On the use of partial least squares (PLS) and balancing for nonlinear model reduction

Model reduction is an important technique to reduce the complexity of nonlinear process models for controller design. The goal is to approximate the model as accurate as possible while at the same time achieve a speedup in computation time. The technique presented in this paper combines balancing with partial least squares (PLS) for achieving a small, control-relevant, reduced-order model. Two specific methods for using PLS are considered: one is for the balanced residualization for ODE systems and the other is for reducing the complexity of algebraic equations in DAE systems. Corresponding to these two cases, a fixed bed reactor (ODE) and a distillation column model (DAE) are studied to illustrate the use of this balancing/PLS combination for model reduction.