Date of Award
Doctor of Philosophy
Dr. Conor Downing
One of the more popular model predictive control approaches, of late, is the generalised predictive controller (GPC). It has received widespread dissemination in academia and numerous successful applications in industry testify to its usefulness. Its popularity partly stems from its superiority over its predecessors, its simplicity and its control acumen. However, since GPC is based on a mathematical model of the process, significant deterioration in performance or even closed-loop instability may arise due to the inevitable discrepancies that arise between the model and the true system. To ensure that the loop stability is not compromised and an adequate level of performance is maintained the control law must be robust with regard to the plant-model mismatch. This may be achieved by manipulating a specific tuning parameter, the T-polynomial. However, the selection of this polynomial is not trivial as its effect is nonmonotonic. Furthermore, in practice T must be chosen to achieve a realistic trade-off between robustness and disturbance rejection performance. Despite this recognition, few of the existing tuning rules address this compromise and no systematic design exists.
The primary contribution of this thesis is the development of a systematic tuning technique for the GPC such that an optimal trade-off between robustness and disturbance rejection is achieved. In this context optimal implies that the design should be just sufficiently robust to cater for the existing model uncertainty, in which case the disturbance rejection performance can be optimised. This design was achieved by minimising, with respect to T, an identical cost function as that employed in //«, control. PerfoiTnance was incorporated by including an additional criterion, the integral of error of the load disturbance response, into the optimisation. This problem formulation does not lend itself to an analytic solution and therefore a numerical technique, known as genetic algorithms, was applied.
Based on the success of this design an auto-tuning tool for the GPC was developed. With this tool the servo performance is dictated by a user-specified nominal closed-loop transfer function while the regulator performance is dictated, as before, by the trade-off between robustness and disturbance rejection. This considerably simplifies the commissioning process. The usefulness of this tool is demonstrated by simulations in Chapter 4 and by real-time experiments in Chapter 5. In both cases the results are compared with previously published tuning techniques. The excellent results obtained support the belief that the advocated design is of practical significance and is the principal contribution of this thesis.
O'Mahony, Thomas, "Robust Generalised Predictive Control - An Optimal Design for Uncertain Systems" (2002). Theses [online].
Available at: https://sword.cit.ie/allthe/460