Abstract
A new optimization-based approach to experiment design for dynamic MIMO identification is presented. Unlike previously proposed methods that consider dynamics and constraints, the proposed method directly addresses the distribution of output data to be generated in the experiment. The problem is formulated as a convex optimization problem with constraints expressed as linear matrix inequalities (LMIs). Three optimization objectives considered are to maximize the smallest singular value and the determinant of the output covariance matrix as well as the determinant of the input covariance matrix. The solution can be implemented in a similar way as previously proposed gain-directional designs based on an estimate of the static gain matrix. The type of perturbation (e.g., RBS, PRBS, multisine signal) can be selected by the user. Various aspects of the method are illustrated by the Wood-Berry distillation column.
Original language | Undefined/Unknown |
---|---|
Pages (from-to) | 7321–7326 |
Journal | IFAC papers online |
Volume | 50 |
Issue number | 1 |
DOIs | |
Publication status | Published - 2017 |
MoE publication type | A1 Journal article-refereed |
Keywords
- system identification
- Multivariable systems
- experiment design
- Input signals
- Linear matrix inequalities
- E-optimality
- Convex optimization
- Control-oriented models
- ill-conditioned systems