Dynamics Based Privacy Preservation for Average Consensus on Directed Graphs

Average consensus is widely used in many applications, such as time synchronization, information fusion, load balancing, and distributed optimization. However, to reach an agreement, conventional average consensus algorithms require all the agents to explicitly exchange their states with neighbors, which inevitably leads to an undesirable disclosure of participating agents' private state information. To address this, we propose a novel privacy preserving average consensus algorithm for directed graphs. By introducing careful-designed randomness into interaction dynamics, our proposed algorithm is able to protect the privacy of participating agents against both internal honest-but-curious attackers and external eavesdroppers. Furthermore, by exploiting the inherent robustness of consensus dynamics, our algorithm guarantees convergence to the exact average consensus value. Numerical simulations are conducted to demonstrate the effectiveness and efficiency of our proposed algorithm.