基于核统计不相关最优鉴别矢量集的GIS局部放电模式识别

GIS局部放电故障诊断对于准确掌握GIS内部的缺陷性质和指导GIS的检修工作有着重要意义.针对线性Fisher鉴别分析用于局部放电故障诊断时存在的问题,文中借鉴核方法思想,提出了一种基于核的统计不相关鉴别矢量集算法(KSUODV),用以解决高维特征空间内的非线性特征提取问题,并且消除了变换后样本特征之间的统计相关性.在对实验室获取的7种缺陷PD三维谱图模式识别试验表明,KSUODV算法的识别性能优于SUODV算法性能,效果良好.

PD Pattern Recognition Based on Kernel Statistical Uncorrelated Optimum Discriminant Vectors in GIS

The diagnosis of gas insulated substation (GIS) partial discharge (PD) is of vital importance to accurately grasp the nature of the defects within the GIS and guide its maintenance. As to the problems of linear Fisher analysis for PD diagnosis, based on kernel method, a Kernel statistical uncorrelated optimum discriminant vectors algorithm (KSUODV) is proposed to solve the problem of non-linear feature extraction in High-dimensional feature space, and has eliminated statistical correlation between transformed sample characteristics. The recognition results of the PD 3-dimen-sional discharge patterns of seven defects obtained in laboratory prove that KSUODV algorithm for identification is better than SUODV algorithm.