Development and analysis of Predictive Functional Control for a range of dynamical processes.

This thesis presents a development and analysis of a low computation Model Predictive Control (MPC) that is known explicitly as Predictive Functional Control (PFC) for different types of dynamical processes. Since the current concept of PFC suffers from several issues such as the weak efficacy of tuning parameter, conservative constrained solution and poor handling of challenging dynamical systems, the prime objective of this research is to develop novel approaches to tackle these limitations while retaining the simplicity of formulation, coding and tuning of PFC. For the first contribution, a Laguerre based PFC (LPFC) is proposed to handle a system with stable and straightforward dynamics where it can provide better consistency between model predictions and actual system behaviour. Consequently, LPFC also improves the overall closed-loop performance including the efficacy of tuning parameter while providing more accurate and less conservative constrained solutions compared to the conventional PFC. For the second contribution, a Pole Shaping PFC (PS-PFC) is developed based on the original idea of the traditional pole cancellation technique to alleviate the effect of undesirable poles when dealing with more challenging dynamical processes such as those with open-loop integrating, oscillating and unstable modes. This approach provides a stable response with less aggressive input demand compared to the pole cancellation method while retaining a similar recursive feasibility property of constraint handling. The third contribution of this research is on the development of an off-line sensitivity analysis to measure the robustness and possible sensitivity trade-off for different PFC structures in response to noise, disturbance and parameter uncertainty. The findings show that although LPFC provides better closed-loop performance than PFC, yet the control loop may become more sensitive to noise. On the other hand, for PS-PFC, since the conventional Independent Model (IM) structure is unable to handle a divergent prediction, a T-filter is proposed where it manages to recover the sensitivity to noise by sacrificing its sensitivity to disturbance and parameter uncertainty.