7th IFAC Workshop on Distributed Estimation and Control in Networked Systems (NecSys18)

Plenary Speakers

  1. Carlos Canudas de Wit (CNRS, GIPSA-Lab)

  2. Jorge Cortes (University of California San Diego)

  3. Florian Dörfler (ETH Zurich)

  4. Antoine Girard (CNRS, L2S)

  5. Julien Hendrickx (UC Louvain)

  6. Paul M.J. Van den Hof (TU Eindhoven)

  7. Steve Morse (Yale University)

  8. Giuseppe Notarstefano (University of Bologna)

  9. Lacra Pavel (University of Toronto)


Carlos Canudas de Wit (CNRS, GIPSA-Lab)

Towards Scale-Free Control of large-scale traffic networks

Abstract: The talk discuss new research lines addressing the optimal control of time-varying dynamic large-scale traffic networks. The driving idea consists in aggregating the whole network in sub-networks having some suitable control properties, allowing the optimal control design to be more tractable. The talk start with a recall of controllability and observability properties in 1-D traffic network and provide a detailed example in how to partition the systems and how to solve the associated time-varying optimal control. Then we discuss recent work on how such ideas can be extended to planar networks in relation with the problem of optimization played over networks and focus our attention on the importance of the topology of the graph and its importance in the structure of optimization setup. Then, we will discuss new avenues for organizing such large-scale complex networks, by proposing a new aggregation procedure that combines, complexity reduction (scale-free graph structure), physical preservation properties, and control targets. Finally, we will present some recent (more technical) results on scale- free control and observation on such aggregated large scale networks.

Jorge Cortes (University of California San Diego)

Understanding the Role of Network Structure in Controlling Complex Network Systems

Abstract: Controllability of complex network systems is an active area of research at the intersection of network science, control theory, and network coordination, with multiple applications ranging from brain dynamics to the smart grid and cyberphysical systems. The basic question is to understand to what extent the dynamic behavior of the entire network can be shaped by changing the states of some of its subsystems, and decipher the role that network structure plays in achieving this. This talk examines this question in two specific instances: characterizing network controllability when control nodes can be scheduled over time and shaping network behavior via inhibitory control. Regarding controllability, we show how time-varying control schedules can significantly enhance network controllability over fixed ones, especially when applied to large networks. Through the analysis of a novel scale-dependent notion of nodal centrality, we show that optimal time-varying scheduling involves the actuation of the most central nodes at appropriate spatial scales. Regarding shaping network behavior, we examine network mechanisms for selective inhibitory control under linear-threshold dynamics. We build on the characterization of key network properties determining basic dynamical properties to enable the targeted inhibition of task-irrelevant subnetworks through control. We show that network stabilizability is solely determined by the latent sub-network of nodes that do not directly receive external stabilizing controls. This allows us to quantify the proportion of nodes that need to be directly controlled for efficient inhibitory control, as well as a computational theory for achieving stabilization through feedback or feedforward control. Finally, we show how the proposed techniques find applicability in a broad class of network analysis and control problems.

Florian Dörfler (ETH Zurich)

Control of Power Converters in Low-Inertia Power Systems

Abstract: The electric power grid is undergoing a period of unprecedented change. A major transition is the replacement of bulk generation based on synchronous machines by renewable generation interfaced via power electronics. The loss of synchronous machines poses a great challenge because today’s power system operation heavily relies on their self-synchronizing dynamics, rotational inertia, and resilient controls. The robust operation of such a low-inertia system based on power electronics is currently regarded as the ultimate bottleneck to massively integrating renewables. As a possible remedy, numerous approaches aim at emulating so-called virtual inertia or other characteristics of synchronous machines. In this talk we investigate “how” virtual inertia can be emulated in a detailed power converter model, which leads us to a nonlinear matching control strategy based on DC rather than AC measurements. Prompted by this finding, we also pursue the question “what else” can be done other than naively emulating synchronous machines. We will discuss a novel and foundational oscillator control approach and report some theoretic results as well as experimental validations.

Antoine Girard (CNRS, L2S)

Verification and synthesis of timing contracts for networked cyber-physical systems

Abstract: Networked cyber-physical systems involve multiple distributed control loops sharing common communication and computation infrastructure. High-confidence design of such systems requires to investigate the tight interaction between the physical and cyber components. In this context, contract-based approaches have been identified as a promising direction for cyber-physical systems design.

In this talk, we advocate the use of timing contracts for analyzing stability and scheduling problems in networked cyber-physical systems. A timing contract associated to a control loop specifies constraints on the time instants at which certain operations are performed such as sampling, actuation, communication or computation. Under timing contracts, the control engineer is responsible for designing control laws that are robust to all possible timing variation specified in the contracts while real-time engineers can focus on designing scheduling protocols so as all control tasks can be executed in agreement with their contract.

In the first part of the talk, we will consider the problem of stability verification: given models of the physical plant and of the controller and a timing contract, verify that the resulting dynamical system is stable. We propose an approach based on the notion of reachable set, and which builds on efficient over-approximation algorithms developed over the past decade. A comparison with approaches based on LMIs is provided. In the second part of the talk, we will consider the scheduling problem: given several control loops, each with a timing contract, to be implemented on a shared distributed infrastructure, synthesize a dynamic scheduler allocating the shared computational resources, which guarantees that all timing contracts are satisfied. We propose an approach based on timed games, whose solution provides a suitable schedule. In the last part of the talk, we will consider the problem of synthesizing suitable timing contracts so as to guarantee at the same time stability and schedulability. Our approach uses monotonicity properties attached to timing contracts to sample efficiently to contract parameter space.

Julien Hendrickx  (UC Louvain)

Open Multi-Agent Systems: arrivals and departures

Abstract: Even though scalability and robustness to agent losses are often cited as advantages of multi-agent systems, almost all theoretical results apply to system with fixed compositions. We consider open systems, that agents continuously leave and join during the process considered. We discuss the general challenges to analyze and design algorithms for such systems. Arrivals and departures keep indeed perturbing the system, forbidding any classical convergence. Moreover, correction mechanisms designed to cope with a small number of arrivals or departures may fail when these events keep taking place.

We focus in particular on averaging, decentralized estimation, and computation of the maximum value among agents present in the system. We also present some fundamental performance limitations in open systems.

Paul M.J. Van den Hof (TU Eindhoven)

Data-driven modeling in linear dynamic networks

Abstract: In many areas of science and technology, the complexity of dynamic systems that are being considered, grows beyond the level of single systems into interconnected networks of dynamic systems. In control and optimization this has led to the development of decentralized and distributed algorithms for control/optimization, as e.g. in multi-agent systems.

From the modelling perspective, data-driven modelling tools are typically developed for relatively simple open-loop and closed-loop structures, while the opportunities for big data handling in the current data science era, are becoming abundant. As a result there is a strong need for the development of data-driven modelling tools for large-scale interconnected dynamic networks.

In this plenary we will highlight the main developments and challenges in this area. Besides setting up a modelling framework, we will address problems of local identification of a particular part of the network, including the selection of the appropriate signals to be measured. The concept of network identifiability is highlighted and the role of structural properties of the network, in terms of its topology/graph, is given strong attention. It is also shown how classical closed-loop identification methods need to be generalized to be able to cope with the new situations.

Steve Morse (Yale University)

Estimating the State of a Linear System Across a Network

Abstract: The problem of estimating the state of a linear system whose measured outputs are distributed across a network has been under study in one form or another for a number of years. Despite this, only recently have provably correct state observers emerged which solve this problem under reasonably non - restrictive assumptions. The aim of this talk is to describe some of these observers. In all cases the observers are intended to estimate the state x of an m-channel, n-dimensional jointly observable, linear system of the form . The system’s state x is simultaneously estimated by m agents assuming agent i senses and receives state of each of its neighbor’s estimators. Neighbor relations are characterized by a directed graph whose vertices correspond to agents and whose arcs depict neighbor relations. Depending on the type of observer being discussed, may be time varying or time-invariant. For the time-invariant case, the observers are themselves time - invariant linear systems. For the time -varying graphs case, the observers are hybrid linear systems consisting of local linear time-invariant estimators and local linear equation solvers, one such pair for each agent. Conditions will be discussed under which both types of observers can obtain asymptotically correct estimates of x at exponential convergence rates which can be freely assigned. It will be explained why the hybrid observer is resilient to abrupt network changes such as the loss of an agent, provided both the network and available measurements are sufficiently redundant entities.

Giuseppe Notarstefano (University of Bologna)

Distributed Optimization in Smart Networks

Abstract: Numerous estimation, learning, decision and control tasks in smart networks involve the solution of large-scale, structured optimization problems in which network agents have only a partial knowledge of the whole problem. Distributed optimization aims at designing local computation and communication rules for the network processors allowing them to cooperatively solve the global optimization problem without relying on any central unit. In this talk I will start presenting some classes of optimization problems arising in cyber-physical networks and show key challenges that arise in a distributed computation framework. Then I will present novel distributed methods to address some of these challenges as: solve big-data problems (in which the high dimension of the decision variable induces limitations on the local computation and communication), guarantee a controlled constraint violation in dynamic problems, or solve mixed-integer (possibly uncertain) programs.

Lacra Pavel (University of Toronto)

Network Games: On the Extension of Nash’s Theory to Networks

Abstract: Networks are ubiquitous around us. Consider the Internet, with its underlying communication, the smart grid/power network, a network of robots engaged in search and rescue, or a group of people interacting over Facebook. They are all instances of large- scale networks with multiple entities/agents that have decision-making capabilities and individual goals, while interacting in a strategic manner. Decision-making in such scenarios can be modeled using game theory. Pioneered by Nash and von Neumann in the fifties for economics, game theory has become a widely applicable framework that provides insights into the factors that govern chance and decision in complex systems in everyday life. However, as a theory, it needs to evolve for today’s globally networked world. What’s needed is a unified framework for network games - a framework that allows both the analysis of strategic networked interactions, and the design of nearest- neighbor agent-interaction/learning rules. These rules should rely on locally available information, minimize superfluous communication, and, by using them, agents should reach an optimal collective state, such as (generalized) Nash equilibrium. This is a topic of increased research interest in recent years, but having a general theory for network games, or games on networks, is far off. In this talk we review some of our group’s contributions towards getting there. Our approach is based on exploiting connections to distributed control of multi-agent systems, and uses tools from passivity, convex optimization and monotone operator theory. We present some examples from wireless networks and social media networks.