Off-Policy Reinforcement Learning for Synchronization and Parameter Identification of Dynamical Complex Networks
Yu‐Ying Lu, Juan Wang, Xiaoyuan Luo, Shaobao Li · 2025
Due to the complexity of dynamical complex networks, both in terms of topology and dynamics, synchronization design based on a well-identified model is often challenging. On the other hand, treating the system as a completely black box also complicates network property analysis. To address this issue, this study investigates the data-driven synchronization problem for dynamical complex networks, alongside the identification of topology and structural parameters. We propose an optimal control method based on off-policy reinforcement learning to learn the synchronization law without requiring prior knowledge of the system dynamics. Using the learning results, we identify the coupling strength and node dynamics through inverse computation of the off-policy reinforcement learning algorithm. The convergence of the learning process is analyzed using Lyapunov stability theory. A key contribution of this work is that both the optimal synchronization policy and network identification are achieved simultaneously, despite the system dynamics being unknown. Finally, simulation results are presented to demonstrate the effectiveness of the proposed algorithm.