A Neural Network Approach for Identification of Continuous-Time Nonlinear Dynamic Systems

Abstract

In this paper, a neural network approach for identifying continuous time nonlinear dynamic systems is presented. The nonlinear dynamic system may be described by a state space model or represented by an input-output relationship. The concept of state-variable filter is employed such that no derivatives of the output or input are required. The weight adjustments are based on a gradient algorithm and can be carried out by a bank of parallel analog filters.