Estimation of dynamic system parameters by neural networks

Abstract

Identification of dynamic systems, operating under correlated noise, is conventionally performed by the generalized least squares algorithm. The Hopfield neural network has been used in connection with the generalized least squares technique to identify the system parameters. A theoretical comparison is made between the conventional generalized least squares and the neural-network-based generalized least squares techniques. This comparison is also supported by the simulated examples. It is shown that the Hopfield-based neural network can perform two fundamental steps of the generalized least squares algorithm in parallel fashion. These steps are the application of least squares routines.<>