Uncooperative Behavior Detection And Isolation For Multi-agent Systems Using Relative Information

In this paper,the distributed uncooperative behavior detection and isolation(UBDI)framework is investigated to improve the security and reliability of multi-agent systems.In recent years,automatic UBDI for multi-agent systems is of growing importance with the growth of the systems' complexity and intelligence.But their inherent decentralized structure makes it more challenging for the lack of a central node monitoring the whole system's activities.Once an uncooperative node appears,its negative impacts may spread out and affect the normal work of the whole system.The existing model-based UBDI methods always have a lot of shortcomings,such as having a complicated structure,occppying too much computing resources,hard to be decentralized and so on.In this paper,we are going to design a new topology-based distributed UBDI framework to overcome the shortcomings of the existing methods.Firstly,a distributed UBDI strategy is proposed for multi-agent systems with single integrater dynamics.The gossip algorithm is used to improve the reliability of the single node's detection result and a new uncooperative behavior recovery method is proposed to reduce the applying restrictions of the UBDI strategy and significantly broadened the scope of applications.Secondly,for multi-agent systems with linear dynamics,an UBDI strategy is proposed using only relative information.By a series of model transformation,a new uncooperative behavior detection model is constructed.It can make the error signals and the state information decoupled and reduce the difficulty of application of the UBDI strategy.What's more,a new distributed uncooperative behavior detection algorithm which can monitor the neighbours' activities by estimating their nominal outputs is created.Finally,to make full use of the group dominance of multi-agent systems,a local communication framework is designed.It can be used to share the detection results of the target agent within a subnetwork.By using the consensus algorithm,each node in the subnetwork can get the globel detection result of the target agent just by communicating with their neighbours,and the reliability of the detection results can be improved without a higher hardware performance requirement.