A realtime adaptive system for dynamics recognition

We propose and investigate an adaptive system for the recognition of the dynamics of an external time series. The system consists of a pool of internal dynamical elements, each of which represents a specific dynamical type. Each of the elements is forced by the external time series to the latter dynamics. We use the absolute value of the control signal, i.e., the strength of forcing, as a criterion for which of the elements out of the pool fits best to the dynamics of the time series. By means of an averaging process the system is able to create a new dynamical element as a “mirror system” of the external one. This adaptation can be performed continuously and impressively quickly even if the dynamical type of the external signal undergoes sudden qualitative changes.