Development of computationally efficient and numerically robust system identification software

Project members: V. Verdult, M. Verhaegen

Over the years several methods for system identification of multivariable linear state-space systems have been developed. To make these methods available to a large community of scientist and engineers in both industry and academics, a reliable and efficient implementation of these methods is needed. The goal of this work is to develop a software package that provides computationally efficient and numerically robust implementations of several system identification methods for multivariable linear state-space systems. The software is developed for use with Matlab and is partly written in the C programming language. It makes use of numerical linear algebra routines from BLAS, LAPACK, and SLICOT.

The software development focuses on the following identification methods: 1) time and frequency domain subspace methods, 2) state-space system identification based on minimizing the prediction error, and 3) state-space system identification by fitting frequency response functions.