| Network security has now become an important issue in the field of automatic control.Because of the integration of physical devices and information technology,the control system needs to deal with network attacks launched by malicious adversaries in addition to ensuring traditional control performance.The security and stability of the control system under network attack cannot barely rely on the communication protocol or encryption technology in the computer field.Analyzing and defense cyber-attack from a control perspective is the underline of system stability.In a distributed system,multiple subsystems are integrated by signal measurement and information exchange,which heavily relies on network communication.Vulnerable links are more likely to exist in a large number of network communications,and hence distributed systems are more vulnerable to network attacks.Distributed systems have a wide range of applications in computing,storage,control,measurement,etc.One of the major application scenarios is multi-agent systems.Multiagent systems use distributed algorithms to make multiple simple agents work together and complete group tasks such as collaborate,formation,optimization,and sensor fusion.At the same time,in a multi-agent system,there is a signal that is the same to all agents in the system,which is called a homologous signal.Common homologous signals include ranged environmental factors(e.g.,temperature,air pressure)or network attacks limited by attack principles(e.g.,homogenous attacks on satellite signals).Therefore,homology attacks in multi-agent systems require further research and analysis by researchers.In this thesis,we study the problem of secure state estimation when sensors in multi-agent systems are attacked.The homologous attack signal occurs in the measurement channel of the agent,so the attack is also called the homologous sensor attack.Without prior knowledge of the agent’s state,secure state estimation uses only the input of each agent and the output corrupted by a homologous sensor attack to reconstruct the state of each agent.Due to the homology of the attack signal,this thesis uses the consensus algorithm to reconstruct the homologous attack signal,and further realizes the reconstruction of the system state.This thesis firstly obtains the necessary and sufficient condition for the system to achieve secure state estimation under homologous sensor attacks,and then progressively designs various types of observers to achieve the above research purposes.The types of secure state observers designed in this thesis include distributed observers with multi-loop structure,Luenberger-like distributed observers and event-triggered distributed observers.Distributed observers with multi-loop structure includes distributed static double-loop observers and distributed online triple-loop observers,which are suitable for offline and online operation,respectively.These two distributed observers utilize the distributed average consensus algorithm as their inner-loop.The static double-loop observer uses the residual update loop as the outer-loop,and realizes that the estimated error of the observer converges exponentially with the number of computations.A double-loop generalized observer framework adapted to various average consensus algorithms is also proposed.The generalized framework can obtain bounded estimation errors.The online triple-loop observer realizes online operation by superimposing the time update loop in addition to the residual update loop.The estimation error of triple-loop observer converges with time,but not the number of computations.The distributed observer design with multi-loop structure mainly studies the relationship between the number of loops and the convergence of the estimation error.The related results show that as long as the multi-agent system satisfies the most basic sufficient and necessary conditions,the distributed observer with multi-ring structure must exist.The distributed Luenberger-like observer focuses on the design of the observer gain and the time delay in the system communication network.For the existence and non-existence of communication delay in the communication network,this thesis designs two distributed 类Luenberger observers and adapts different consensus estimation strategies for them.The distributed Luenberger-like observer greatly reduces the amount of computation and communication per time step and maintains the estimation performance.The Luenberger-like observer is more suitable for online operation than the distributed observer with multi-loop structure.The sufficient conditions for the Luenberger-like observers converge to the real value are given by biaffine matrix inequalities.The biaffine matrix inequality can guide the design of the observer gain and the selection of parameters.Distributed event-triggered observers design triggering conditions for the information transmission of Luenberger-like distributed observers.Only when the state of the relevant agent satisfies the event-triggered condition,the communication of this agent is triggered,otherwise the agent remains silent.The distributed event-triggered observer also considered the time delay of communication,out-of-order and finally obtains a bounded estimation error.Since the upper bound of the estimation error is positively related to the trigger threshold of the event trigger condition,the designer can adjust the trigger threshold to meet different design requirements.The simulation results show that various observers can achieve the expected performance under the given conditions.The final chapter of this thesis gives a brief overview of the future direction of secure state estimation under multi-agent homologous attacks.For example,the global positioning system is widely used in military,navigation and other fields,and the observer proposed in this thesis provides a new positioning system security scheme from the perspective of cybernetics. |
