Christof Vermeersch

Research

Globally optimal least-squares system identification and model reduction

The identification, analysis, and manipulation of dynamical systems emerge naturally in almost every branch of industry and economy. Moreover, even in their daily lives, people are continuously confronted with dynamical systems. Today, there is a growing need in techniques to closely monitor these systems and automate the accompanying control tasks. Systems generate, in one way or another, data, exhibiting temporal and/or spatial correlations. Well-suited models based on that data, make the analysis and manipulation of systems possible. The advent of some modern technologies, like the internet-of-things, industry 4.0, or the big data hype, only accelerates the need for techniques to translate the staggering amounts of data into useful models.
My research is about deriving globally optimal least-squares models from observed input-output data and large high-order models. Current state-of-the-art algorithms do not guarantee to solve these problems exactly. The cornerstone of this proposal is therefore to perform system identification and model reduction in a globally optimal least-squares way. To achieve this, I use numerical linear algebra techniques and multidimensional realization theory to solve the underlying optimization problems. 

Publications