Information Transfer-based Topology Identification of Dynamic Multi-agent Systems

Abstract

This paper proposes a data-driven approach to identify the communication network topology of dynamic multi-agent systems using information transfer. In particular, we focus on the discrete-time phase-coupled oscillators which have a directed tree structure. Our proposed approach is a protocol that adjusts the agents’ feedback gain due to synchronization while identifying the communication network topology, which is unknown in advance by an operator outside the network, by using the difference of two transfer entropy values based on the limited state information under transition. After revealing that the steady-state phase allocation is determined by the natural frequencies of the oscillators, we show that oscillations are reduced and fast synchronization is achieved by appropriately adjusting the feedback gain based on the network topology obtained by the proposed method through numerical examples using simple network structures. We also discuss the limitations of the proposed approach.