A unified approach to linear design and predictive control of constrained systems

This thesis studies the use of model predictive control (MPC) for linear systems from a design perspective. Our focus is on establishing a unified approach of linear design methods and MPC so that the benefits of the two can be obtained in a well-defined way for practical applications. For this purpose, we have considered the design question from both the control and communication point of view. On the control side, starting from a basic assumption that an unconstrained pre-stabilizing controller is available, we have considered several closely-related issues. Specifically, we have proposed novel tuning techniques so that an MPC controller can either replace an existing controller or gradually improve the control performance based on the latter. We have presented a detailed stability proof for these techniques. In this thesis we have also discussed the role of the observer in robust MPC. As such, we have considered systems with unstructured uncertainty and presented some design methods of robust policies. We have shown via theoretical analysis and numerical simulation that the choice of the observer makes a key difference in the resulting closed-loop performance. On the communication side, we have presented a sparse communication strategy for networked control systems (NCS) based on the singular value decomposition (SVD) of the Hessian of the quadratic performance index generally considered in MPC and the unconstrained optimal controller. The singular vectors are employed to generate an orthonormal basis function expansion of the unconstrained solution to the infinite horizon optimal control problem. The proposed control law is deduced from the former unconstrained controller based on cost reduction consideration. We have presented a thorough study of the associated stability analysis and have shown the advantages of the proposed method via simulation studies.