Research On Clustering Methods Of Multi-Source Partial Discharge Signals

Electrical equipment will collide during production and transportation or will continue to age during long-term use,resulting in partial damage.Non-penetrating breakdown will occur in the damaged part of electrical equipment,resulting in partial discharge.Partial discharge is divided into different types,and different equipment maintenance measures will be adopted for different types of partial discharge.Through the phase resolved partial discharge(PRPD)map,the type of partial discharge signal can be judged.However,when the multi-source partial discharge signal and the interference pulse signal are mixed together,the PRPD pattern will overlap,which makes it impossible to identify the type of partial discharge based on the PRPD pattern.Therefore,studying the clustering analysis method of multi-source partial discharge signals is helpful to separate the overlapping PRPD patterns,which is of great significance for improving the ability of the monitoring system to recognize the types of partial discharges.This thesis analyzes the requirements for cluster analysis of multi-source partial discharge signals,that is,the clustering process requires less human intervention,the number of clusters cannot be determined in advance,the clustering time requirement is low,and clusters of different shapes can be found.With the advantages and disadvantages of classical clustering method,an improved clustering analysis method for multi-source partial discharge signals based on grid division is proposed.Firstly,based on the idea of grid division,a grid space is generated according to the upper and lower scale relationship of the characteristic data set,and the original data set is associated with the grid space.Then,the Gaussian kernel function is used to estimate the local density of each grid,and the relative distance idea of the density peak clustering method is used to generate the decision value,thereby automatically determining the cluster center and the number of clusters.Finally,the non-clustering center grids are classified according to the relative distance relationship and the outliers are processed.Due to the use of grid division ideas,the clustering time is greatly reduced,and the clustering center is automatically determined during the clustering process,thereby effectively improving the clustering effect of partial discharge signals.However,when categorizing nonclustering centers,the "domino" effect that is prone to appear will cause the clustering effect to deteriorate.To this end,this thesis introduces grid weights and grid boundary detection to further improve the multi-source partial discharge signal clustering method to improve the clustering effect.Finally,the clustering method studied in this thesis is verified through experiments.The experimental results show that: compared with the classical clustering method,the clustering method based on grid division studied in this thesis can automatically determine the clustering center.At the same time,the grid division technology used makes the clustering time drop significantly.The clustering method based on weighted grid and grid boundary detection will further improve the quality of clustering at the expense of execution time.The experimental results prove that the clustering method studied in this thesis has important reference value for the clustering analysis of multisource partial discharge signals.