WP1: Local identification in dynamic networks
The objective is to provide a general theory and effective algorithms for local identification in a dynamic network. After defining the particular (sub)network of interest, and the options for measuring/actuating signals as well as locations of potential noise contributions, the theory should provide appropriate methods for identifying models for the dynamic modules as well as for the noise disturbances, and should provide guidelines for setting up the best experiments. Both parametric and non-parametric models will be considered. Algorithms should be scalable to reasonably sized networks (100 nodes).
Researchers involved: Paul Van den Hof, Giulio Bottegal, Harm Weerts, Karthik Raghavan Ramaswamy.
WP2: Topology identification
We will develop a coherent theory and related methods and tools for the identification of the topology of a dynamic network on the basis of measured data. This includes the detection of the presence of feedback loops, the detection of (correlation) structures in disturbance signals, and in general all dynamical relationships between node variables. We cover the most common situation in the systems and control field of networks described by directed graphs , thereby pre-specifying the direction of information flow in each module of the network. Additionally we also cover the situation of non-directed graphs where only links between node variables are described without pre- specifying inputs and outputs, so without arrows in the network block diagrams.
Researchers involved: Paul Van den Hof, Giulio Bottegal, Harm Weerts, Lizan Kivits, Shengling Shi
WP3: Data acquisition and continuous-time (CT) models
There are two important reasons for particularly addressing continuous-time models: (a) because many first principles models are formulated in continuous-time, and (b) because in a dynamic network it will not be realistic to assume that all data are sampled synchronously, which renders a single discrete-time model inadequate. In WP3 we address all questions and issues that relate to data sampling and communication, including aspects of data links with variable communication delays, and (structured) continuous-time models.
Researchers involved: Paul Van den Hof, Tijs Donkers, Giulio Bottegal
WP4: Identification for distributed control
The control of large-scale dynamic networks is a great challenge. Building and maintaining models that can accurately serve as a basis model-based distributed control, requires identification methods that provide the relevant model information. Rather than at all times capturing all dynamic in the network, we have to face the question, which dynamic information is most essential to be extracted from the measurements. In line with the earlier scientific development of the concept of “identification for control”, we will develop theoretical concepts and methods for a new target field of “identification for distributed control”.
Researchers involved: Paul Van den Hof, Tom Steentjes, Lizan Kivits, Mircea Lazar
WP5: Software and applications
We will develop a software suite that will contain the algorithms that are developed in this project, in a well-documented way, so as to make the developed methods accessible for interested users. In order to validate the developed theory and methods, as well as to be inspired by relevant applications, case studies will be performed in different domains of science / engineering.
Researchers involved: Paul Van den Hof, Mohsin Siraj, Jobert Ludlage, Shengling Shi