Research On On-line Intelligent Recognition Of GIS Partial Discharge Based On UHF Method

With the acceleration of China's industrialization process,UHV(Ultra High Voltage)power grids have developed rapidly.GIS(Gas Insulated Switchgear)is widely used in high-voltage substations due to its small footprint,long maintenance intervals,and low electromagnetic hazards.Due to many uncontrollable factors during production,installation,transportation and long-term operation,partial discharge of GIS equipment often occurs.Therefore,timely diagnosis of partial discharge faults of GIS equipment is of great significance to the safe and stable operation of power systems.The main research content of this article is to identify the UHF(Ultra High Frequency)signal generated by partial discharge of GIS equipment.First,the generation mechanism of UHF signals in GIS is described,the propagation characteristics of UHF signals are analyzed,and the common fault types when partial discharge occurs are summarized.A partial discharge detection device is designed based on the characteristics of the UHF signal during partial discharge.The device consists of an UHF sensor,a data acquisition device,and a centralized unit.The UHF signal is collected and transformed by the detection device and transmitted to the computer,where the real-time data is intelligently analyzed.Based on statistical features,time-domain features and frequency-domain features,22 sets of feature parameters were extracted.Principal component analysis method is used to calculate the covariance matrix of feature parameters and solve their eigenvalues.The order of principal components is determined by the cumulative contribution rate,and dimensionality reduction data can be obtained.Using dimensionality reduction data as input samples,BP neural network is used to identify faults,and the optimal structure of the network is determined one by one.However,the performance of BP neural network is easily affected by weights and thresholds.Therefore,genetic algorithms are used to optimize the weights and thresholds of BP neural networks.From the experimental analysis,it can be seen that the optimized algorithm has higher recognition accuracy than the original algorithm,and the average failure recognition rate is 91.5%.The sample data verifies that the method can effectively identify partial discharge faults of GIS equipment.