Dynamic system identification using neural networks

A practical neural network design method for the identification of both the direct transfer function and inverse transfer function of an object plant is proposed. As a practical application of the direct transfer function identifier, a nonlinear plant simulator is also proposed. Simulated and experimental results for a second-order plant show that identification can be satisfactorily achieved and that neural network identifiers can represent nonlinear plant characteristics very well. The characteristics of a neural network direct controller with a feedback control loop, which uses the learning results of the inverse transfer function identifier, is also proposed and confirmed