Study On Gis Partial Discharge Type Recognition Method Based On The Non-dispersed Infrared

Gas Insulated Switchgear(GIS) has been widely used in the power system because of its small footprint, high reliability and long maintenance cycle.But some routine prevention testings are difficult to implement effectively in GIS because of its exquisite manufacturing and installation process, complex structure and closed tank. Therefore seeking effective GIS partial discharge detection and pattern recognition method has an important value to assure safe operating of GIS, master partical discharge fault characteristics of GIS and insulation state of equipment, as well as arrange the repair work of GIS.The main contents of this paper are as follows:1. Set up the experimental platform. According to the actual GIS equipment strucuture and material, this paper has designed four kinds of typical partial discharge physical model: free particle, insulator bubbles, metal protrusions and insulator dirt. The electric field simulation analysis of typical fault type is opposed using ANSYS. The feasibility of the model is assessed.2. Partial discharge test. Electrochemical sensor method, gas chromatoaraphy and the infrared spectroscopic are analysised through the contrast experiment. The infrared spectroscopic is used to detect the level of SF6 decomposition at different fault type and different fault time, this lays a data foundation for further fault feature extraction and fault type identification.3. Extract the data characteristics. The data characteristics of various faule types are extracted using curve fitting method. The difference of gas content ratio and gas content fitted curve is contrastly analysed. The extracted data characteristics are verified using clustering algorithm and the physical meaning of gas content ratio is proposed from the definition perspective of gas content ratio.4. Fault type identification. The bionic pattern recognition algorithm is introduced to identify GIS fault types and a method to change the connecting direction is put forward to improve the recognition rate of the algorithmic. The recognition rate before and after changing the connecting direction is analyzed. The identification results of SVM algorithm are analyzed in comparison to the above result.