NEURAL NETWORKS FOR IDENTIFICATION OF DYNAMIC SYSTEMS

Abstract

Identification is the determination of a system within a specified class of system, on the basis of inputs and outputs. Such determination means that some variables, characterizing the given system are chosen and some relation are defined in the form of formula or graphs. In this dissertation, a single layer ANN has been developed to identify the parameters of the linear dynamic system whose states and derivatives of states are given. Gradient descent algorithm has been used to learn the network. This algorithm made the learning very fast and provides global results. By this method, a non-linear system has also been identified in the form of a linear system about its operating point. Further, the effect of change in learning rate has been studied. This method has been successfully implemented on three sample systems and the results of identification of system parameter are reported