摘要： 基于二维特征矩阵的二维特征融合(2DFF)方法——二维主成分分析法能够降低特征矩阵的维数,达到特征融合的目的,但该方法仅在特征向量维数相近的情况下效果较好。传统2DFF特征矩阵构造方法需要在每个特征向量后补0以形成二维特征矩阵,在特征向量维数相差较大时补0个数较多,破坏原始特征向量属性,使最终识别率降低。针对该问题,提出一种基于奇异值分解(SVD)的二维特征矩阵构造方法,该方法将所有特征向量首尾相接组合成一维特征向量,利用SVD的分解特性,在保持特征信号相位不变的情况下,将一维综合特征向量分解成二维特征矩阵,避免大量补0导致信号特性的改变。实验结果表明,该方法在各特征向量维数相差较大的情况下,可获得比在向量后直接补0的特征矩阵构造方法更高的识别率。

关键词: 状态识别, 高速列车故障诊断, 特征融合, 二维特征矩阵, 主成分分析

Abstract: The Two-dimensional Feature Fusion(2DFF) method based on two-dimensional feature matrix,i.e.,two-dimensional principal component analysis can the goal of feature fusion by decreasing the dimensions of the feature matrix,but it performs well got only when the difference in the dimensions of feature vectors is small.Some zeros after every single feature vector to get a two-dimensional feature matrix in the construction method of feature matrix of traditional 2DFF,which may change attributes of original feature vector at the condition that the difference in dimension of each feature vectors is huge and decreases the identification rate.Since the disadvantage above,a new construction method of feature matrix based on Singular Value Decomposition(SVD) is proposed.The new method groups all feature vectors end to end as a new one-dimensional feature vector which is decomposed into a two-dimensional feature matrix by keeping the phase of the signal unchanged based on the decomposition feature of SVD.Experimental result shows that the new method has a higher identification rate than traditional 2DFF feature construction method difference in the dimensions of feature vectors.

Key words: state identification, fault diagnosis of high-speed train, feature fusion, two-dimensional feature matrix, Principal Component Analysis(PCA)

中图分类号: 

• TP391