Data-Based Collaborative Learning for Multiagent Systems Under Distributed Denial-of-Service Attacks

• Published in: IEEE transactions on cognitive and developmental systems Vol. 16; no. 1; pp. 75 - 85
• Main Authors: Xu, Yong, Wu, Zheng-Guang
• Format: Journal Article
• Language: English
• Published: Piscataway IEEE 01.02.2024; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Algorithms; Communication channels; Control algorithms; Control systems; Control theory; Denial of service attacks; Denial-of-Service (DoS) attacks; Denial-of-service attack; Federated learning; Feedback control; Machine learning; Multi-agent systems; Multiagent systems; multiagent systems (MASs); Observers; Output feedback; Protocols; Reinforcement learning; reinforcement learning (RL); Riccati equation; Synchronization; Tracking; Tracking control; Transient response
• ISSN: 2379-8920; 2379-8939
• DOI: 10.1109/TCDS.2022.3172937

This article employs a reinforcement learning (RL) technique to investigate the distributed output tracking control of heterogeneous multiagent systems (MASs) under multiple Denial-of-Service (DoS) attacks. Different from existing results where the dynamic of the leader is known for partial or all agents, the leader's system matrix is completely unknown for each follower in this article. To learn the leader system matrix, a data-based learning mechanism is first proposed using the idea of the data-driven method. Then, under the data-based learning mechanism, a resilient predictor subject to multiple DoS attacks is exploited to provide the estimation of the leader's state for each agent, where adversaries attack different communication links independently. Moreover, a resilient dynamic output feedback controller is proposed to solve the output tracking control problem based on the output regulation theory. To consider the transient responses of agents, an RL-based dynamic output feedback controller is developed to realize the optimal output tracking control by solving discounted algebraic Riccati equations (AREs) in both offline and online ways. Theoretical analysis shows that the secure output tracking control of MASs can be achieved under the proposed data-based resilient learning control algorithm. Finally, a numerical example is provided to verify the effectiveness of theoretical analysis.