摘要：Kernel principal component analysis (KPCA) has been recently proven to be a powerful dimensionality reduction tool for monitoring nonlinear processes. However, the KPCA based monitoring method suffers from several drawbacks. First, the KPCA method depends strongly on its kernel function, but its selection of kernel parameters is problematic. Second, the underlying manifold structure of the data is not considered in process modelling. To overcome these deficiencies, this paper proposes a new process monitoring technique named extended maximum variance unfolding (EMVU). Because the global and local structures of process data probably change in some abnormal states, global and local graphs are designed to exploit the underlying geometrical structure. The feasibility and validity of the EMVU based process monitoring scheme are investigated through a simple numerical example simulation process. The experimental results demonstrate that the EMVU based nonlinear process monitoring method is a good alternative method to the KPCA-based monitoring method.
关键词：process monitoring; nonlinear; improved kernel principal component analysis; extended maximum variance unfolding; example simulation
