Optimal Hankel-norm identification of dynamical systems (Q1367174)

A linear time-invariant model is considered in case of misfit between it and an observed time series. The misfit resembles a data weighted gap between the model and the behavior of a specific for a given data linear exact model. The minimization of the misfit function is expressed as minimization of the Hankel norm of an operator that is associated with the data and with the co-inner kernel representation of the model. So the misfit function is computed by formalizing Pareto-optimal models. Characterization of the model complexity and misfit in terms of the kernel representation is provided. The key result of the paper is that all Pareto-optimal models are parametrized as exact models of data sets obtained from Hankel-norm approximations of the original data.