Research On Distributed Parameter Estimation In Adversarial Network Environment

With the rapid development of the information age,artificial intelligence has gradually been popularized into people's daily lives,which brings great convenience to people's daily lives and improves the quality of people's life.The rapid development of the information age is closely related to the development of sensors.Compared with the previous sensors,the current sensors have a series of advantages such as small size,high accuracy,low power consumption,low price,strong communication ability,etc.,which has laid a strong foundation for modern artificial intelligence.Wireless sensor networks(WSNs),which consist of a series of sensors,have been widely used.WSNs are often used to detect some unknown parameters that people are interested in.That is,WSNs are used for parameter estimation.In this paper,WSNs with the distributed diffusion strategy are used to estimate an unknown parameter.The distributed diffusion strategy is that each sensor node and its neighbor nodes in WSNs cooperate to estimate the unknown parameter.In WSNs,nodes usually work in a secure network environment.However,when an attacker appears,the network will be in an adversarial network environment.The last parameter estimated by WSNs will be offset from the unknown parameter,that is,the wrong parameter is estimated.Therefore,this paper is devoted to studying how to estimate the correct unknown parameter in the adversarial and distributed network environment.Firstly,this paper studies the impact of external attacks in WSNs,that is,WSNs are attacked by channel attacks.Combining data selection strategies,this paper proposes a distributed data selection diffusion least mean square(DLMS)algorithm.The proposed algorithm can both censor abnormal data and select updated data with sufficient innovation.In order to make the proposed algorithm more robust against channel attacks,the distributed data selection DLMS algorithm with credibility weight is designed based on the characteristic of data selection.This paper analyzes the mean performance and mean square performance of the proposed algorithm theoretically,and then verifies the robustness of the proposed algorithm through simulation examples.The simulations' results show that the distributed data selection diffusion DLMS algorithm is significantly better than other data selection algorithms.And when the wireless sensors are in an adversarial network environment,the distributed data selection DLMS algorithm with credibility weight is strongly robust.Then,this paper studies the changes of the network model parameters when the WSNs is attacked by the internal attack,that is,when it is attacked by the false data injection(FDI)attack.This paper proposes a distributed DLMSKL algorithm based on Kullback-Leibler(KL)divergence.When the proposed algorithm detects the nodes that were attacked by FDI attack,this paper proposes three different strategies to weaken the impact of FDI attack.This paper first analyzes the mean performance and mean square performance of DLMS in a secure environment.Based on the above mathematical analysis,this paper analyzes the mean performance and mean square performance of the DLMSKL algorithm with different strategies.Meanwhile,this paper analyzes the rationality of the threshold design strategy used in detecting damaged nodes.Finally,simulations' examples are used to verify the robustness of the proposed algorithm with different strategies.The experimental results show that the proposed algorithms are robust against FDI attack,and the method of removing damaged nodes is the most robust.And when the wireless sensors are in an adversarial network environment,the distributed DLMSKL algorithm with removing damaged nodes is significantly better than the algorithms in other literatures.