A non-parametric normalized stochastic approximation algorithm for on-line identification of multivariable systems

Abstract

A non-parametric algorithm is proposed for on-line identification of linear multivariable discrete-time systems from the measurements of input-output data contaminated with noise. It is based on the estimation of the Markov parameters of the system from the measured data using a non-parametric normalized stochastic approximation algorithm. The state-space representation for the system is then obtained using an efficient algorithm for minimal realization. Results of simulation are included indicating that the scheme works successfully even when the output measurements are contaminated with considerable amounts of noise.