Kernel principal component analysis with reduced complexity for nonlinear dynamic process monitoring

This paper proposes a new reduced kernel method for monitoring nonlinear dynamic systems on reproducing kernel Hilbert space (RKHS). Here, the proposed method is a concatenation of two techniques proposed in our previous studies, the reduced kernel principal component (RKPCA) Taouali et al. (Int J Adv Manuf Technol, 2015) and the singular value decomposition-kernel principal component (SVD-KPCA) (Elaissi et al. (ISA Trans, 52(1), 96–104, 2013)) The proposed method is entitled SVD-RKPCA. It consists at first to identify an implicit RKPCA model, that approaches “properly” the system behavior, and after that to update this RKPCA model by SVD of an incremented and decremented kernel matrix using a moving data window. The proposed SVD-RKPCA has been applied successfully for monitoring of a continuous stirred tank reactor (CSTR) as well as a Tennessee Eastman process (TEP).