Security consensus control for multi-agent systems under DoS attacks via reinforcement learning method

This paper is concerned with the security consensus control issue for discrete-time multiagent systems (MASs) on the basis of a reinforcement learning (RL) approach. Considering the effects of denial-of-service (DoS) attacks, a novel control protocol is proposed to deal with the H infinity consensus problem. Firstly, a Q-learning algorithm is put forward under the directed graph, which can obtain the target gain matrices without any system dynamics information. In addition, the obtained gain matrices and Lyapunov function are employed to demonstrate that the MASs can reach security consensus. Moreover, the proof of H infinity consensus under undirected graphs is derived using the designed Q-learning algorithm. In the end, the simulation experiments are given to verify the correctness of the designed control strategy.