New Approach to Recursive Identification for ARMAX Systems

For the multivariate ARMAX system A(z)yk=B(z)uk-1+C(z)wk recursive algorithms are proposed for estimating coefficients of A(z), B(z), and C(z) and the covariance matrix Rw of wk, assuming that the orders of A(z) , B(z) , and C(z) are known and the control uk can be arbitrarily chosen. The new method consists in on-line solving the algebraic equations associated with ARMAX on the basis of observed data. The algorithms are easily computable, and the almost sure convergence of the algorithms is proved under reasonable conditions.