Experience-based integral reinforcement learning consensus for unknown multi-agent systems

This paper investigates an optimal consensus control problem and proposes a policy iteration algorithm based on online integral reinforcement learning for nonlinear multi-agent systems with unknown dynamics. Introducing a critic-actor neural network into the traditional policy iteration avoids the identification of unknown dynamics. To address the issue of local optima in online learning, an experience-based weight-tuning law is introduced to ensure the persistence of excitation conditions during the training phase. The theoretical results show that the system is asymptotically stable, and the network weights converge. Finally, the effectiveness and correctness are verified by several simulation studies.