Exponentially weighted aggregation of models for Wiener-Hammerstein system modelling

In the paper the problem of Wiener-Hammerstein system (LNL) modelling by exponentially weighted aggregation of models is discussed. The class of systems under consideration admits almost any LNL objects even with infinite memory and virtually arbitrary nonlinear characteristic. Given the set of various possible models of the true system (being e.g. the result of prior identification attempts), the proposed approach yields mixed model that efficiently mimics input-output relationship of the genuine LNL system. Detailed analysis of the approach demonstrates good theoretical properties of the resulting (aggregated) model, whereas numerical simulations illustrate practical aspects of the considered method.