Neural network modeling of dynamical systems
Jan T. Białasiewicz, Donald I. Soloway · 2002
A recurrent backpropagation neural network suitable for modelling dynamical systems is analyzed. It is shown that the weight matrices of the neural network model determine, with reasonable accuracy, the impulse response of the modelled dynamical system. By analyzing this impulse response with the eigensystem realization algorithm (ERA), one can obtain state-space representation of the original system. Simulation results are presented.>