Research On Theory And Applications Of Model Predictive Control

Model Predictive Control (MPC), also known as moving or receding horizon control, has originated in industry as a real-time computer control algorithm to solve linear multivariable problems that have constraints and time delays.This dissertation takes a comprehensive and deep insight in model predictive control algorithm, and presents some renovation points after summarizing a lot of previous research reports in these fields.The main contents and renovation points in this dissertation include the following:1. The multi-objective and layered steady-optimization method of model predictive control for stable and integrating systems is proposed from the application background of optimization and control for complex industrial processes. Firstly, the meanings of multi-objective and layered optimization is explained in detail and the steady-state optimization based on quadratic programming is presented to solve the multi-objective and layered optimization in the context of two-stage model predictive control formulation.2. A robust design method for predictive controller parameter based on Min-Max ruler is presented that takes into account model uncertainty. Simulation results show that it enables predictive controller to keep better control performance when the plant's dynamic character change and predictive controller parameters needn't be designed again.3. An off-line algorithm of constrained model predictive control based on piecewise-linear feedback control law is presented to solve its potentially high on-line computational demand. The piecewise-linear feedback control law is off-line computed by LMI optimization problems and the on-line control computation reduces to the simple evaluation of the defined piecewise linear function, which makes the online computational efficiency significantly improved.4. A stabilizing receding horizon control algorithm based on piecewise-linearterminal weighting matrix is proposed for discrete linear systems with constraint. The piecewise-linear terminal weighting matrix is off-line computed by LMIoptimization problems and the on-line computation reduces to the simple switch of the piecewise-linear terminal weighting matrix. The main advantage of this approach is that a trade-off between feasibility and optimality is obtained.5. A multivariable process identification based on asymptotic black-box theoryis studied. Firstly, a high-order MIMO ARX model and its frequency error bound isestimated from identification data and low-order SISO models is obtained fromhigh-order MIMO ARX model. It solves the problem of model order determinationand model validation for multivariable processes. It improves model quality and reduces time for plant test and data analysis and reduces disturbance to unitoperation during plant test.At the end of this dissertation, the author also gives some suggestion to thefurther research in these fields.