Event Triggered Strategy For Robust Model Predictive Control

In practical applications, there exist various uncertainties, including modelling errors, unknown parameters, noises, and external disturbances, etc. Robust model predictive control (MPC) is an effective design approach for uncertain systems due to its superiority to constraints, uncertainties, multivariable issues and inaccurate models.Currently, numerous efforts have been made for the robust MPC on uncertain systems. Nevertheless, the traditional time-triggered robust MPC algorithm has more conservatism because of the huge online computation, which limits the applications in actual industrial fields. In contrast, the event-triggered robust MPC has more superiority due to its specific control mechanism. Hence, it is of theoretical and practical significances to discuss the event-triggered robust MPC for uncertain systems.In this thesis, we focus on presenting the applicable event-triggered robust MPC strategies with less online computation, lower optimization complexity and favourable robust performance for the systems with bounded disturbances and constraints. The main results include:(1) The event-triggered robust MPC method for linear discrete-time system is proposed under the consideration of addictive bounded disturbances and input/output constraints. The constraint tightening approach is adopted to establish the optimization problem, and the robust feasible control sequences for the subsequent events are conducted based on the optimal control sequence at event-triggered time instant. In addition, the event-triggered condition is derived to ensure the robust stability of the controlled system by means of the input-to-state stability theory. It is shown that the proposed method not only avoids the computation complexity, but also reduces the online calculation.(2) For the large scale system with every subsystem subjected to constraints and bounded disturbances, an event-triggered decentralized robust MPC approach is proposed, in which the event-triggered robust model predictive controller is designed for every subsystem. Especially, the associated information of global state and the addictive bounded disturbance are regarded as a whole, and the bounded set of this term can be calculated. It shows that the proposed methodology effectively improves the general reliability. Furthermore, it guarantees the closed-loop robust stability and reduces the requirements of computing capabilities for CPUs.(3) For the case that the states are unmeasurable, an event-triggered robust output feedback model predictive controller is designed, which consists of an asymptotically stable Luenberger state estimator and an event-triggered robust model predictive controller. The robust invariant set is derived and the event-triggered robust model predictive controller is designed. This proposed method ensures that the actual system maintains in the tube with the trajectory of state estimation as the center and the invariant set of estimation error as the width.