Research On Attack Detection And Cooperative Control In Multi-agent System

The multi-agent system(MAS)uses the communication and coordination and cooperation between the agents to solve large and complex tasks,which is favored by experts and scholars in many related fields such as artificial intelligence and distributed computing.It also has broad application prospects in many practical projects such as smart grids,smart transportation,UAV formation flying,and multi-intelligent robots.Multi-agent system can complete tasks well in a complex environment,but it is also vulnerable to external malicious attacks due to its open characteristics.Therefore,the important part of the research content of this thesis emphasizes on the security issues of multi-agent system.This thesis aims at studying the multi-agent system that is subject to false data injection attack(FDI),and propounds the corresponding algorithms to keep the system from the damage of the attack.The prime objective of this thesis is to detect whether there is FDI attack in the system and determine the location of the attack,that is,to find out the attacked agent,and put forward a strategy to enable the damaged multi-agent system to continue to carry out cooperative control.This paper chooses to carry out analysis and discussion based on the first-order MAS and the second-order MAS,and obtains good research results.In the following,we will introduce the specific research results of this paper.1.Firstly,the first-order multi-agent system under FDI attack is studied,and a monitor is designed to detect and identify FDI attacks.The monitor is mainly based on Luenberger observer and relies upon the technology from distributed control.We construct the corresponding attack detection filter for each agent,and the output of the filter is the state residual,which is the basis to judge whether there is FDI attack in the multi-agent system.Since the agents can directly communicate with each other,we separate the neighbor state of each agent system,and do not consider this part when constructing the residual filter corresponding to the attack recognition for the purpose of reducing the influence of neighbor nodes on identifying the attacked agent.According to the output of the filter,it can be directly judged whether the agent is attacked.2.Then we conduct research on the second-order multi-agent system attacked by injection of false data.Regarding the detection of FDI attack in the second-order MAS,we choose the state estimation method based on unknown input observer,and assume that only one agent in the system is attacked.The attack signal of each agent is sequentially decoupled from the second-order multi-agent system and regarded as an unknown input to construct a UIO to obtain the state estimation error,and then use the error corresponding to all agents to determine which agent is attacked.This is a centralized attack detection method we propose.Improve the method to obtain a distributed attack detection method,that is,choose to divide every three agents that can form a topological network with a spanning tree into a group,and then only need to calculate the residuals of two of the agents for each group.Then it can be judged whether there is an agent attacked by FDI in the group,and the same judgment can be made on each group in turn.Then,we propose another distributed attack detection method,considering the combination of UIO and Luenberger observer.In this method,the state of neighbor nodes is selected as the unknown input to construct UIO,and then use the estimated state to design a residual filter.The residual signal can be used to determine whether the agent is attacked by FDI.Similarly,after the attack set is obtained,attack isolation is implemented so that the remaining normal agents can then perform coordinated control to complete the task.3.Finally,a cooperative control strategy is proposed to enable the multi-agent system injected with false data attacks to get rid of external influences and continue to collaborate to complete tasks.After obtaining the corresponding attack set according to the attack detection methods of the first-order and second-order multi-agent systems mentioned in this article,we consider to use the strategy of attack isolation to separate all the agents destroyed by FDI attack from the MAS,so that the remaining agents not attacked form a new multi-agent network and realize the state consistency of the first-order and second-order multi-agent systems respectively.Then the agents can continue the cooperative control.In this paper,we also give the consistent value of the state that the first-order MAS finally converges to,and the values that achieves consistency with the position and velocity state of the second-order MAS.The feasibility and accuracy of all the proposed algorithms in the thesis are vertified by the implementation of simulation experiments on MATLAB platform through design examples.For both the first-order MAS and the second-order MAS that are under FDI attack,the agents destroyed by the attack could be found successfully,and the state of the system could converge to the same value again after the implementation of attack isolation.