Performance enhancement of model predictive control through robust design and closed-loop identification

Model Predictive Control (MPC) is an advanced process control strategy that is commonplace in today's computer aided process control technology. The general philosophy underlying predictive control is that by embedding process knowledge into the controller, improved performance can be attained. The significant reasons for the success of MPC in the process industry especially for multivariable systems include the presence of constraints, process non-linearities, model uncertainty, and unique performance criteria to name a few.; Since process knowledge is embedded in the controller in the form of a process model, it is imperative that the controller design take into account inaccuracies of the process model. In this work, we develop and study methods to account for these inaccuracies, and to design the controller in a flexible manner so as to optimize the performance of the closed-loop system. Studied first are the implications of prior developed robust design conditions on the stability and performance of multivariable systems. A tuning methodology that guarantees robust stability and the best possible performance under robust stability conditions is presented and studied. Following this, a new approach to the identification of finite impulse response (FIR) models (typically used by MPC schemes) is provided. This novel approach relies on identifying a compressed set of FIR coefficients thereby eliminating the effects of data overfitting. The FIR model compression is achieved via the discrete wavelet transform of the FIR. The efficacy of this approach is demonstrated through comparison with existing schemes through simulation studies. Finally, a new algorithm for simultaneous MPC and identification (MPCI) is presented. This algorithm is a new formulation of the original version of MPCI. In this formulation we use a frequency domain persistent of excitation condition on the inputs. This approach when compared with the original form of MPCI, is shown to have superior behavior in terms of computational speed. This aspect makes it extremely attractive from the industrial viewpoint of practicality.; Presented herein is work that is believed to be an essential building block of the plant-wide process control structure that the process industry will be witness to in the near future.