Clustering Separation Method Based on Multi-Source Partial Discharge Signal Data Stream

Abstract

In partial discharge(PD) detection, due to the simultaneous and constantly changing phenomenon of multiple discharge sources and on-site interference sources, it is difficult to effectively separate and identify multiple PD sources. An efficient adaptive efficient adaptive online data stream clustering algorithm (EAOStream) is proposed. The algorithm uses natural neighborhoods to create K-dimensional (KD) trees to improve the efficiency of querying neighbors. That is, the adaptive neighborhood radius and the area density are obtained through the characteristics of the flow data, which can search locally and form clusters, and realize the real-time online separation of multiple local discharge sources. The superiority of EAOStream is verified in the artificial data set and the real data set. After comparing EAOStream with the traditional DenStream and SE-Stream algorithms, it is applied to the pattern recognition of gas-insulated substation faults. Experimental test results show that the clustering accuracy of EAOStream in the real network intrusion detection, the forest cover type, and the multi-source PD signal data sets reaches 95.28%, 98.47%, and 97.23%, verifying the practicability and effectiveness of the algorithm in fault diagnosis of gas-insulated substations.

Keywords

• data stream
• cluster separation
• local adaptation
• natural neighborhood
• partial discharge (pd)