Model Reduction for Large Scale Systems with Application to a Pulp Digester

Abstract

Existing model reduction algorithms tend to be ill-conditioned for process systems with larger than 300 states. An approach for the reduction of high-state dimension systems with block diagonal structure is introduced. The block diagonal model structure is common for chemical process models in which numerical approximations are used to model distributed parameter behavior. The procedure is applied to a linearized continuous pulp digester model with 1050 states: an accurate reduced-order model with 54 states is obtained in less than 3 minutes. The procedure is also modified to use a set of transformation matrices, obtained at a nominal operating point, to reduce the model in less than 5 seconds. With this development, the utilization of high-order, nonlinear process models in nonlinear model predictive control is feasible.