Identification of nonlinear dynamic objects by pattern recognition methods (Q1062026)

A new method of identifying nonlinear dynamical objects based on pattern recognition methods is discussed. Identified objects are described by a functional Volterra series. The relative semiinvariants are used as features. To choose the optimal model a nearest neighbor algorithm is used with Euclidean distance measure. The asymptotic properties of mutual relative semiinvariants are derived and a computational experiment in the identification of typical nonlinear systems is given.