Resilient Average Consensus in Presence of False Data Injection Attacks

Average consensus is important for distributed multi-agent systems, with applications ranging from network synchronization, load balancing for parallel processor, distributed information fusion, average cardinal number estimation for database, to decentralized control. In distributed average consensus algorithms, each node updates its state based on corresponding coupling weights and received messages from its neighbors. However, in presence of false data injection attacks, which might be caused by external attackers or internal faults, conventional average consensus algorithms fail. To address this, we propose a resilient average consensus algorithm which is able to ensure the convergence of normal nodes to accurate average value of their initial states. More specifically, the proposed algorithm utilizes adaptive coupling weights from Q-consensus to gradually exclude faulty nodes. Meanwhile, inspired by surplus algorithm, the proposed algorithm augments an auxiliary variable for each node to eliminate the accumulated messages from faulty nodes and further track the average shift amount of normal nodes, which is key to achieving accurate average consensus. Numerical simulations confirm the effectiveness and efficiency of our proposed algorithm.