Research On Dynamic Consensus Problems In Distributed Multi-Agent Network

Distributed multi-agent system mainly relies on the mutual communication,coordination and cooperation between intelligent nodes to form a complex network system.It plays an important role in various fields such as intellion,smart cities,cluster structures and sensor networks.Consensus is the basic problem of distributed multi-agent system control,and it has been extensively and deeply studied in the past ten years.In the current research on consensus problems,the most of them assume that the communication of all agents are reliable,the agent is always stable,and the topology is fixed,and so on.The distributed control of multi-agent systems in actual applications will face many problems inevitably,such as unstable network communication,malicious attacks on agents,and agent failures.This paper mainly studies the consensus problem when the multi-agent system has adversarial agents in the network.The main research content and results achieved are as follows:1.Focusing on the consensus control strategy of a distributed linear multi-agent system under adversarial nodes attack.Giving the consensus condition which satisfies strict state feedback under the ideal situation without adversarial nodes.Then,considering the adversarial nodes attack,designing a strategy which based on neighbor node monitoring and isolation to identify and isolate adversarial nodes.Under this designed,each agent has a tolerance for adversarial node,and this design that it does not directly isolate suspected adversarial nodes,but adopt a correction strategy based on state deviation to correct the malicious deviation.Under this method,it not only can reduce network fluctuations effectively but also achieve the identification,isolation of adversarial nodes and the back online of the node which is change into good node from adversarial node.Finally,numerical simulation and a six-degree-of freedom UAV distributed consensus simulation experiment are given to demonstrate the effectiveness of the proposed algorithm.2.Studying the model predictive control problem of the distributed linear multi-agent system based on hypergraph.Proposing an optimization scheme of distributed model predictive control with H-CADMM optimization algorithm.Because there are virtual fusion centers in the hypergraph structure,it can effectively improve the convergence speed of the network.3.Considering the model predictive control problem of the distributed linear multi-agent system based on hypergraph under adversarial nodes.Considering adversarial nodes attack,designing an algorithm that the child nodes adopt the algorithm which is based on neighbor node monitoring and isolation scheme,while the virtual fusion central nodes adopt the strategy based on Byzantine thought.It can effectively improve the anti-interference performance of the virtual fusion central node.Proposing an adaptive correction strategy based on state history data to deal with suspected adversarial nodes within the tolerance range.Finally,numerical simulation is given to demonstrate the effectiveness of the proposed algorithm.