Nonlinear dynamical systems control using a new RNN temporal learning strategy

摘要

The ability of recurrent neural networks (RNN) to handle time-varying input/output through its own temporal operation is discussed. A new class of continuous-time (CT) RNN is proposed and it is proved that any finite time trajectory of a given n-dimensional dynamical CT system with input can be approximated by the internal state of the output units of an RNN. The proposed RNNs are extended for temporal processing.