A tutorial overview on multivariable identification algorithms for process systems

The widespread adoption of multivariable optimization based control systems, usually known as Model Predictuve Control (MPC) technologies, have boosted the research and development activities in numerous related fields. In particular, systems identification methods gained tremendous importance especially for large multivariable processes, which pose significant challenges to control engineers due to large selling times, process noise, correlations among (output and/or input) variables, operational constraints. In the first part of this talk, an overview of the basic steps of a multivariable system identification application project will be presented, covering the main theoretical foundations of (open-loop and closed-loop) data collection, data treatment, identification algorithms, and model validation. In the second part of this talk, a tutorial review of subspace identification algorithms will be given, covering both standard algorithms (e.g. N4SID, MOESP, etc.) and advanced parsimonious methods which can be consistently applied to closed-loop data. Several examples of process systems will serve to illustrate and support the exposition of theory and design concepts.