Dynamic Event-Triggered IRL Algorithms for Nonzero-Sum Games of DC Microgrids

This paper proposes an integral reinforcement learning (IRL)-based optimal control method to solve the nonzero-sum (NZS) game problem of DC microgrids with partially unknown dynamics under the dynamic event-triggered mechanism. Initially, the islanded DC microgrid with buck or boost converters is modeled in state-space form. The IRL method is introduced to relax the requirement of the system drift dynamics. Subsequently, a single neural network (NN) is designed to approximate the performance index. Furthermore, this study introduces a novel dynamic event-triggered scheme to address the NZS game problem while reducing computation and communication burdens. By using Lyapunov’s theorem, it is demonstrated that both the system states and the estimation error of the critic NN weights are ultimately uniformly bounded. In the meantime, it is also proven that there is no Zeno behavior here. Finally, this work provides two examples to confirm that the designed algorithms are accurate and efficient.