Privacy-Preserving Average Consensus for Multi-agent Systems with Directed Topologies

Abstract

In the process of forming average consensus, the privacy that the agents do not want to disclose may be maliciously speculated and used by others. To avoid breaches of privacy for multi-agent systems subject to directed topologies, we propose a novel privacy-preserving average consensus algorithm that employs an improved Laplacian-type control protocol. It is shown that all agents can achieve accurate average consensus without the weight-balance condition despite directed topologies. To ward off internal malicious agents, we add edge-based zero-sum interference signals in the process of transferring information. Thus, by introducing a private parameter, all agents can be protected against malicious eavesdroppers who know the entire topology and can intercept communication links. Two simulation examples are presented to demonstrate the validity of our algorithms for realizing the average consensus under the impacts of malicious adversaries.