New gradient based identification methods for multivariate pseudo-linear systems using the multi-innovation and the data filtering

This paper proposes parameter identification methods for multivariate pseudo-linear autoregressive systems. First, a multivariate generalized stochastic gradient (M-GSG) algorithm is presented as a comparison basis. In order to improve the parameter estimation accuracy, a multivariate multi-innovation generalized stochastic gradient (M-MI-GSG) algorithm and a filtering based multivariate generalized stochastic gradient (F-M-GSG) algorithm are presented by means of the multi-innovation identification theory and the data filtering technique. The simulation results confirm that the proposed algorithms are more effective than the M-GSG algorithm.