Data-driven control of network systems: Accounting for communication adaptivity and security

Gang Wang, Wenjie Liu, Yifei Li, Xin Wang, Jian Sun, Jie Chen · 2025

Read Data-driven control of network systems: Accounting for communication adaptivity and security by Gang Wang, Wenjie Liu, Yifei Li, Xin Wang, Jian Sun, Jie Chen (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2511-15044.txt.

Introduction

Gang Wang, Wenjie Liu, and four co-authors survey data-driven control for networked systems, where many linked units must be steered together. Such systems run everything from self-driving fleets to smart power grids, but their links are imperfect and their interconnections are intricate. Learning controllers straight from data has become the practical response to those modeling headaches.

The survey organizes the recent literature around four pain points: communication delays, sampling that happens at irregular times, network security under attack, and fully distributed setups with no central boss. For each, it reviews how data-driven designs identify the system, build the controller, and still guarantee stability. The map should help engineers pick methods that fit their constraints.

Catalog record

CREATORGang Wang, Wenjie Liu, Yifei Li, Xin Wang, Jian Sun, Jie Chen
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE135 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2511-15044

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Data-driven control of network systems: Accounting for communication adaptivity and security Gang Wang, Wenjie Liu, Yifei Li, Xin Wang, Jian Sun, Jie Chen Abstract Over the past decades, network systems have surged in significance, driven by merging technological advancements. These systems play pivotal roles in diverse applications ranging from autonomous driving to smart grids, yet they confront complexities arising from network imperfections and intricate interconnections, which challenge system identification, controller design, as well as stability and performance analysis. This survey provides an in-depth exploration of network systems from most recent data-driven perspective, across four key issues: communication delay, aperiodic sampling, network security, and distributed configurations. By doing so, this survey enhances our comprehension of the challenges and theoretical innovations within the realm of network systems. \ArticleType Review \Year 2025 \Month \Vol \No \DOI \ArtNo \ReceiveDate \ReviseDate \AcceptDate \OnlineDate \AuthorMark \AuthorCitation Data-driven control of network systems: Accounting for communication adaptivity and security keywords Data-driven control, Networked control, Aperiodic sampling, Resilient control, Distributed control 1 Introduction The last two decades have borne witness to an unprecedented surge of interest in network systems, driven by the rapid evolution and merge of computing, communication, networking, and data-centric learning technologies. This heightened enthusiasm finds its roots in the pivotal role that network systems now occupy within contemporary engineering systems. These systems span a vast spectrum of engineering applications, including autonomous driving, intelligent manufacturing, smart grids, and remote healthcare [ 14 , 41 , 38 , 98 , 132 , 101 ] . Concurrently, industries are intensifying their call for systematic methodologies capable of adeptly modeling, analyzing, designing, and enhancing network systems in a robust, resilient, and sample-efficient manner, where sensor and control data traverse intricate digital communication channels. This escalating demand underscores the pressing need for innovative solutions and continued research in the ever-evolving domain of network systems. Yet, amidst this technological fervor, it is paramount to acknowledge that this type of system is not immune to several network-induced challenges, including communication imperfections, safety, security, and scalability. Numerous efforts have been made addressing these challenges when an explicit system model is available; see e.g., [ 91 , 77 , 81 , 88 , 104 ] . However, as the scale of the modern networked systems growing, e.g., robotics [ 6 ] , biology [ 54 ] , or human-in-the-loop systems [ 129 ] , modeling using first principle or identifying an exact system model from data become difficult [ 30 ] , which highly restricts the practical implementation of the aforementioned control patterns. Very recently, a convergence of a hundred researchers from diverse backgrounds within the IEEE Control Systems Society gathered to collectively contemplate a scientific roadmap for the future of our control discipline [ 30 ] . To quote from Section 4 of the “Control for Societal-Scale Challenges Roadmap 2030 ” [ 4 ] and Introduction of [ 30 ] report: “One of the major developments in control over the past decade—and one of the most important moving forward—is the interaction of machine learning and control systems.” Since the key feature of machine learning is to learn a control law directly from data, this statement indicates the increasing importance of data-centric methods in control theory, referred to as data-driven control [ 44 , 127 ] . Most recently, the fundamental lemma in [ 115 ] has attracted reviving interest in this literature due to its advantage in designing data-driven controllers with rigorous theoretical guarantees. To be specific, this lemma provides a non-parametric representation of a linear time-invariant (LTI) system using a trajectory of the system. Inspired by this work, a number of applications and generalizations have been developed, including model predictive control (MPC) [ 8 , 66 , 80 , 46 , 110 , 111 , 116 ] , optimal control [ 24 , 26 , 93 , 67 , 102 ] , robust control [ 130 , 45 ] , dissipativity analysis [ 97 , 55 ] , nonlinear control [ 36 , 74 ] , among others. In the remainder of this review, we focus on the fundamental lemma-based data-driven method, which we will refer to simply as the data-driven control method, and embark on an expedition through the multifaceted landscape of network systems. Our exploration is categorized into four themes of concern: communication delay, aperiodic transmission, network security, and distributed configuration. These themes are arguably the most representative issues of network systems and have been considered in, e.g., [ 100 , 126 ] . 1.1 Communication delay Communication delay is a prevalent factor in almost all practical engineering systems, including power systems, vehicle suspension systems, and communication networks. This temporal discrepancy can lead to performance degradation and even instability. The primary challenge revolves around stability analysis, specifically determining the maximal allowable upper bound (MAUB) for the delay. Model-based approaches. When the exact system model is known, several methodologies have been developed, among which the Lyapunov-Krasovskii stability theorem plays an important role due to its adaptability to time-varying delay. Guided by this theorem, extensive efforts have been made to reduce the conservatism of stability conditions by enlarging the MAUB. To this aim, two aspects are often taken into account. One is to develop a tighter estimation method of the derivative of the Lyapunov functional, while the other lies in designing a suitable Lyapunov functional. More references can be found in [ 34 , 65 , 103 , 112 , 18 , 21 , 22 ] . Data-driven approaches. So far, data-driven solutions to this problem have only been considered in [ 87 , 107 ] , where the delays are assumed to be known constants during offline data collection. Under this assumption, data-based stability conditions under noise-free, measurement noise-corrupted, process noise-corrupted data were derived. Although the assumption on exact time-delay removes most of the difficulties, their results can be seen as a starting point in this area, leading to several possible research directions. 1.2 Aperiodic transmission Central to the investigation of network systems is the challenge of determining the optimal execution frequency for sensors, controllers, and/or actuators. Striking the right balance between communication cost and overall system performance is the crux of this matter. Conventional periodic or time-triggered schemes, characterized by fixed transmission rates, often result in inefficient utilization of communication resources. To tackle this issue, a resource-efficient scheduling approach for data transmissions, known as event-based control, has been widely studied. There are two effective event-based approaches, namely, event-triggered control (ETC) and self-triggered control (STC) [ 91 , 104 , 40 ] . Model-based approaches. An ETC periodically or continuously assesses the system’s state/output to determine if predefined conditions warrant data transmission [ 1 , 39 ] . In contrast, an STC determines the next transmission time by sequentially comparing the predicted future state/output with the most recent triggered state/output. Consequently, under STCs, sensors can be completely deactivated between sampling times, thereby conserving energy and extending the sensor’s lifespan, [ 5 , 35 , 76 ] . Data-driven approaches. In the literature of data-driven aperiodic transmission, numerous efforts have been made, including both data-driven ETCs and STCs. Specifically, various data-driven ETCs have been proposed with respect to noise-free and noisy data, different types of systems, and different event-triggered strategies (ETSs) [ 75

Read the complete text