Cluster-based resilient distributed estimation through adversary detection

Security becomes increasingly important due to various attacks from adversaries in wireless sensor networks. This work considers a resilient distributed estimation of an unknown parameter with a cluster-based approach when some agents are adversarial. A two-phase algorithm is adopted to perform parameter estimation and detect attacks. First, a cluster scheme is proposed to make sure that each cluster is connected. Then, the attack is detected and estimation is achieved with a consensus +innovation estimator in each cluster. Finally, the cluster heads combine the consensus estimates in each cluster and exchange with other cluster heads to achieve unknown parameter estimation. In addition, the detection sensitivity under different cluster schemes is also compared. Numerical examples illustrate that the proposed cluster-based approach can improve the convergence rate and detection sensitivity.