Dynamic event-triggered optimized control for nonlinear multi-agent systems via reinforcement learning.

Abstract

The practical deployment of high-order nonlinear multi-agent systems (MASs) is often hindered by two critical bottlenecks: limited communication resources and unavoidable environmental uncertainties. To overcome these challenges, an innovative optimized consensus tracking control framework with a dynamic event-triggered mechanism is established. For handling the unknown nonlinearities, multilayer perceptrons (MLPs) are employed as adaptive function approximators, and their parameter tuning is guided by a multi-agent Actor-Critic reinforcement-learning (RL) mechanism embedded with consensus information. The optimized controller is equipped with a dynamic event-triggered control (DETC) mechanism that adaptively regulates sampling-error thresholds online, thereby reducing the burden on data exchange and processing. By utilizing the Lyapunov analysis method, closed-loop stability is ensured and Zeno behavior is effectively prevented. To demonstrate the performance of the presented approach, numerical studies are conducted on a representative multi-electromechanical system. The results show that the proposed method not only significantly reduces communication overhead but also demonstrates superior robustness against sensor noise compared to existing independent learning approaches, providing a reliable solution for resource-constrained networked control systems.