Unknown parameter identification of parameterized system using multi-layered neural network

A system identification algorithm using multilayered neural networks is proposed. This neural network identification is not intended to match the input-output properties of a given system, but rather to estimate the unknown parameters of the system. Thus, the outputs of the neural network become the estimation of the parameters. After defining an identification structure and an identifier error, it is shown that the parameters can be updated by the error-back-propagation algorithm by means of the steepest descent method. The stability and convergence are proved by Lyapunov analysis so that both the identifier error and the rate of the parameter error are shown to converge to zero. Simulation studies verify the proposed identification algorithm