Localized Learning of Robust Controllers for Networked Systems with Dynamic TopologyDownload PDF

08 Jun 2020 (modified: 05 May 2023)L4DC 2020Readers: Everyone
Abstract: This paper addresses the problem of controller synthesis for networked systems with dynamic and unknown topology. Such networked systems arise in applications such as the power grid when lines between consumers are downed due to natural causes, or in multi-agent robotic networks and embedded sensor networks, both of which tend to be structurally complex and prone to high faults. We develop a robust and adaptive system-level controller synthesis approach to this problem. The algorithm is described sequentially in three settings: centralized, localized, and iterative localized for multiple modifications over time. In particular, the iterative localized scheme is designed around networks which switch between topological configurations arranged in a finite-state Markov Chain; the subsystems additionally perform consensus to estimate its transition probabilities. The advantages of our method are twofold. First, as a result of the localized implementation of the system-level approach, our controller can be extended to large-scale networks. Second, achieving exact consensus or learning precisely where the network structure has changed is not necessary to stabilize the overall network due to the robust nature of our controller. To demonstrate performance, we simulate the iterative localized algorithm for a simple centered hexagon network with 4 different topological states.
0 Replies

Loading