Study of GIS PD Defect Classification Based on Adaptive Affinity Propagation Clustering
Da Peng Duan, Wei Qiang Qi, Hong Jing Liu, Xu Cheng, Yan Ran Li, Xiao Xin Chen, Hai Feng Ye · Applied Mechanics and Materials · 2013
In order to classify and identify different types of defects in GISs, this paper introduces a novel adaptive affinity propagation clustering (adAP) algorithm, which is used for classification of UHF signals of four kinds of typical defects. The algorithm has overcome the conventional affinity propagation (AP) algorithms defects, namely inability to acquire cluster number accurately and inability to cease oscillations, by technologies of adaptively scanning preference space to search for the number of clusters to find the optimal clustering results, adaptively adjusting the damping factor to eliminate oscillation and adaptively escaping oscillations. Finally, validity verification is performed over classification results using a variety of validity indexes. It is discovered that adAP algorithm can accurately tell the number of classes and realize classification of different types of defects. It is also superior to conventional AP algorithm in regards of the processing time and the number of loops.