Optimal recursive algorithm for identification of time-varying systems

Abstract

07Oct1992|07Jan1993|08Jan1993 New developments for recursive parameter identification are presented for deterministic and stochastic non-linear systems with rapidly time-varying parameters. The new algorithm is based upon applying a linear combination of the equations determined by the ordinary least-squares method. The particular form of the resulting estimation error generic to recursive filters is highlighted and it retains provable convergence at low convergence rates and is well suited to real-time applications. The stability theory basis for the proof of this scheme allows its use in a noiseless feedback environment. The estimation of the output error of the identified system is shown to be consistent, despite convergent parameter estimate inconsistency. The class of systems encompassed by the linear in the parameters formulation is shown to allow consistent output estimation despite severe alteration of the system structure within a known class at a priori unspecified instants.