Distributed Estimation And Decision-making Algorithms Over Adaptive Networks

Distributed estimation of unknown parameters based on adaptive networks has been widely applied in environmental monitoring,precision agriculture,military and aerial surveillance.However,the existing distributed estimation algorithms mostly assume that nodes are in the secure network environment.In the adversarial networks under false data injection(FDI)attacks,the malicious data injected by FDI attackers would propagate across the network via the diffusion strategy,resulting in severe deterioration in the estimation accuracy of these algorithms.In this thesis,the FDI detection,the resilient estimation strategy and the game theory based decision-making are investigated to enhance the estimation accuracy in the adversarial networks under FDI attacks.Specifically,the main contents are summarized as follows:Firstly,the correntropy-based resilient distributed estimation algorithm over adversarial networks is developed,in which each node utilizes the correntropy-based discrimination scheme followed by a state perception scheme to detect FDI attacks in a distributed manner.Further,the reference neighbor is selected at each compromised node to restrain the propagation of malicious data.As is validated in the simulations,the proposed resilient distributed estimation algorithm could effectively detect FDI attacks as well as accurately estimate the unknown parameter in the adversarial networks,with its estimation performance close to that of the renowned diffusion Least Mean Square(d LMS)free of attack.Secondly,the mean and mean-square performance of the proposed resilient distributed estimation algorithm is analyzed in this thesis.Then,a global constrained optimization problem with respect to the reference neighbor is formulated based on the upper bound of the MSD.The distributed optimal reference neighbor selection scheme is further proposed.The simulation results validate the estimation performance enhancement of the proposed optimal reference neighbor selection scheme in the scenarios of both heavy and time-varying FDI attacks.Finally,the decision-making process of each node in the adversarial networks under FDI attacks is explored based on the non-cooperative game theory,and accordingly,the resilient distributed game is proposed.Specifically,the proposed secure reception reward avoids the impact of FDI attacks on the evaluation of the estimation accuracy improvement,and the introduced rescue earning promotes secure nodes to provide assistance to the compromised nodes.Further,the best response rules for both compromised and secure nodes are also derived.Illustrative simulations validate that each node could adaptively make its own decision considering both the communication cost and rescue earning in the adversarial networks,and the rescue earning contributes to the resilient distributed estimation of unknown parameters.