Skip to Main Content
The method of proper orthogonal decomposition (POD) has been proven to be very useful for constructing low dimensional models of large scale systems. However, despite the model order reduction, low-order models derived from truncations of POD bases remain computationally intensive for the simulation of large scale linear time-varying (LTV) and nonlinear models. The main bottleneck lies in the requirement to have full spatial information from the original model to construct the reduced-order models. In this paper, we propose criteria to select a suitable subset of the original spatial coordinate system using information from the snapshot matrix and the POD basis functions. We show that the states of the POD-based reduced order model can be estimated much more efficiently by conducting projections on these selected states. The method is applied to a representative industrial model of a glass feeder.