Event-triggered learning-based control for output tracking with unknown cost functions

In this paper, a two-layer event-triggered learning-based control framework is proposed to address extremum seeking problem in networked control systems with limited communication resources and unknown cost function. In this framework, the lower layer is an event-triggered controller to drive the output to track the given setpoints generated from the upper layer, where a learning-based optimizer is developed to approach the extremum of the unknown cost function. Specifically, in the lower layer, an event-triggered output controller, based on a high-gain extended state observer, is designed to tackle uncertainties and disturbances. In the upper layer, a nonparametric gradient model is established, and then the gradient descent method is applied to generate setpoints for the tracking control. The update of the learning and optimization process is determined by the tracking performance of the lower layer. The stability and Zeno-freeness of the proposed event-triggered controller is proved. Furthermore, the dependence of the convergence rate of the proposed learning-based extremum seeking algorithm on the designed parameters is also explicitly characterized. Finally, the effectiveness of the proposed framework is validated by numerical examples.