Reinforcement learning-based secure synchronization for two-time-scale complex dynamical networks with malicious attacks

Abstract

This paper studies the secure synchronization problem for two-time-scale complex dynamical networks with unknown dynamics information and malicious attacks. The challenge is that under complex dynamic networks with unknown dynamic information, all system matrices are unknown. To ameliorate this conundrum, we design a reinforcement learning algorithm, conjoined with a full-order processing of singularly perturbed parameters, with the express objective of securing the stability of the synchronization error system within a context devoid of adversarial intrusions. Second, in order to reduce the impact of malicious attacks, a switching function is developed based on feedback gain. Besides, it is proved that the proposed distributed controllers can still guarantee the convergence of the synchronization error system under malicious attacks. Finally, two numerical examples are given to illustrate the applicability and effectiveness of the proposed algorithm.