关键词: 核主成分分析, 局部切空间排列, 数据降维, 故障分类

Abstract: In order to solve the problem that kernel principal component analysis (KPCA) and local tangent space alignment (LTSA) algorithm can not keep both the global and local structure characteristics of data in the process of dimensionality reduction, this paper uses linear superposition of kernel function to derive a nonlinear dimensionality reduction algorithm which combines KPCA algorithm with LTSA algorithm. The dimensionality reduction algorithm can make a fault dataset maintaining both global distance relationship and local neighborhood relationship between data samples after dimensionality reduction. Computational experiments show that this algorithm can accurately extract the global and local structural characteristics contained in the fault dataset, and make the results of fault classification clearer and more accurate as well as more effective.

Key words: kernel principal component analysis, local tangent space alignment, data dimensionality reduction, fault classification