Abstract - Nikolaou - 9615979 Constrained model predictive control (MPC) is a technique that uses a process model in an on-line optimization scheme to meet various control objectives for a process whose inputs and outputs are constrained. There are always discrepancies between the model that MPC uses and the real world. Such discrepancies may arise from inaccuracies in the initial model used by an MPC system, or from variation of a process with time. Increased MPC autonomy, the main thrust of this project, refers to the capability of an MPC system to handle, with minimal human intervention, a wide range of situations where a process model could be inaccurate. Situations that will be addressed in this research are: o Time-invariant processes for which an inaccurate linear of nonlinear model and bounds on the inaccuracy are available or need to be developed. o Time-varying processes under constrained MPC feedback. More specifically, the following problems will be investigated: o How MPC systems can be structured and tuned (off- or on-line) to ensure robust performance for a certain level of process/process-model mismatch. Conversely, how much modeling accuracy is required for robust MPC performance. o How a constrained MPC systems that monitors its performance, can modify itself if necessary, e.g., by performing simultaneous control and closed-loop process identification.
