基于改进结构保持数据降维方法的故障诊断研究

传统基于核主成分分析(Kernel principal component analysis, KPCA)的数据降维方法在提取有效特征信息时只考虑全局结构保持而未考虑样本间的局部近邻结构保持问题, 本文提出一种改进全局结构保持算法的特征提取与降维方法.改进的特征提取与降维方法将流形学习中核局部保持投影(Kernel locality preserving projection, KLPP)的思想融入核主成分分析的目标函数中, 使样本投影后的特征空间不仅保持原始样本空间的整体结构, 还保持样本空间相似的局部近邻结构, 包含更丰富的特征信息.上述方法通过同时进行的正交化处理可避免局部子空间结构发生失真, 并能够直观显示出低维结果, 将低维数据输入最近邻分类器, 以识别率和聚类分析结果作为衡量指标, 同时将所提方法应用于故障诊断中.使用AVL Boost软件模拟的柴油机故障数据和田纳西(Tennessee Eastman, TE)化工数据仿真, 验证了所提方法的有效性.

Research on Fault Diagnosis of Data Dimension Reduction Based on Improved Structure Preserving Algorithm

The traditional data reduction method based on kernel principal component analysis(KPCA)only considers the global structure preservation when extracting effective feature information,but does not take the problem of local neighbor structure retention between samples into consider.An improved feature extraction and dimension reduction of global structure preservation algorithm is proposed which integrates the idea of kernel locality preserving projection(KLPP)of manifold learning into the objective function of kernel principal component analysis,so that the feature space after the sample projection not only remains the whole original sample space.But also maintains a local neighbor structure with similar sample space which contains more feature information.Distortion of the local subspace structure can be avoided by simultaneous orthogonalization,and the low-dimensional results can be visually displayed.The low-dimensional data is inputed into the nearest neighbor classifier,using the recognition rate and cluster analysis results as a measurement.At the same time,the proposed method is applied to fault diagnosis.The diesel engine fault data simulation simulated by AVL Boost software and Tennessee Eastman(TE)chemical data simulation verify the effectiveness of the proposed algorithm.