Dual-model predictive control

Model Predictive Control (MPC) uses a mathematical model of the process to predict the future process output trajectory. A multi-step optimization problem is then formulated which gives an optimal control action even in the presence of hard constraints. MPC is "optimal" in the sense of minimization of a user-specified performance index but the traditional design does not include any way of insuring stability and/or robustness (to model error). Three important issues are covered in this thesis: (1) process modelling; (2) predictive control design and tuning; (3) closed loop analysis of stability and robustness. All three are important requirements in both academic and industrial applications.; A dual-model formulation is developed to represent the process. The dual-model, specified in state space form, combines the advantages of both the Finite Step Response (FSR) and the Deterministic Auto Regression and Moving Average (DARMA) model forms. Defining the output predictions directly as the states, the dual-model generates explicit output predictions for use in the control calculations and is also convenient for expansion to multivariable process modelling and identification. Two important related issues, state estimation and parameter estimation, are discussed in detail. For optimal state estimation, the standard observer theory can be simplified when applied to the specific structure of the dual-model. For parameter identification, the extended Kalman filter gives predictive-control-relevant model identification based on experimental data.; Predictive control design includes many user specified control tuning parameters. Physically, the general effect of traditional tuning parameters is intuitive and easily understood, but the selection of specific values is quite 'ad hoc'. A new method which results in better dynamic matrix conditioning is developed for choosing numerical values of tuning parameters. Then, two simple new tuning parameters, {dollar}alpha{dollar} and {dollar}beta{dollar}, are introduced to fine tune the servo and regulatory performance respectively. With {dollar}alpha{dollar} and {dollar}beta{dollar}, the controller can be adjusted on-line to obtain the best trade-off between robustness and servo/regulatory performance.; Two important closed loop system issues, robustness and constrained stability, are also discussed. Using the modelling errors in parametric form and matrix perturbation theory, simpler and less conservative robust stability criteria are developed by using the special structure of the dual-model formulation. When MPC has active constraints, its closed loop control structure is changed. The traditional constrained stability analysis procedures are applied and some difficulties in practical applications are explored. Then, two new approaches to handle hard constraints are proposed.; The theoretical research results developed in this thesis are incorporated into a new predictive control scheme, Dual-Model Predictive Control (DMPC). DMPC provides enhanced functionality (e.g. control-relevant identification, constraint handling) and more flexibility (e.g. MIMO identification and control) for practical control applications.