Model predictive control for nonlinear systems

Model predictive control (MPC) has been known as one of the most famous techniques due to its optimized control performance and powerful ability to handle the constraints. Therefore, it has been widely used in process control systems. However, there are two disadvantages in MPC. The one is it requires quite heavy computation due to solving an optimization control problem (OCP) in each sampling instant. Overcoming this shortage becomes urgent when extending MPC scheme to fast control systems such as robotic systems. The other disadvantage of MPC is that it depends on the precise model of the plant. However, uncertainties such as external disturbances and parameter perturbations exist in most of systems due to environment variation, measurement noises, load variation, and so on. We seek to overcome the first disadvantage of MPC by, on the one hand, co-designing the event- and self-triggered mechanism in MPC to reduce the frequency of solving the OCP, and, on the other hand, shrinking the prediction horizon and thereby decreasing the computational complexity of the OCP. For the second shortage of MPC, we will develop learning-based MPC scheme to estimate and compensate the uncertainties of the system.