maintenance of GIS.In the present methods of measuring the GIS PD signal, the pulse current method is the only method of quantitative measurement in the existing standards, and it is applied in the GIS type tes,factory test and field test. After measuring the PD signal, the test personnel conduct the failure analysis mainly by the experience and there is great subjectivity, so it can not get a good recognition result. Although there is much research about the pattern recognition of PD signal, but there is not much research about the pattern recognition of GIS PD signal measured by the pulse current method. This paper studies the feature extraction and the design of pattern recognition device of GIS PD signal measured by the pulse current method systematically. It includes the statistical operator of two-dimensional spectra and the fractal characteristics of grayscale for the feature extraction and the two types of the pattern recognition device basing on the BP network and the Support Vector Machines. Finally, combine two features with two pattern recognition devices one-on-one, and it will get four types of pattern recognition program.In order to test the effect of four types of pattern recognition program, this paper conducts the PD test of four typical defects in the GIS fault simulation device and gets a large number of PD test data. Deal with the test data in accordance with the four pattern recognition program. Experimental results show that one of the four pattern recognition program is effective and feasible, and the program uses the fractal characteristics of grayscale for input and Support Vector Machines for GIS PD pattern recognition device. The program has a lot of advantages such as lower input,less training time,better network stability and higher rate of recognition accuracy.