Application issues of state space model predictive control and formulation for robust model predictive control

This dissertation focuses on two issues in Model Predictive Control (MPC): application and robust control formulation. The first objective of this dissertation is to provide control engineers with a better understanding of the State Estimation based Model Predictive Control (SEMPC) technique and some tuning guidelines for robustness. By reviewing some major developmental trends in MPC research, this dissertation focuses on the state space formulation. Different state estimation scenarios based on the available disturbance models to improve the quality of the output prediction are investigated. In order to provide practical on/off-line tuning rules of the SEMPC controllers for industrial applications, this dissertation identifies a set of on-line tunable parameters of the SEMPC controllers that have direct, independent, and well understood effects on the closed-loop response speed and robustness.;To demonstrate the advantages of applying SEMPC to solve challenging chemical process control problems, a benchmark process control problem, the Shell Control Problem (SCP), is used as a case study. The SCP embodies most of the critical elements of challenging industrial process control problems (e.g. unmeasured disturbances, model uncertainty, input/output constraints, optimization objective conflicting with control requirements, failure-prone sensors, non-square system, etc.) and therefore serves as a good test problem for investigation of potential benefits and pitfalls of the new technique. We demonstrate in the case study that, while the theory for the SEMPC technique is rigorously laid out, it is nontrivial for practicing engineers to correctly formulate practical objectives within the theoretical framework. By formulating and analyzing a series of different SEMPC controller designs for the SCP, this dissertation highlights some of the possible difficulties that engineers may encounter in applying SEMPC to practical control problems and shows how these difficulties can be efficiently overcome.;The second objective of this dissertation is to explicitly address deterministic bounded parametric model uncertainty in MPC formulation. A mini-max algorithm that minimizes the predicted worst-case closed-loop error is formulated as a Dynamic Programming problem. It is proven that, in the limiting case of an infinite prediction horizon, the proposed robust MPC algorithm guarantees to give rise to asymptotically closed-loop stable result for the entire model set. The Dynamic Programming formulation provides insights into the essence of the problem. However, this rigorous robust MPC algorithm suffers from numerical problems and cannot be implemented in practice. Suboptimal yet computationally efficient robust MPC algorithms are studied. A convex model structure, FIR model, is used to develop suboptimal algorithms so that the robust MPC algorithms can be applied in practice.