一种基于特征子空间的改进动态核主元分析方法

Improved method of dynamic kernel principal component analysis based on feature subspace

针对传统的动态核主成分分析(dynamic kernel principal component analysis,DKPCA)用于大样本数据集的故障检测时,占用计算机内存大、计算复杂度高等不足,提出一种基于特征子空间的DKPCA算法(EFS-DKPCA)。该方法通过构建具有较小维数特征子空间上的正交基来简化核矩阵K,从而降低DKPCA的计算复杂性。与DKPCA方法相比,该方法具有更高的计算效率,且只需较小的计算机存储空间。将该方法应用于TE(tennessee eastman)过程,仿真结果显示,两者诊断结果大致相同,而所需时间大大减小,说明了本算法的有效性。

For large sample data sets, traditional DKPCA occupancy a lot of computer memory and large computation, in order to solve these problems, this paper proposed an improved DKPCA based on effective feature subspace （EFS-DKPCA）. The new method based on a orthonormal basis of the sub-space spanned by the training samples mapped onto the smaller feature space to simplify K, thereby reducing DKPCA computational complexity. When applied to process monitoring, the EFS-DKPCA-based method was more efficient in computation and needed less computer memory than DKPCA-based methods. Computer simulation of TE process demonstrates the effectiveness and efficiency of the proposed method.