A neural network based identification-control paradigm via adaptive prediction

A neural-network-based control scheme with an identification-prediction-control paradigm for a linear (nonlinear) single-input single-output (SISO) system is proposed. A Hebbian or projection learning algorithm is used for neuronal identification of the system, an adaptive d-step ahead prediction method is derived for anticipating its dynamic behavior, and the control law is obtained on the basis of the prediction. Therefore, one has two control loops: the inner loop for identification and the outer loop for control. The proposed control architecture has better performance in the case of linear processes. It is emphasized that the learning mechanisms of artificial neural networks can relax some restrictions on the process