Modified Subspace Identification Method for Building a Long-Range Prediction Model for Inferential Control

Abstract

In a chemical plant involving a series of processing units, it is beneficial to have a model that can accurately forecast the behavior of downstream variables based on upstream measurements. Such a model can be useful in feedforward and inferential control of the downstream variables to compensate for various upstream disturbances. However, creating such a dynamic model can be very difficult. The conventional multivariable identification approach based on minimizing single-step-ahead prediction error, can result in models leading to poor prediction and control in the described context. To alleviate this difficulty. we propose a modification to the conventional subspace identification method geared towards accurate k-step-ahead prediction, where k is a number chosen according to the estimated dead time. It is shown that the modified subspace identification method call be used in conjunction with the k-step prediction error minimization (PEM) . Using an illustrative examples involving six mixing units with a recycle loop, we demonstrate the improvement that is possible from adopting the suggested modification.