Obtaining neural networks based state space models using time-lagged neurons

Obtaining discrete-time state space models directly from experimental input-output data is the focus of this paper. We use a neural network structure that allows direct mapping of the state space model since the hidden neurons have self-feedback plus feedback from the other neurons of the hidden layer. Simulation results from preliminary studies indicate that this tool could be developed into a very promising technique in the field of dynamic system modeling and identification. The method can be applied to both linear and nonlinear systems.