SCO Seminar - Henk van Waarde, University of Groningen

Title: Mini-course on data-driven control - part 1

Abstract:

Advances in sensing technology have led to a massive increase in available data. There is an enormous potential for using these data to guide the design of controllers for a wide range of applications, including power systems, robotics and biological systems. However, before such controllers can be employed in practice, a number of fundamental issues need to be addressed. Indeed, how can we make sure that data-driven control design comes with the same stability and performance guarantees traditionally associated with model-based control? And how to develop data-driven control algorithms that are able to cope with big data sets?

The purpose of this two-part mini-course is to provide an overview of some of the main ideas in data-driven analysis and control. The mini-course is theoretical in nature and the material will be presented in a tutorial style.

In this first part, I will begin with some historical perspectives, going back to the literature on persistency of excitation and the fundamental lemma by Willems and coauthors. We will see how this literature gave rise to a number of so-called direct data-driven control methods, i.e., approaches that map data directly into control policies without the intermediate step of system identification. Thereafter, I will introduce the notion of informative data as a general framework for studying a variety of data-driven analysis and control problems. The framework will be applied to obtain data-driven Hautus tests for controllability analysis, and conditions for data-driven stabilization using exact data.