Modeling of switching-mode nonlinear system by exponentially weighted aggregation

We consider a class of non-linear dynamic systems with finite memory and nonlinear characteristic that reveals time-varying behavior. It is assumed that the considered systems can change their mode in the a priori known range of modes but the switching moments are unknown and cannot be directly detected. In the considered approach, based on the noisy measurements of the system output, we apply exponentially weighted aggregation techniques to estimate noise-free counterparts of the possessed output observations. Theoretical properties of the method as well as exemplary numerical simulations are also described and discussed in the paper.