Improving Tracking in Optimal Model Predictive Control

The thesis deals with the improvement in the tracking in model predictive control(MPC). The main motivation is to explore high embedding performance controllers with constraint handling capabilities in a simple fashion. There are several techniques available for effectively using an infinite horizon rather than a finite horizon. First, there has been relatively little discussion so far on how to make effective use of advance information on target changes in the predictive control literature. While earlier work has indicated that the default solutions from conventional algorithms are often poor, very few alternatives have been proposed. This thesis demonstrates the impact of future information about target changes on performance, and proposes a pragmatic method for identifying the amount of future information on the target that can be utilised effectively in infinite horizon algorithms. Numerical illustrations in MATLAB demonstrate that the proposal is both systematic and beneficial. This thesis introduces several important issues related to model predictive control (MPC)tracking that have been hitherto neglected in the literature, by first deriving a control law for future information about target changes within optimal predictive control (OMPC) for both nominal and constraints cases. This thesis proposes a pragmatic design for scenarios in which the target is unreachable. In order to deal with an unreachable target, the proposed design allows an artificial target into the MPC optimisation problem. Numerical illustrations in MATLAB provide evidence of the efficacy of the proposals. This thesis extends efficient, robust model predictive control (MPC) approaches for Linear Parameter-Varying (LPV) systems to tracking scenarios. A dual-mode approach is used and future information about target changes is included in the optimisation tracking problem. The controller guarantees recursive feasibility by adding an artificial target as an extra degree of freedom. Convergence to admissible targets is ensured by constructing a robustly invariant set to track any admissible target. The efficacy of the proposed algorithm is demonstrated by MATLAB simulation. The thesis considers the tractability of parametric solvers for predictive control based optimisations, when future target information is incorporated. It is shown that the inclusion of future target information can significantly increase the implied parametric dimension to an extent that is undesirable and likely to lead to intractable problems. The thesis then proposes some alternative methods for incorporating the desired target information, while minimising the implied growth in the parametric dimensions, at some possibly small cost to optimality. Feasibility is an important issue in predictive control, but the influence of many important parameters such as the desired steady-state, the target and the current value of the input is rarely discussed in the literature. At this point, the thesis makes two contributions. First, it gives visibility to the issue that including the core parameters, such as the target and the current input, vastly increases the dimension of the parametric space, with possible consequences on the complexity of any parametric solutions. Secondly, it is shown that a simple re-parametrization of the degrees of freedom to take advantage of allowing steady-state offset can lead to large increases in the feasible volumes, with no increases in the dimension of the required optimisation variables. Simulation with MAT LAB 2017a provides the evidence of the efficacy of all proposals.