基于自适应仿射传播聚类的GIS局放缺陷分类研究

Pattern Recognition of Partial Discharge in GIS Based on the Adaptive Affinity Propagation Clustering Algorithm

为对GIS中不同缺陷类型进行分类和识别,笔者介绍了一种新型自适应仿射传播聚类(adAP)算法,用于对4种典型缺陷UHF信号进行分类。该算法通过自适应扫描偏向参数空间来搜索聚类个数以寻找最优聚类结果、自适应调整阻尼因子来消除震荡以及自适应逃离震荡等技术,克服了传统仿射传播聚类(AP)算法无法准确获知聚类个数以及无法消除震荡的缺点。最后,通过采用多种聚类有效性指标对分类结果进行有效性验证,发现adAP算法不仅可以准确地给出分类个数,实现不同缺陷类型的分类,而且在处理时间和循环次数上均优于传统AP算法。

An adaptive affinity propagation clustering adAP） algorithm is proposed for pattern recognition of UHF signals from four insulation defects induced partial discharges in GIS. The adAP algorithm searches the number of dusters for finding the optimal clustering solution by adaptive scanning of parameters space, and adaptively adjusts damping factors to eliminate oscillations and adaptively escape from oscillations when the damping adjustment technique fails, thus overcomes two corresponding limitations of the conventional affinity propagation clustering （AP） algorithm. In addition, the clustering results of the proposed adAP algorithm is verified by several clustering validity indexes, indicating its accuracy in the number of clusters and in the classification of different types of defects, and showing better performances in processing time and cycle number than the conventional AP algorithm.