Event-Triggered Safety Control Strategy in Learning-Based Model Predictive Control of Partially Unknown Nonlinear System

In this study, a novel learning-based model predictive control (LMPC) framework that uses a Lipschitz interpolation (LI)-based nonparametric estimation method is developed for the optimal control of a partially unknown Lipschitz discrete system subject to noise and constraints. This work focuses on the design and theoretical analysis of an online model learning method and an event-triggered constraint handling strategy. First, a sampled data set construction mechanism for an LI-based prediction function is developed to ensure a bounded estimation deviation with low computational complexity. Second, a relaxed barrier function is introduced as a safe control strategy to address the hard state constraint. To overcome the contradiction between the satisfaction of state constraints and the efficient online solving of the optimal control problem, a model confidence-based triggering mechanism is investigated to reduce redundant execution of the safety control strategy. Finally, numerical simulations are conducted to verify the effectiveness of the proposed LMPC framework.