Consistency analysis of subspace identification methods based on a linear regression approach (Q1592896)

The paper deals with the subspace estimation method of a linear state-space model using the linear least squares approach. The main contribution is the consistency analysis of the extended observability matrix estimate, computed directly from the input-output measurement data, and the observation that the consistency can be achieved by including a suitable number of past data in the regression model. Sufficient algebraic (matrix rank) type consistency conditions are given and the consistency proof is extended to the parameters in the deterministic part of the system description. The considerations are illustrated by means of simulation examples.