| Model predictive control (MPC) is a paradigm that emerged in the late 1970's, and has developed considerably over the last two decades. The two versions which appear to have had the most acceptability are one derived by Clarke et al. [24] and is called GPC for generalized predictive control, and secondly that derived by Cutler and Ramaker [30] called DMC for dynamic matrix control.; Multirate systems are common in industry, especially in chemical process control. The synthesis and analysis of GPC for single-rate systems has drawn considerable attention since the 1980's. On the other hand multirate GPC, practically useful as it may be, has not received much attention. Most of the research in the multirate control area deals with the time-varying nature of multirate systems ([54, 99]) and hence makes the design and analysis more complex. The use of the lifting technique as introduced by Kranc [50] allows one to convert multirate systems into single-rate and linear time-invariant (LTI) equivalent systems. However, its use also presents a challenge in the synthesis problem: the lifted controllers should satisfy certain causality constraint. The main purposes of this thesis are: (1) To develop GPC controllers for multirate systems, taking into account the causality design condition. (2) To develop a sampled-data GPC scheme for multirate systems so that the continuous-time closed-loop performance is optimized, and the inter-sample behavior is improved. (3) To perform robust stability analysis of a general class of multirate MPC controllers in the presence of model-plant mismatch (MPM). Note that GPC belongs to the MPC class of controllers which are all model based algorithms.; The thesis focuses on the synthesis and analysis of GPC for multirate systems, as well as the more general and special case of multirate systems such as non-uniformly sampled systems and fast-control, slow-sampling, dual-rate systems. In addition to these, related topics such as the multirate digital redesign problem and the continuous time delay estimation problem are also investigated in this study. |
