Subspace based methods for continuous-time model identification of MIMO systems from filtered sampled data

This article introduces a new identification method for continuous-time MIMO state space models from sampled input output data. The proposed approach consists more precisely in combining filtering techniques with a specific subspace algorithm. Two filtering methods (the reinitialised partial moments and the Poisson moment functionals) are considered to circumvent the time derivative problem inherent in continuous-time modelling. The developed subspace algorithm belongs to the MOESP method family. A particular attention is payed to the construction of the instrumental variable used to supply consistent and accurate estimates in a noisy framework. The benefits of the proposed algorithms in comparison with existing methods are illustrated with a simulation study.