Research On Secure Distributed Adaptive Diffusion Estimation Algorithm Over Adversarial Networks

With wireless network applications,chip miniaturization increased computing power,and the advances in signal processing technology,distributed information processing technology based on wireless sensor network(WSN)has attracted the attention of scholars and industry as a research hotspot.The distributed adaptive algorithms widely used to solve the problem of distributed parameter estimation due to their simple structure and easy implementation.The distributed adaptive network is composed of a group of sensor nodes with data processing and communication capabilities.It obtains the estimation of system unknown target parameter through using adaptive algorithms and collaboration strategies.However,due to the openness of the network and the characteristic of collaboration between nodes,distributed information processing is vulnerable to network attacks.False data injection attack(FDIA)is one of the most typical attacks.An adversarial network is a network with attackers.When the network encounters the false data injection attack,the nodes in the network will no longer be able to perform the parameter estimation task correctly,resulting in an incorrect target parameter estimation result.Therefore,this thesis studies the secure distributed adaptive estimation algorithm in the distributed network.Firstly,this thesis analyzes the impact of the false data injection attack on the estimation performance of the diffusion least mean square(DLMS)algorithm,and derives the expression of the steady-state mean square deviation(MSD)in the adversarial network environment.The research results show that the estimation performance of the traditional DLMS algorithm deteriorates severely when malicious nodes launch attacks.Aiming at the problem that the estimation performance of the DLMS algorithm severely deteriorates under the false data injection attack,this thesis proposes an adaptive clustering-based secure DLMS(ACS-DLMS)algorithm.In the proposed ACS-DLMS algorithm,the node finds a reliable reference value in the non-cooperative estimation value of the neighbor nodes,and designs an adaptive clustering fusion criterion based on minimizing the instantaneous MSD and the reference value.In order to make the algorithm more robust to false data injection attacks,we propose a data discarding strategy based on minimizing the local cost function of the node.The simulation results show that the ACS-DLMS algorithm has good estimation performance in the adversarial network,and it has achieved a good tradeoff between security and estimation accuracy.To solve the problem of slow convergence of adaptive clustering fusion strategy in adversarial networks,this thesis proposes a trust-based secure DLMS(TS-DLMS)algorithm.In the proposed TS-DLMS algorithm,the node uses the local non-cooperative estimation to construct a threshold test to determine whether the neighbor node data is abnormal.Moreover,an adaptive threshold update rule is designed.Besides,in order to track the behaviors of nodes better and reduce the impact of false alarm detection on estimation performance,we propose a trust evaluation mechanism based on abnormal data detection results and data deviations,and the fusion weights are adjusted accordingly according to the trust value of neighbor nodes.The simulation results show that the TS-DLMS algorithm has better steady-state estimation performance and robustness under the false data injection attack.Two distributed adaptive diffusion estimation algorithms proposed in this thesis effectively prevent the propagation of false data in the network,and have a good application prospect in adversarial network environments where the false data injection attack exists.