An efficient recursive identification algorithm for ARMAX model

In this paper the problem of ARMAX model identification is studied and an efficient recursive robust identification algorithm applicable to ARMAX model is proposed. When applying directly LS method to ARMAX model estimation, the asymptotical bias appears. In order to estimate the bias, inspiring from the bias compensation least squares (BCLS) algorithm that is proposed by authors, a set of auxiliary linear backward predictors are introduced. With the help of those found orthogonal properties, a solution to the estimate of the bias is built. Consequently the consistent estimate for ARMAX model can be obtained via compensating the estimated bias. Moreover in order to satisfy the need of on-line identification of ARMAX model, recursive processing of the proposed algorithm is given. Simulation results are presented to illustrate the effectiveness of the proposed algorithm.