Improved subspace identification with prior information using constrained least squares

Abstract

Subspace identification incorporating prior information has been proven to be effective in obtaining state-space models with improved accuracy. Available algorithms, however, can require prohibitively demanding computations. The incorporation of prior information, for example, steady-state gain, time constant and zero transfer functions, in subspace identification is investigated using constrained least squares (CLS). The method exploits the interpretation of subspace identification as an optimal multi-step ahead predictor and reduces the identification to solve an optimisation problem with equality constraints describing the prior information. The standard multivariable output-error state-space subspace algorithm is further examined using the same CLS approach to incorporate dc gain information. Simulation results show that the proposed algorithm provides computationally efficient approach for subspace identification with satisfactory parameter variances, and the method is equally applicable to both single-input single-output and multiple-input multiple-output systems.