MODELING, ANALYSIS, AND CONTROL OF COMPLEX NETWORKS AND CYBER-PHYSICAL SYSTEMS
Modeling, Analysis, and Control of
Complex Networks and Cyber-Physical Systems
Ischia, Biblioteca Comunale Antoniana
Saturday, June 29
Sunday, June 30
Abstracts of the contributions
Distributed Monitoring and Fault-Tolerant Control: Scalable Plug & Play Tools and Industry 4.0 Perspective
This lecture deals with a class of systems that are becoming ubiquitous in the current and future "distributed world" made by countless "nodes", which can be cities, computers, people, etc., and interconnected by a dense web of transportation, communication, or social ties. In an increasingly "smarter" planet, it is expected that such interconnected systems will be safe, reliable, available 24/7, and of low-cost maintenance – the Industry 4.0 vision. Therefore, faults and malfunctions need to be detected promptly and their source and severity should be diagnosed so that corrective actions can be taken as soon as possible. Once a fault is detected, the faulty subsystem can be unplugged to avoid the propagation of the fault in the interconnected large-scale system. Analogously, once the issue has been solved, the disconnected subsystem can be re-plugged-in.
In the talk, an adaptive approximation-based distributed fault diagnosis approach for large-scale nonlinear systems will be dealt with, by exploiting a "divide et impera" approach in which the overall diagnosis problem is decomposed into smaller sub-problems, which can be solved within “local” computation architectures. The distributed detection, isolation and identification task is broken down and assigned to a network of "Local Diagnostic Units", each having a "local view" of the system. The following step is the integration of a distributed model predictive control scheme and a distributed fault diagnosis architecture. Being distributed, the whole model of the large-scale system is never used in any step of the design process. We refer to this kind of decentralized synthesis as plug & play design, if - in addition - the plug-in and unplugging operations can be performed through a procedure for automatically assessing whether the operation does not spoil stability and constraint satisfaction for the overall large-scale system.
In the lecture, the connection is finally worked out with Virtual Commissioning which is the very recent trend in the process industry to make the dream of plug & work installation of a reliable and efficient automation system become a reality.
Smart Buildings: Monitoring, Control and Fault Tolerance
Modern buildings are complex systems of structures and technology aimed at providing a safe and comfortable environment for the occupants. Recent advances in information and communication technologies have generated significant interest in developing smart buildings, which provide much greater capabilities in terms of energy efficiency, safety, security, interactivity, as well as in terms of mitigating environmental impact. New components for smart buildings, such as sensors, actuators, controllers, embedded systems and wireless communications, are becoming readily available. Moreover, the Internet-of-Things (IoT) technology is already having a significant impact on developments related to smart buildings. The objective of this presentation is to provide an overview of current advances in smart buildings and to present some results on monitoring, control and fault tolerance of Heating, Ventilation and Air-Conditioning (HVAC) systems, which are crucial components of smart buildings. Various estimation, learning and feedback control algorithms will be presented and illustrated, and directions for future research will be discussed.
The role of non-normal dynamics for the efficient information transmission in networks
While there exists a consistent number of contributions in computational neuroscience that support the hypothesis that non-normality makes the information transmission more efficient in neuronal networks, no quantitative analysis has been proposed, analysis that would allow the comparison of different network topologies. In this contribution we propose a model that puts in a precise frame the arguments proposed in those papers. Based on this model we propose a metric that, through the Shannon capacity of a suitable channel build on the network, enables to quantify the information transmission efficiency. This allows to confirm that non-normality enhance the channel capacity, but only in the high noise regime. Finally we specify which non-normality degree plays a role in this enhancement.
Structure preserving reduction of networks: classical control methods
Most of the current reduction techniques for networks of systems rely on clustering, because they easily preserve the network structure in the reduced model. Application of the classical methods that are very relevant from a control systems perspective, such as balancing and moment matching based model reduction, generally do not preserve the network structure. In this presentation we focus on the generalisation of these classical methods so that both the amount of nodes and the (passive) dynamics of nodes are reduced, while preserving relevant network and (passive) dynamics structures. The developments are done for networks of linear systems, as well as networks with Lur’e dynamics on the nodes. Furthermore, a priori error bounds will be provided, optimal weight allocation for the reduced order network is considered and relevant small and large scale examples will be used to illustrate the results.
Dynamics and regularities in the structure: opportunities for control
In this talk I will start discussing the notion of symmetries and equitable partitions for complex networks and then illustrate how they impact the dynamics on the networks, focusing in particular on consensus and synchronization. In the second part of the talk I will show how these notions could be exploited to design control strategies for inducing synchronization in a subset of the nodes of the network or multi-consensus regimes.
Emerging control problems in complex networks
Complex networks theory was born from the need of explaining how ensembles of interacting units may give rise to fascinating emerging behaviors that cannot be explained by only looking at the individual dynamics. Accordingly, it could be considered the natural framework for studying a wide range of phenomena in very diverse disciplines. However, classic complex networks theory is based on a set of simplifying assumptions to guarantee analytical tractability, which immediately appear inconsistent when we aim at controlling real-world complex systems.
Optimal control of networks: energy scaling and open challenges
Recent years have witnessed increased interest from the scientific community regarding the control of complex dynamical networks. Some common types of networks examined throughout the literature are power grids, communication networks, gene regulatory networks, neuronal systems, food webs, and social systems. Optimal control studies strategies to control a system that minimize a cost function, for example the energy that is required by the control action.
We show that by controlling the states of a subset of the nodes of a network, rather than the state of every node, the required energy to control a portion of the network can be reduced substantially. The energy requirements exponentially decay with the number of target nodes, suggesting that large networks can be controlled by a relatively small number of inputs, as long as the target set is appropriately sized.
Control Theory for Practical Cyber-Physical Security
In this talk, we discuss how control theory can contribute to the analysis and design of secure cyber-physical systems. We start by reviewing conditions for detectability and impact of data attacks targeting feedback control loops running over a field communication network. We investigate three different attack scenarios: Sensor attacks, actuator attacks, and coordinated actuator and sensor attacks. In particular, we highlight how a physical understanding of the controlled process can guide us in the allocation of counter measures and limit the possible impact of attacks.
Distributionally Robust Learning with Applications to Health Analytics
We will present a distributionally robust optimization approach to learning predictive models, using general loss functions that can be used either in the context of classification or regression. Motivated by medical applications, we assume that training data are contaminated with (unknown) outliers. The learning problem is formulated as the problem of minimizing the worst case expected loss over a family of distributions within a certain Wasserstein ball centered at the empirical distribution obtained from the training data. We will explore the generality of this approach, its robustness properties, its ability to explain a host of "ad-hoc" regularized learning methods, and we will establish rigorous out-of-sample performance guarantees.
Beyond predictions, we will discuss methods that can leverage the robust predictive models to make decisions and offer specific personalized prescriptions and recommendations to improve future outcomes. We will provide some examples of medical applications of our methods, including predicting hospitalizations for chronic disease patients, predicting hospital length-of-stay for surgical patients, and making treatment recommendations for diabetes and hypertension.
E-bikes as a paradigm to design human-in-the-loop cyber-physical systems
In this talk we present the design of a cyber-physical control system for an intelligent e-bike. The system, which is deployed on a real-world testbed, leverages tools from data analytics, stochastic processes and control to manage the interactions between the cyclist and the bike motor. Our ultimate goal is to influence the cycling behavior and an application concerned with the regulation of the cyclist breathing rate to minimize his/her intake of environmental pollution is discussed. After presenting experimental and theoretical results, we outline and generalize some of the main challenges of our design, which are underpinned by dynamical systems and control theory.
Distributed Resilient Control of Dynamic Flows in Transportation Networks
Ever-growing loads, limited infrastructure capacity, and new technologies enabling the use of intelligence at unprecedented levels have created significant new challenges in the control of transportation networks. Due to their level of interconnectedness and the complex interactions between cyber and physical layers and human decision makers, these systems may exhibit inefficiencies and fragilities. This talk will present recent results on resilience and efficiency of distributed control architectures for dynamic flow networks with with applications to traffic signal control, dynamic pricing, and route guidance systems.
Multi-agent Map-building: Kalman Filtering for Gaussian Processes
The proliferation of large scale smart multi-agent systems, also known as Internet-of-Things, Networked Control Systems, Wireless sensor and actuator networks, Cyber-physical Systems, etc., are providing us with a wealth of data with unprecedented time-space resolution which can trigger the next technological revolution. However, this trend is also posing a formidable challenge, often referred as Data Tsunami, which requires the analysis of a large-scale correlated time-series. In this talk, the problem of estimating a map will be addressed, first in a static scenario and later in a dynamic scenario, based on noisy measurements collected by a large number of sensors in the presence of unreliable communication. In particular, pros and cons of parametric and non-parametric estimation approaches will be discussed and some strategies are proposed to merge ideas from control theory such as Gauss-Markov estimators and Kalman Filtering, and from Machine Learning such as Gaussian regression, Karhunen-Loève kernel expansions and Nystrom method.