Probabilistic PCA based spatio-temporal multi-modeling for distributed parameter processes

Data-based modeling of unknown distributed parameter systems (DPSs) is very challenging due to their infinite-dimensional, nonlinear and even time-varying dynamics. To get a low-order model for applications, the principal component analysis (PCA) is often used. However, as a linear dimension reduction, it only leads to one set of fixed spatial bases. Therefore a good performance for nonlinear and time-varying DPSs could not be guaranteed. In this study, a probabilistic PCA based spatio-temporal multi-modeling is proposed. Due to its multi-modeling mechanism, a better performance can be achieved, which is demonstrated by simulations.