Distributed Design of Ultra Large-Scale Control Systems: Progress, Challenges, and Prospects

TL;DR

This paper addresses the challenge of designing control for ultra large-scale systems (ULSS) by formalizing a multilayer graph-based modeling framework that captures dynamics, sensing, performance, and communication across many subsystems. It articulates four feasibility requirements—topology, design, computation/memory, and communication—and evaluates existing decentralized and distributed approaches (consensus, H2/H∞ BMI, convex relaxation, clustering, DMPC, and special-structure convex formulations) against these criteria. The analysis concludes that current methods do not yet satisfy all ULSS feasibility demands, highlighting the need for distributed design with guaranteed stability and safety, and emphasizing the role of consistency in achieving scalable, real-time implementations. The paper identifies promising directions such as performance-driven consensus design, distributed dynamic clustering on multi-layer architectures, and consistency-based robust distributed design as the most impactful paths forward for ULSS applications in sectors like agriculture, power grids, and space systems.

Abstract

The transition from large centralized complex control systems to distributed configurations that rely on a network of a very large number of interconnected simpler subsystems is ongoing and inevitable in many applications. It is attributed to the quest for resilience, flexibility, and scalability in a multitude of engineering fields with far-reaching societal impact. Although many design methods for distributed and decentralized control systems are available, most of them rely on a centralized design procedure requiring some form of global information of the whole system. Clearly, beyond a certain scale of the network, these centralized design procedures for distributed controllers are no longer feasible and we refer to the corresponding systems as ultra large-scale systems (ULSS). For these ULSS, design algorithms are needed that are distributed themselves among the subsystems and are subject to stringent requirements regarding communication, computation, and memory usage of each subsystem. In this paper, a set of requirements is provided that assures a feasible real-time implementation of all phases of a control solution on an ultra large scale. State-of-the-art approaches are reviewed in the light of these requirements and the challenges hampering the development of befitting control algorithms are pinpointed. Comparing the challenges with the current progress leads to the identification and motivation of promising research directions.

Figures (8)

• Figure 1: Satellites in one of the shells of the Starlink mega-constellation with communication links depicted as yellow edges.
• Figure 2: Block diagram of actuated subsystem $\mathcal{S}_i$.
• Figure 3: Illustrative example of theformation controltask of a constellation of satellites on a single orbital plane (not to scale).
• Figure 4: Topology of the formation control task of a constellation of satellites, depicted for a fraction of the satellite network in Fig. \ref{['fig:eg_sat']}. The four layers of interactions: (i) dynamics; (ii) sensor output; (iii) performance output; and (iv) communication are depicted separately and according to the topology described.
• Figure 5: Scheme of a centralized design for the illustrative network in Fig. \ref{['fig:eg_sat']}. Local information is shared from each subsystem to the central computational entity, which centrally computes a design procedure in real time. The local controllers are then deployed to each of the subsystems, which locally implement the working phase.