Adaptive multivariable process controller using model switching and attribute interpolation

Abstract

An adaptive multivariable process control system includes a multivariable process controller, such as a model predictive controller, having a multivariable process model characterized as a set of two or more single-input, single-output (SISO) models and an adaptation system which adapts the multivariable process model. The adaptation system detects changes in process inputs sufficient to start an adaptation cycle and, when such changes are detected, collects process input and output data needed to perform model adaptation. The adaptation system next determines a subset of the SISO models within the multivariable process model which are to be adapted, based on, for example, a determination of which process inputs are most correlated with the error between the actual (measured) process output and the process output developed by the multivariable process model. The adaptation system then performs standard or known model switching and parameter interpolation techniques to adapt each of the selected SISO models. After the adaptation of one or more of the SISO models, the resulting multivariable process model is validated by determining if the adapted multivariable process model has lower modeling error than the current multivariable process model. If so, the adapted multivariable process model is used in the multivariable controller.