Design And Implementation Of Distributed Estimation Algorithms To Resist FDI Attacks

With the emergence of the Internet of Things concept,various communication networks such as wireless sensor networks and the Internet of Vehicles have seen a boom in development.The stability and efficiency of distributed algorithms have led to their widespread use in communication networks,with the attendant problems arising from the practical application of distributed signal processing techniques.Distributed algorithms will be made more complex in order to cope with more realistic requirements,and the security of the large amount of communication data that exists in such a complex network is a concern for all.Currently,distributed algorithms have been extended with a number of algorithms suitable for different application scenarios depending on the number of tasks,signal types,cost functions,etc.For example,the distributed multi-task estimation is extended according to the number of tasks in networks,and distributed graph signal processing techniques are developed after the type of signal evolved from simple spatiotemporal sequences to graph signals with unstructured properties.Therefore,this thesis will examine the problem of resisting false data injection attacks in three scenarios: distributed single-task estimation algorithms,distributed multi-task estimation algorithms and distributed graph signal algorithms.The research work is divided into three main points:(1)The distributed single-task estimation algorithm against false data injection attacks is investigated,an attack detection strategy based on the generalized maximum correntropy criterion and a corresponding secure distributed algorithm are proposed.This thesis uses the generalized maximum correntropy criterion based on its insensitivity to outliers to detect outliers in the prediction error.And the estimated value of the node with small prediction error is used to replace the estimated value of the abnormal node.Thus a corresponding attack detection strategy is proposed.In related work,the assumption for the number of attacked nodes in each node’s neighbourhood is usually less than half the number of nodes in the neighbourhood.In contrast,a more relaxed assumption is made in this thesis,where the algorithm maintains satisfactory performance when the number of attacked nodes in the network increases to even more than half the number of all nodes.Simulation experiments found that the proposed algorithm achieves better steady-state performance compared to related work when false data injection attacks are present in the network.And simulation experiments are conducted to compare the selection of parameters and thresholds used in the attack detection strategy.(2)The distributed multi-task estimation algorithm against false data injection attacks is investigated,an attack detection strategy suitable for multi-task environments based on the generalized maximum correntropy criterion and a corresponding secure distributed multi-task algorithm are proposed.Since it is not appropriate to directly introduce the attack-resistance approach from the distributed single-task estimation into the distributed multi-task estimation,the attack detection strategy based on the generalized correlation entropy criterion is adapted for a multitasking environment in this work.A regular term related to other clusters is added to the cost function and only the neighbor nodes in the same cluster are used in the fusion step.In addition,the presence of attacks on neighbouring nodes of each node becomes more complex as the number of tasks in the network increases.For this case,this thesis proposes an attack hypothese that is applicable to multi-task environments.Then,convergence analysis in the mean and mean square sense is carried out theoretically.In the simulation experiments,both the proposed algorithm and the related work are effective against the false data injection attacks,but the proposed algorithm has better performance as the number of attacked nodes increases or even exceedes half of the number of all nodes in the network.The simulation experiments also compare the performance of the algorithms obtained using different parameters and thresholds in multi-task environments.(3)Distributed graph signal processing with false data injection attacks is investigated,and corresponding secure distributed graph signal processing algorithm is proposed using an attack detection strategy based on the generalized maximum correntropy criterion.In the distributed graph signal diffusion least mean square algorithm,the data model becomes similar to that in classical signal processing when the original graph signal is made dynamic by introducing a time dimension.The parameter to be estimated at this point is the graph filter coefficient.This thesis migrates the attack detection strategy applied to classical signals in distributed processing to graph signals for defending against false data injection attacks.The simulation experimental results show that the proposed attack detection strategy based on the generalized maximum correntropy criterion is also applicable to distributed graph signal processing algorithms with false data injection attacks.