Online Learning-Based Event-Triggered Model Predictive Control With Shrinking Prediction Horizon for Perturbed Nonlinear Systems

Abstract

This article proposes an online learning-based event-triggered model predictive control (OLEMPC) scheme for constrained nonlinear systems with state-dependent uncertainties. The scheme incorporates both the nominal and the learned models to ensure favorable theoretical properties during online learning. A composite measurement-triggering strategy is devised to reduce the number of state measurements as well as solving the optimization problems. This strategy attenuates the conservatism in measurement and triggering through combining the event- and self-triggering approaches. By implementing the algorithm, both state measurement and triggering frequency further decrease with the online refinement of the prediction model, and the prediction horizon adaptively shrinks as the state approaches the terminal region. It is shown that the feasibility of the optimization problem and stability of the closed-loop system are guaranteed. Simulation results verify the effectiveness of this scheme in ensuring closed-loop performance and alleviating computational burden.