Data-driven modelling in dynamic networks

In many areas of science and technology, increased complexity and interconnections of systems is a strong motivation for developing modelling, control and optimization methods for dynamic networks. While in the control field attention is paid to decentralized, distributed and networked control, as e.g. in multi agent systems, data-driven modelling is still dominantly restricted to considering simple open-loop and closed-loop structures. In this seminar we consider several questions that appear when addressing the problem of data-driven modelling in structured linear dynamic networks, and we will set up a framework for addressing those questions. They include identification of a particular module (local identification) within the network and sensor location selection for achieving consistency. The concept of network identifiability is introduced and it is shown how classical closed-loop identification methods can be generalized to the dynamic network situation